Upload folder using huggingface_hub
Browse files- README.md +27 -15
- adapter_config.json +9 -9
- adapter_model.bin +2 -2
- checkpoint-126/README.md +202 -0
- checkpoint-126/adapter_config.json +39 -0
- checkpoint-126/adapter_model.safetensors +3 -0
- checkpoint-126/global_step126/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt +3 -0
- checkpoint-126/global_step126/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt +3 -0
- checkpoint-126/global_step126/mp_rank_00_model_states.pt +3 -0
- checkpoint-126/latest +1 -0
- checkpoint-126/rng_state_0.pth +3 -0
- checkpoint-126/rng_state_1.pth +3 -0
- checkpoint-126/scheduler.pt +3 -0
- checkpoint-126/special_tokens_map.json +30 -0
- checkpoint-126/tokenizer.json +0 -0
- checkpoint-126/tokenizer.model +3 -0
- checkpoint-126/tokenizer_config.json +48 -0
- checkpoint-126/trainer_state.json +935 -0
- checkpoint-126/training_args.bin +3 -0
- checkpoint-126/zero_to_fp32.py +592 -0
- checkpoint-168/README.md +202 -0
- checkpoint-168/adapter_config.json +39 -0
- checkpoint-168/adapter_model.safetensors +3 -0
- checkpoint-168/global_step168/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt +3 -0
- checkpoint-168/global_step168/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt +3 -0
- checkpoint-168/global_step168/mp_rank_00_model_states.pt +3 -0
- checkpoint-168/latest +1 -0
- checkpoint-168/rng_state_0.pth +3 -0
- checkpoint-168/rng_state_1.pth +3 -0
- checkpoint-168/scheduler.pt +3 -0
- checkpoint-168/special_tokens_map.json +30 -0
- checkpoint-168/tokenizer.json +0 -0
- checkpoint-168/tokenizer.model +3 -0
- checkpoint-168/tokenizer_config.json +48 -0
- checkpoint-168/trainer_state.json +1237 -0
- checkpoint-168/training_args.bin +3 -0
- checkpoint-168/zero_to_fp32.py +592 -0
- config.json +7 -9
- tokenizer_config.json +1 -0
README.md
CHANGED
@@ -7,11 +7,10 @@ base_model: ai21labs/Jamba-v0.1
|
|
7 |
model-index:
|
8 |
- name: out
|
9 |
results: []
|
10 |
-
datasets:
|
11 |
-
- chargoddard/Open-Platypus-Chat
|
12 |
---
|
13 |
|
14 |
-
|
|
|
15 |
|
16 |
[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
|
17 |
<details><summary>See axolotl config</summary>
|
@@ -33,7 +32,6 @@ chat_template: chatml
|
|
33 |
dataset_prepared_path:
|
34 |
val_set_size: 0.01
|
35 |
output_dir: ./out
|
36 |
-
hub_strategy: every_save
|
37 |
|
38 |
sequence_len: 4096
|
39 |
sample_packing: true
|
@@ -80,9 +78,6 @@ logging_steps: 1
|
|
80 |
xformers_attention:
|
81 |
flash_attention: true
|
82 |
|
83 |
-
loss_watchdog_threshold: 5.0
|
84 |
-
loss_watchdog_patience: 3
|
85 |
-
|
86 |
warmup_steps: 10
|
87 |
evals_per_epoch: 4
|
88 |
saves_per_epoch: 4
|
@@ -96,9 +91,23 @@ special_tokens:
|
|
96 |
|
97 |
</details><br>
|
98 |
|
99 |
-
|
|
|
|
|
100 |
It achieves the following results on the evaluation set:
|
101 |
-
- Loss: 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
102 |
|
103 |
## Training procedure
|
104 |
|
@@ -109,8 +118,11 @@ The following hyperparameters were used during training:
|
|
109 |
- train_batch_size: 1
|
110 |
- eval_batch_size: 1
|
111 |
- seed: 42
|
|
|
|
|
112 |
- gradient_accumulation_steps: 8
|
113 |
-
- total_train_batch_size:
|
|
|
114 |
- optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
|
115 |
- lr_scheduler_type: cosine
|
116 |
- lr_scheduler_warmup_steps: 10
|
@@ -120,11 +132,11 @@ The following hyperparameters were used during training:
|
|
120 |
|
121 |
| Training Loss | Epoch | Step | Validation Loss |
|
122 |
|:-------------:|:-----:|:----:|:---------------:|
|
123 |
-
| 0.
|
124 |
-
| 0.
|
125 |
-
| 0.
|
126 |
-
| 0.
|
127 |
-
| 0.
|
128 |
|
129 |
|
130 |
### Framework versions
|
|
|
7 |
model-index:
|
8 |
- name: out
|
9 |
results: []
|
|
|
|
|
10 |
---
|
11 |
|
12 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
13 |
+
should probably proofread and complete it, then remove this comment. -->
|
14 |
|
15 |
[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
|
16 |
<details><summary>See axolotl config</summary>
|
|
|
32 |
dataset_prepared_path:
|
33 |
val_set_size: 0.01
|
34 |
output_dir: ./out
|
|
|
35 |
|
36 |
sequence_len: 4096
|
37 |
sample_packing: true
|
|
|
78 |
xformers_attention:
|
79 |
flash_attention: true
|
80 |
|
|
|
|
|
|
|
81 |
warmup_steps: 10
|
82 |
evals_per_epoch: 4
|
83 |
saves_per_epoch: 4
|
|
|
91 |
|
92 |
</details><br>
|
93 |
|
94 |
+
# out
|
95 |
+
|
96 |
+
This model is a fine-tuned version of [ai21labs/Jamba-v0.1](https://huggingface.co/ai21labs/Jamba-v0.1) on the None dataset.
|
97 |
It achieves the following results on the evaluation set:
|
98 |
+
- Loss: 0.9651
|
99 |
+
|
100 |
+
## Model description
|
101 |
+
|
102 |
+
More information needed
|
103 |
+
|
104 |
+
## Intended uses & limitations
|
105 |
+
|
106 |
+
More information needed
|
107 |
+
|
108 |
+
## Training and evaluation data
|
109 |
+
|
110 |
+
More information needed
|
111 |
|
112 |
## Training procedure
|
113 |
|
|
|
118 |
- train_batch_size: 1
|
119 |
- eval_batch_size: 1
|
120 |
- seed: 42
|
121 |
+
- distributed_type: multi-GPU
|
122 |
+
- num_devices: 2
|
123 |
- gradient_accumulation_steps: 8
|
124 |
+
- total_train_batch_size: 16
|
125 |
+
- total_eval_batch_size: 2
|
126 |
- optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
|
127 |
- lr_scheduler_type: cosine
|
128 |
- lr_scheduler_warmup_steps: 10
|
|
|
132 |
|
133 |
| Training Loss | Epoch | Step | Validation Loss |
|
134 |
|:-------------:|:-----:|:----:|:---------------:|
|
135 |
+
| 0.6274 | 0.01 | 1 | 1.0298 |
|
136 |
+
| 0.4403 | 0.25 | 42 | 0.9768 |
|
137 |
+
| 0.4417 | 0.5 | 84 | 0.9675 |
|
138 |
+
| 0.4451 | 0.75 | 126 | 0.9652 |
|
139 |
+
| 0.4616 | 1.0 | 168 | 0.9651 |
|
140 |
|
141 |
|
142 |
### Framework versions
|
adapter_config.json
CHANGED
@@ -20,18 +20,18 @@
|
|
20 |
"rank_pattern": {},
|
21 |
"revision": null,
|
22 |
"target_modules": [
|
23 |
-
"
|
24 |
-
"
|
|
|
|
|
|
|
|
|
25 |
"k_proj",
|
|
|
26 |
"in_proj",
|
|
|
27 |
"out_proj",
|
28 |
-
"
|
29 |
-
"down_proj",
|
30 |
-
"router",
|
31 |
-
"up_proj",
|
32 |
-
"o_proj",
|
33 |
-
"v_proj",
|
34 |
-
"x_proj"
|
35 |
],
|
36 |
"task_type": "CAUSAL_LM",
|
37 |
"use_dora": false,
|
|
|
20 |
"rank_pattern": {},
|
21 |
"revision": null,
|
22 |
"target_modules": [
|
23 |
+
"x_proj",
|
24 |
+
"o_proj",
|
25 |
+
"up_proj",
|
26 |
+
"v_proj",
|
27 |
+
"down_proj",
|
28 |
+
"dt_proj",
|
29 |
"k_proj",
|
30 |
+
"q_proj",
|
31 |
"in_proj",
|
32 |
+
"gate_proj",
|
33 |
"out_proj",
|
34 |
+
"router"
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
],
|
36 |
"task_type": "CAUSAL_LM",
|
37 |
"use_dora": false,
|
adapter_model.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:485eddbdd38763c9850251f824a3877a9ddc0db9c893fae77c6ad1f9eb0d379c
|
3 |
+
size 531653306
|
checkpoint-126/README.md
ADDED
@@ -0,0 +1,202 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
library_name: peft
|
3 |
+
base_model: ai21labs/Jamba-v0.1
|
4 |
+
---
|
5 |
+
|
6 |
+
# Model Card for Model ID
|
7 |
+
|
8 |
+
<!-- Provide a quick summary of what the model is/does. -->
|
9 |
+
|
10 |
+
|
11 |
+
|
12 |
+
## Model Details
|
13 |
+
|
14 |
+
### Model Description
|
15 |
+
|
16 |
+
<!-- Provide a longer summary of what this model is. -->
|
17 |
+
|
18 |
+
|
19 |
+
|
20 |
+
- **Developed by:** [More Information Needed]
|
21 |
+
- **Funded by [optional]:** [More Information Needed]
|
22 |
+
- **Shared by [optional]:** [More Information Needed]
|
23 |
+
- **Model type:** [More Information Needed]
|
24 |
+
- **Language(s) (NLP):** [More Information Needed]
|
25 |
+
- **License:** [More Information Needed]
|
26 |
+
- **Finetuned from model [optional]:** [More Information Needed]
|
27 |
+
|
28 |
+
### Model Sources [optional]
|
29 |
+
|
30 |
+
<!-- Provide the basic links for the model. -->
|
31 |
+
|
32 |
+
- **Repository:** [More Information Needed]
|
33 |
+
- **Paper [optional]:** [More Information Needed]
|
34 |
+
- **Demo [optional]:** [More Information Needed]
|
35 |
+
|
36 |
+
## Uses
|
37 |
+
|
38 |
+
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
|
39 |
+
|
40 |
+
### Direct Use
|
41 |
+
|
42 |
+
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
|
43 |
+
|
44 |
+
[More Information Needed]
|
45 |
+
|
46 |
+
### Downstream Use [optional]
|
47 |
+
|
48 |
+
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
|
49 |
+
|
50 |
+
[More Information Needed]
|
51 |
+
|
52 |
+
### Out-of-Scope Use
|
53 |
+
|
54 |
+
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
|
55 |
+
|
56 |
+
[More Information Needed]
|
57 |
+
|
58 |
+
## Bias, Risks, and Limitations
|
59 |
+
|
60 |
+
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
|
61 |
+
|
62 |
+
[More Information Needed]
|
63 |
+
|
64 |
+
### Recommendations
|
65 |
+
|
66 |
+
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
|
67 |
+
|
68 |
+
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
|
69 |
+
|
70 |
+
## How to Get Started with the Model
|
71 |
+
|
72 |
+
Use the code below to get started with the model.
|
73 |
+
|
74 |
+
[More Information Needed]
|
75 |
+
|
76 |
+
## Training Details
|
77 |
+
|
78 |
+
### Training Data
|
79 |
+
|
80 |
+
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
|
81 |
+
|
82 |
+
[More Information Needed]
|
83 |
+
|
84 |
+
### Training Procedure
|
85 |
+
|
86 |
+
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
|
87 |
+
|
88 |
+
#### Preprocessing [optional]
|
89 |
+
|
90 |
+
[More Information Needed]
|
91 |
+
|
92 |
+
|
93 |
+
#### Training Hyperparameters
|
94 |
+
|
95 |
+
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
|
96 |
+
|
97 |
+
#### Speeds, Sizes, Times [optional]
|
98 |
+
|
99 |
+
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
|
100 |
+
|
101 |
+
[More Information Needed]
|
102 |
+
|
103 |
+
## Evaluation
|
104 |
+
|
105 |
+
<!-- This section describes the evaluation protocols and provides the results. -->
|
106 |
+
|
107 |
+
### Testing Data, Factors & Metrics
|
108 |
+
|
109 |
+
#### Testing Data
|
110 |
+
|
111 |
+
<!-- This should link to a Dataset Card if possible. -->
|
112 |
+
|
113 |
+
[More Information Needed]
|
114 |
+
|
115 |
+
#### Factors
|
116 |
+
|
117 |
+
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
|
118 |
+
|
119 |
+
[More Information Needed]
|
120 |
+
|
121 |
+
#### Metrics
|
122 |
+
|
123 |
+
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
|
124 |
+
|
125 |
+
[More Information Needed]
|
126 |
+
|
127 |
+
### Results
|
128 |
+
|
129 |
+
[More Information Needed]
|
130 |
+
|
131 |
+
#### Summary
|
132 |
+
|
133 |
+
|
134 |
+
|
135 |
+
## Model Examination [optional]
|
136 |
+
|
137 |
+
<!-- Relevant interpretability work for the model goes here -->
|
138 |
+
|
139 |
+
[More Information Needed]
|
140 |
+
|
141 |
+
## Environmental Impact
|
142 |
+
|
143 |
+
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
|
144 |
+
|
145 |
+
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
|
146 |
+
|
147 |
+
- **Hardware Type:** [More Information Needed]
|
148 |
+
- **Hours used:** [More Information Needed]
|
149 |
+
- **Cloud Provider:** [More Information Needed]
|
150 |
+
- **Compute Region:** [More Information Needed]
|
151 |
+
- **Carbon Emitted:** [More Information Needed]
|
152 |
+
|
153 |
+
## Technical Specifications [optional]
|
154 |
+
|
155 |
+
### Model Architecture and Objective
|
156 |
+
|
157 |
+
[More Information Needed]
|
158 |
+
|
159 |
+
### Compute Infrastructure
|
160 |
+
|
161 |
+
[More Information Needed]
|
162 |
+
|
163 |
+
#### Hardware
|
164 |
+
|
165 |
+
[More Information Needed]
|
166 |
+
|
167 |
+
#### Software
|
168 |
+
|
169 |
+
[More Information Needed]
|
170 |
+
|
171 |
+
## Citation [optional]
|
172 |
+
|
173 |
+
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
174 |
+
|
175 |
+
**BibTeX:**
|
176 |
+
|
177 |
+
[More Information Needed]
|
178 |
+
|
179 |
+
**APA:**
|
180 |
+
|
181 |
+
[More Information Needed]
|
182 |
+
|
183 |
+
## Glossary [optional]
|
184 |
+
|
185 |
+
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
|
186 |
+
|
187 |
+
[More Information Needed]
|
188 |
+
|
189 |
+
## More Information [optional]
|
190 |
+
|
191 |
+
[More Information Needed]
|
192 |
+
|
193 |
+
## Model Card Authors [optional]
|
194 |
+
|
195 |
+
[More Information Needed]
|
196 |
+
|
197 |
+
## Model Card Contact
|
198 |
+
|
199 |
+
[More Information Needed]
|
200 |
+
### Framework versions
|
201 |
+
|
202 |
+
- PEFT 0.10.0
|
checkpoint-126/adapter_config.json
ADDED
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"alpha_pattern": {},
|
3 |
+
"auto_mapping": null,
|
4 |
+
"base_model_name_or_path": "ai21labs/Jamba-v0.1",
|
5 |
+
"bias": "none",
|
6 |
+
"fan_in_fan_out": null,
|
7 |
+
"inference_mode": true,
|
8 |
+
"init_lora_weights": true,
|
9 |
+
"layer_replication": null,
|
10 |
+
"layers_pattern": null,
|
11 |
+
"layers_to_transform": null,
|
12 |
+
"loftq_config": {},
|
13 |
+
"lora_alpha": 32,
|
14 |
+
"lora_dropout": 0.05,
|
15 |
+
"megatron_config": null,
|
16 |
+
"megatron_core": "megatron.core",
|
17 |
+
"modules_to_save": null,
|
18 |
+
"peft_type": "LORA",
|
19 |
+
"r": 16,
|
20 |
+
"rank_pattern": {},
|
21 |
+
"revision": null,
|
22 |
+
"target_modules": [
|
23 |
+
"x_proj",
|
24 |
+
"o_proj",
|
25 |
+
"up_proj",
|
26 |
+
"v_proj",
|
27 |
+
"down_proj",
|
28 |
+
"dt_proj",
|
29 |
+
"k_proj",
|
30 |
+
"q_proj",
|
31 |
+
"in_proj",
|
32 |
+
"gate_proj",
|
33 |
+
"out_proj",
|
34 |
+
"router"
|
35 |
+
],
|
36 |
+
"task_type": "CAUSAL_LM",
|
37 |
+
"use_dora": false,
|
38 |
+
"use_rslora": false
|
39 |
+
}
|
checkpoint-126/adapter_model.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:be949fb50dbe53d64e9f4375ed327e36ae2368bf683cfdaf813667de418fcbb8
|
3 |
+
size 531611600
|
checkpoint-126/global_step126/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f6998ecb2f6d60f27d30371a8ce207805cf8a48dd8f296036fff813d22b65593
|
3 |
+
size 797643792
|
checkpoint-126/global_step126/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1f504e0638f11f71fa63e7587596b481c876e9ec17a2e4703833ab1713a855b1
|
3 |
+
size 797644432
|
checkpoint-126/global_step126/mp_rank_00_model_states.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:271bfc43b751ef756538a0f6b19ee11530e5d5c9f68f5e3c3b086877dc1d3a6a
|
3 |
+
size 1345660121
|
checkpoint-126/latest
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
global_step126
|
checkpoint-126/rng_state_0.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:601237d8f7ccca9157ec802c2d8dbf105a33ca51949abaf17e8caac019fa582f
|
3 |
+
size 14512
|
checkpoint-126/rng_state_1.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:0b453660e9304c04b047fceaa4cad80dfb1b50b01880351c794e0268dc4d5d75
|
3 |
+
size 14512
|
checkpoint-126/scheduler.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:fb0b78cebb6cf02731e93f8b5296f19d99fc5821bb80f3f750f5f7f3c9de25e2
|
3 |
+
size 1064
|
checkpoint-126/special_tokens_map.json
ADDED
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"bos_token": {
|
3 |
+
"content": "<|startoftext|>",
|
4 |
+
"lstrip": false,
|
5 |
+
"normalized": false,
|
6 |
+
"rstrip": false,
|
7 |
+
"single_word": false
|
8 |
+
},
|
9 |
+
"eos_token": {
|
10 |
+
"content": "<|endoftext|>",
|
11 |
+
"lstrip": false,
|
12 |
+
"normalized": false,
|
13 |
+
"rstrip": false,
|
14 |
+
"single_word": false
|
15 |
+
},
|
16 |
+
"pad_token": {
|
17 |
+
"content": "<|pad|>",
|
18 |
+
"lstrip": false,
|
19 |
+
"normalized": false,
|
20 |
+
"rstrip": false,
|
21 |
+
"single_word": false
|
22 |
+
},
|
23 |
+
"unk_token": {
|
24 |
+
"content": "<|unk|>",
|
25 |
+
"lstrip": false,
|
26 |
+
"normalized": false,
|
27 |
+
"rstrip": false,
|
28 |
+
"single_word": false
|
29 |
+
}
|
30 |
+
}
|
checkpoint-126/tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
checkpoint-126/tokenizer.model
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:02fd6530b8ede0eedd8e509fcab32da7b1dd04c8119f8498c787100f13112713
|
3 |
+
size 1124742
|
checkpoint-126/tokenizer_config.json
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"add_bos_token": true,
|
3 |
+
"add_eos_token": false,
|
4 |
+
"added_tokens_decoder": {
|
5 |
+
"0": {
|
6 |
+
"content": "<|pad|>",
|
7 |
+
"lstrip": false,
|
8 |
+
"normalized": false,
|
9 |
+
"rstrip": false,
|
10 |
+
"single_word": false,
|
11 |
+
"special": true
|
12 |
+
},
|
13 |
+
"1": {
|
14 |
+
"content": "<|startoftext|>",
|
15 |
+
"lstrip": false,
|
16 |
+
"normalized": false,
|
17 |
+
"rstrip": false,
|
18 |
+
"single_word": false,
|
19 |
+
"special": true
|
20 |
+
},
|
21 |
+
"2": {
|
22 |
+
"content": "<|endoftext|>",
|
23 |
+
"lstrip": false,
|
24 |
+
"normalized": false,
|
25 |
+
"rstrip": false,
|
26 |
+
"single_word": false,
|
27 |
+
"special": true
|
28 |
+
},
|
29 |
+
"3": {
|
30 |
+
"content": "<|unk|>",
|
31 |
+
"lstrip": false,
|
32 |
+
"normalized": false,
|
33 |
+
"rstrip": false,
|
34 |
+
"single_word": false,
|
35 |
+
"special": true
|
36 |
+
}
|
37 |
+
},
|
38 |
+
"bos_token": "<|startoftext|>",
|
39 |
+
"chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
|
40 |
+
"clean_up_tokenization_spaces": false,
|
41 |
+
"eos_token": "<|endoftext|>",
|
42 |
+
"model_max_length": 1000000000000000019884624838656,
|
43 |
+
"pad_token": "<|pad|>",
|
44 |
+
"spaces_between_special_tokens": false,
|
45 |
+
"tokenizer_class": "LlamaTokenizer",
|
46 |
+
"unk_token": "<|unk|>",
|
47 |
+
"use_default_system_prompt": false
|
48 |
+
}
|
checkpoint-126/trainer_state.json
ADDED
@@ -0,0 +1,935 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"best_metric": null,
|
3 |
+
"best_model_checkpoint": null,
|
4 |
+
"epoch": 0.7488855869242199,
|
5 |
+
"eval_steps": 42,
|
6 |
+
"global_step": 126,
|
7 |
+
"is_hyper_param_search": false,
|
8 |
+
"is_local_process_zero": true,
|
9 |
+
"is_world_process_zero": true,
|
10 |
+
"log_history": [
|
11 |
+
{
|
12 |
+
"epoch": 0.01,
|
13 |
+
"grad_norm": 0.30475762367257353,
|
14 |
+
"learning_rate": 2e-05,
|
15 |
+
"loss": 0.6274,
|
16 |
+
"step": 1
|
17 |
+
},
|
18 |
+
{
|
19 |
+
"epoch": 0.01,
|
20 |
+
"eval_loss": 1.0297596454620361,
|
21 |
+
"eval_runtime": 150.6014,
|
22 |
+
"eval_samples_per_second": 1.653,
|
23 |
+
"eval_steps_per_second": 0.83,
|
24 |
+
"step": 1
|
25 |
+
},
|
26 |
+
{
|
27 |
+
"epoch": 0.01,
|
28 |
+
"grad_norm": 0.2798137150913395,
|
29 |
+
"learning_rate": 4e-05,
|
30 |
+
"loss": 0.6362,
|
31 |
+
"step": 2
|
32 |
+
},
|
33 |
+
{
|
34 |
+
"epoch": 0.02,
|
35 |
+
"grad_norm": 0.3178684451319545,
|
36 |
+
"learning_rate": 6e-05,
|
37 |
+
"loss": 0.6299,
|
38 |
+
"step": 3
|
39 |
+
},
|
40 |
+
{
|
41 |
+
"epoch": 0.02,
|
42 |
+
"grad_norm": 0.33651284916847835,
|
43 |
+
"learning_rate": 8e-05,
|
44 |
+
"loss": 0.6391,
|
45 |
+
"step": 4
|
46 |
+
},
|
47 |
+
{
|
48 |
+
"epoch": 0.03,
|
49 |
+
"grad_norm": 0.23984915388648712,
|
50 |
+
"learning_rate": 0.0001,
|
51 |
+
"loss": 0.6071,
|
52 |
+
"step": 5
|
53 |
+
},
|
54 |
+
{
|
55 |
+
"epoch": 0.04,
|
56 |
+
"grad_norm": 0.20514040410017348,
|
57 |
+
"learning_rate": 0.00012,
|
58 |
+
"loss": 0.5996,
|
59 |
+
"step": 6
|
60 |
+
},
|
61 |
+
{
|
62 |
+
"epoch": 0.04,
|
63 |
+
"grad_norm": 0.1950018128286362,
|
64 |
+
"learning_rate": 0.00014,
|
65 |
+
"loss": 0.6298,
|
66 |
+
"step": 7
|
67 |
+
},
|
68 |
+
{
|
69 |
+
"epoch": 0.05,
|
70 |
+
"grad_norm": 0.14246019947393238,
|
71 |
+
"learning_rate": 0.00016,
|
72 |
+
"loss": 0.5108,
|
73 |
+
"step": 8
|
74 |
+
},
|
75 |
+
{
|
76 |
+
"epoch": 0.05,
|
77 |
+
"grad_norm": 0.15792014279750227,
|
78 |
+
"learning_rate": 0.00018,
|
79 |
+
"loss": 0.5529,
|
80 |
+
"step": 9
|
81 |
+
},
|
82 |
+
{
|
83 |
+
"epoch": 0.06,
|
84 |
+
"grad_norm": 0.1517889177511264,
|
85 |
+
"learning_rate": 0.0002,
|
86 |
+
"loss": 0.5433,
|
87 |
+
"step": 10
|
88 |
+
},
|
89 |
+
{
|
90 |
+
"epoch": 0.07,
|
91 |
+
"grad_norm": 0.1372121219437277,
|
92 |
+
"learning_rate": 0.00019998023297700658,
|
93 |
+
"loss": 0.5856,
|
94 |
+
"step": 11
|
95 |
+
},
|
96 |
+
{
|
97 |
+
"epoch": 0.07,
|
98 |
+
"grad_norm": 0.16740807394942855,
|
99 |
+
"learning_rate": 0.00019992093972273018,
|
100 |
+
"loss": 0.5546,
|
101 |
+
"step": 12
|
102 |
+
},
|
103 |
+
{
|
104 |
+
"epoch": 0.08,
|
105 |
+
"grad_norm": 0.13512320693394078,
|
106 |
+
"learning_rate": 0.00019982214367819328,
|
107 |
+
"loss": 0.6193,
|
108 |
+
"step": 13
|
109 |
+
},
|
110 |
+
{
|
111 |
+
"epoch": 0.08,
|
112 |
+
"grad_norm": 0.16169796294070152,
|
113 |
+
"learning_rate": 0.0001996838839014696,
|
114 |
+
"loss": 0.5495,
|
115 |
+
"step": 14
|
116 |
+
},
|
117 |
+
{
|
118 |
+
"epoch": 0.09,
|
119 |
+
"grad_norm": 0.16796913812281988,
|
120 |
+
"learning_rate": 0.00019950621505224273,
|
121 |
+
"loss": 0.5035,
|
122 |
+
"step": 15
|
123 |
+
},
|
124 |
+
{
|
125 |
+
"epoch": 0.1,
|
126 |
+
"grad_norm": 0.1800514764162192,
|
127 |
+
"learning_rate": 0.00019928920737019733,
|
128 |
+
"loss": 0.5083,
|
129 |
+
"step": 16
|
130 |
+
},
|
131 |
+
{
|
132 |
+
"epoch": 0.1,
|
133 |
+
"grad_norm": 0.170432124866908,
|
134 |
+
"learning_rate": 0.0001990329466472502,
|
135 |
+
"loss": 0.632,
|
136 |
+
"step": 17
|
137 |
+
},
|
138 |
+
{
|
139 |
+
"epoch": 0.11,
|
140 |
+
"grad_norm": 0.19129325489749488,
|
141 |
+
"learning_rate": 0.00019873753419363336,
|
142 |
+
"loss": 0.4813,
|
143 |
+
"step": 18
|
144 |
+
},
|
145 |
+
{
|
146 |
+
"epoch": 0.11,
|
147 |
+
"grad_norm": 0.1459357988760762,
|
148 |
+
"learning_rate": 0.00019840308679784207,
|
149 |
+
"loss": 0.4973,
|
150 |
+
"step": 19
|
151 |
+
},
|
152 |
+
{
|
153 |
+
"epoch": 0.12,
|
154 |
+
"grad_norm": 0.192594730984382,
|
155 |
+
"learning_rate": 0.00019802973668046363,
|
156 |
+
"loss": 0.5291,
|
157 |
+
"step": 20
|
158 |
+
},
|
159 |
+
{
|
160 |
+
"epoch": 0.12,
|
161 |
+
"grad_norm": 0.859025467969139,
|
162 |
+
"learning_rate": 0.0001976176314419051,
|
163 |
+
"loss": 0.5296,
|
164 |
+
"step": 21
|
165 |
+
},
|
166 |
+
{
|
167 |
+
"epoch": 0.13,
|
168 |
+
"grad_norm": 0.13366297885670222,
|
169 |
+
"learning_rate": 0.000197166934004041,
|
170 |
+
"loss": 0.4819,
|
171 |
+
"step": 22
|
172 |
+
},
|
173 |
+
{
|
174 |
+
"epoch": 0.14,
|
175 |
+
"grad_norm": 0.15698714419747645,
|
176 |
+
"learning_rate": 0.00019667782254580374,
|
177 |
+
"loss": 0.5409,
|
178 |
+
"step": 23
|
179 |
+
},
|
180 |
+
{
|
181 |
+
"epoch": 0.14,
|
182 |
+
"grad_norm": 0.10995943735837355,
|
183 |
+
"learning_rate": 0.00019615049043274205,
|
184 |
+
"loss": 0.5108,
|
185 |
+
"step": 24
|
186 |
+
},
|
187 |
+
{
|
188 |
+
"epoch": 0.15,
|
189 |
+
"grad_norm": 0.10796742192788925,
|
190 |
+
"learning_rate": 0.00019558514614057609,
|
191 |
+
"loss": 0.5215,
|
192 |
+
"step": 25
|
193 |
+
},
|
194 |
+
{
|
195 |
+
"epoch": 0.15,
|
196 |
+
"grad_norm": 0.11641740089490231,
|
197 |
+
"learning_rate": 0.00019498201317277828,
|
198 |
+
"loss": 0.5012,
|
199 |
+
"step": 26
|
200 |
+
},
|
201 |
+
{
|
202 |
+
"epoch": 0.16,
|
203 |
+
"grad_norm": 0.1120175962893241,
|
204 |
+
"learning_rate": 0.00019434132997221345,
|
205 |
+
"loss": 0.474,
|
206 |
+
"step": 27
|
207 |
+
},
|
208 |
+
{
|
209 |
+
"epoch": 0.17,
|
210 |
+
"grad_norm": 0.1218171278782483,
|
211 |
+
"learning_rate": 0.0001936633498268728,
|
212 |
+
"loss": 0.5216,
|
213 |
+
"step": 28
|
214 |
+
},
|
215 |
+
{
|
216 |
+
"epoch": 0.17,
|
217 |
+
"grad_norm": 0.11718521115928844,
|
218 |
+
"learning_rate": 0.0001929483407697387,
|
219 |
+
"loss": 0.4856,
|
220 |
+
"step": 29
|
221 |
+
},
|
222 |
+
{
|
223 |
+
"epoch": 0.18,
|
224 |
+
"grad_norm": 0.12611471038571026,
|
225 |
+
"learning_rate": 0.00019219658547282067,
|
226 |
+
"loss": 0.4823,
|
227 |
+
"step": 30
|
228 |
+
},
|
229 |
+
{
|
230 |
+
"epoch": 0.18,
|
231 |
+
"grad_norm": 0.11106871615269753,
|
232 |
+
"learning_rate": 0.00019140838113540346,
|
233 |
+
"loss": 0.4869,
|
234 |
+
"step": 31
|
235 |
+
},
|
236 |
+
{
|
237 |
+
"epoch": 0.19,
|
238 |
+
"grad_norm": 0.1416503230360699,
|
239 |
+
"learning_rate": 0.00019058403936655233,
|
240 |
+
"loss": 0.5341,
|
241 |
+
"step": 32
|
242 |
+
},
|
243 |
+
{
|
244 |
+
"epoch": 0.2,
|
245 |
+
"grad_norm": 0.10761396399791698,
|
246 |
+
"learning_rate": 0.00018972388606192125,
|
247 |
+
"loss": 0.4304,
|
248 |
+
"step": 33
|
249 |
+
},
|
250 |
+
{
|
251 |
+
"epoch": 0.2,
|
252 |
+
"grad_norm": 0.10975376180434356,
|
253 |
+
"learning_rate": 0.0001888282612749132,
|
254 |
+
"loss": 0.4646,
|
255 |
+
"step": 34
|
256 |
+
},
|
257 |
+
{
|
258 |
+
"epoch": 0.21,
|
259 |
+
"grad_norm": 0.12848879670359908,
|
260 |
+
"learning_rate": 0.00018789751908224338,
|
261 |
+
"loss": 0.4972,
|
262 |
+
"step": 35
|
263 |
+
},
|
264 |
+
{
|
265 |
+
"epoch": 0.21,
|
266 |
+
"grad_norm": 0.11904721819683833,
|
267 |
+
"learning_rate": 0.00018693202744395827,
|
268 |
+
"loss": 0.505,
|
269 |
+
"step": 36
|
270 |
+
},
|
271 |
+
{
|
272 |
+
"epoch": 0.22,
|
273 |
+
"grad_norm": 0.12249852034224981,
|
274 |
+
"learning_rate": 0.00018593216805796612,
|
275 |
+
"loss": 0.5396,
|
276 |
+
"step": 37
|
277 |
+
},
|
278 |
+
{
|
279 |
+
"epoch": 0.23,
|
280 |
+
"grad_norm": 0.12453395046646995,
|
281 |
+
"learning_rate": 0.00018489833620913642,
|
282 |
+
"loss": 0.4917,
|
283 |
+
"step": 38
|
284 |
+
},
|
285 |
+
{
|
286 |
+
"epoch": 0.23,
|
287 |
+
"grad_norm": 0.12585770374422164,
|
288 |
+
"learning_rate": 0.00018383094061302766,
|
289 |
+
"loss": 0.5079,
|
290 |
+
"step": 39
|
291 |
+
},
|
292 |
+
{
|
293 |
+
"epoch": 0.24,
|
294 |
+
"grad_norm": 0.11095271476322731,
|
295 |
+
"learning_rate": 0.00018273040325430574,
|
296 |
+
"loss": 0.4812,
|
297 |
+
"step": 40
|
298 |
+
},
|
299 |
+
{
|
300 |
+
"epoch": 0.24,
|
301 |
+
"grad_norm": 0.12968115101635422,
|
302 |
+
"learning_rate": 0.00018159715921991612,
|
303 |
+
"loss": 0.5106,
|
304 |
+
"step": 41
|
305 |
+
},
|
306 |
+
{
|
307 |
+
"epoch": 0.25,
|
308 |
+
"grad_norm": 0.10933018515590627,
|
309 |
+
"learning_rate": 0.00018043165652707649,
|
310 |
+
"loss": 0.4403,
|
311 |
+
"step": 42
|
312 |
+
},
|
313 |
+
{
|
314 |
+
"epoch": 0.25,
|
315 |
+
"eval_loss": 0.9767947196960449,
|
316 |
+
"eval_runtime": 152.0343,
|
317 |
+
"eval_samples_per_second": 1.638,
|
318 |
+
"eval_steps_per_second": 0.822,
|
319 |
+
"step": 42
|
320 |
+
},
|
321 |
+
{
|
322 |
+
"epoch": 0.26,
|
323 |
+
"grad_norm": 0.11525719626792096,
|
324 |
+
"learning_rate": 0.00017923435594615744,
|
325 |
+
"loss": 0.482,
|
326 |
+
"step": 43
|
327 |
+
},
|
328 |
+
{
|
329 |
+
"epoch": 0.26,
|
330 |
+
"grad_norm": 0.12962154411778218,
|
331 |
+
"learning_rate": 0.00017800573081852122,
|
332 |
+
"loss": 0.5452,
|
333 |
+
"step": 44
|
334 |
+
},
|
335 |
+
{
|
336 |
+
"epoch": 0.27,
|
337 |
+
"grad_norm": 0.12555700120045588,
|
338 |
+
"learning_rate": 0.0001767462668693908,
|
339 |
+
"loss": 0.5084,
|
340 |
+
"step": 45
|
341 |
+
},
|
342 |
+
{
|
343 |
+
"epoch": 0.27,
|
344 |
+
"grad_norm": 0.11427565378293324,
|
345 |
+
"learning_rate": 0.00017545646201582303,
|
346 |
+
"loss": 0.5191,
|
347 |
+
"step": 46
|
348 |
+
},
|
349 |
+
{
|
350 |
+
"epoch": 0.28,
|
351 |
+
"grad_norm": 0.10974901402857151,
|
352 |
+
"learning_rate": 0.00017413682616986185,
|
353 |
+
"loss": 0.4703,
|
354 |
+
"step": 47
|
355 |
+
},
|
356 |
+
{
|
357 |
+
"epoch": 0.29,
|
358 |
+
"grad_norm": 0.11781465084480325,
|
359 |
+
"learning_rate": 0.00017278788103694943,
|
360 |
+
"loss": 0.4548,
|
361 |
+
"step": 48
|
362 |
+
},
|
363 |
+
{
|
364 |
+
"epoch": 0.29,
|
365 |
+
"grad_norm": 0.10781807228559999,
|
366 |
+
"learning_rate": 0.000171410159909675,
|
367 |
+
"loss": 0.476,
|
368 |
+
"step": 49
|
369 |
+
},
|
370 |
+
{
|
371 |
+
"epoch": 0.3,
|
372 |
+
"grad_norm": 0.12502639462035098,
|
373 |
+
"learning_rate": 0.00017000420745694254,
|
374 |
+
"loss": 0.5084,
|
375 |
+
"step": 50
|
376 |
+
},
|
377 |
+
{
|
378 |
+
"epoch": 0.3,
|
379 |
+
"grad_norm": 0.10718920826593327,
|
380 |
+
"learning_rate": 0.00016857057950864132,
|
381 |
+
"loss": 0.5093,
|
382 |
+
"step": 51
|
383 |
+
},
|
384 |
+
{
|
385 |
+
"epoch": 0.31,
|
386 |
+
"grad_norm": 0.10040549880547282,
|
387 |
+
"learning_rate": 0.0001671098428359037,
|
388 |
+
"loss": 0.4644,
|
389 |
+
"step": 52
|
390 |
+
},
|
391 |
+
{
|
392 |
+
"epoch": 0.32,
|
393 |
+
"grad_norm": 0.11778478994740472,
|
394 |
+
"learning_rate": 0.00016562257492703757,
|
395 |
+
"loss": 0.4725,
|
396 |
+
"step": 53
|
397 |
+
},
|
398 |
+
{
|
399 |
+
"epoch": 0.32,
|
400 |
+
"grad_norm": 0.1008386031049932,
|
401 |
+
"learning_rate": 0.000164109363759222,
|
402 |
+
"loss": 0.5121,
|
403 |
+
"step": 54
|
404 |
+
},
|
405 |
+
{
|
406 |
+
"epoch": 0.33,
|
407 |
+
"grad_norm": 0.1170302528140235,
|
408 |
+
"learning_rate": 0.000162570807566056,
|
409 |
+
"loss": 0.4766,
|
410 |
+
"step": 55
|
411 |
+
},
|
412 |
+
{
|
413 |
+
"epoch": 0.33,
|
414 |
+
"grad_norm": 0.1104526773884303,
|
415 |
+
"learning_rate": 0.00016100751460105243,
|
416 |
+
"loss": 0.4886,
|
417 |
+
"step": 56
|
418 |
+
},
|
419 |
+
{
|
420 |
+
"epoch": 0.34,
|
421 |
+
"grad_norm": 0.10467920768691032,
|
422 |
+
"learning_rate": 0.00015942010289717105,
|
423 |
+
"loss": 0.4703,
|
424 |
+
"step": 57
|
425 |
+
},
|
426 |
+
{
|
427 |
+
"epoch": 0.34,
|
428 |
+
"grad_norm": 0.11551406829220555,
|
429 |
+
"learning_rate": 0.00015780920002248484,
|
430 |
+
"loss": 0.4837,
|
431 |
+
"step": 58
|
432 |
+
},
|
433 |
+
{
|
434 |
+
"epoch": 0.35,
|
435 |
+
"grad_norm": 0.11133818831887894,
|
436 |
+
"learning_rate": 0.0001561754428320771,
|
437 |
+
"loss": 0.5148,
|
438 |
+
"step": 59
|
439 |
+
},
|
440 |
+
{
|
441 |
+
"epoch": 0.36,
|
442 |
+
"grad_norm": 0.11281448423273216,
|
443 |
+
"learning_rate": 0.00015451947721626676,
|
444 |
+
"loss": 0.4561,
|
445 |
+
"step": 60
|
446 |
+
},
|
447 |
+
{
|
448 |
+
"epoch": 0.36,
|
449 |
+
"grad_norm": 0.13934126997471205,
|
450 |
+
"learning_rate": 0.00015284195784526195,
|
451 |
+
"loss": 0.5069,
|
452 |
+
"step": 61
|
453 |
+
},
|
454 |
+
{
|
455 |
+
"epoch": 0.37,
|
456 |
+
"grad_norm": 0.11851655387640142,
|
457 |
+
"learning_rate": 0.00015114354791034225,
|
458 |
+
"loss": 0.5094,
|
459 |
+
"step": 62
|
460 |
+
},
|
461 |
+
{
|
462 |
+
"epoch": 0.37,
|
463 |
+
"grad_norm": 0.12909148374566123,
|
464 |
+
"learning_rate": 0.0001494249188616723,
|
465 |
+
"loss": 0.581,
|
466 |
+
"step": 63
|
467 |
+
},
|
468 |
+
{
|
469 |
+
"epoch": 0.38,
|
470 |
+
"grad_norm": 0.11070161341925377,
|
471 |
+
"learning_rate": 0.00014768675014285062,
|
472 |
+
"loss": 0.4585,
|
473 |
+
"step": 64
|
474 |
+
},
|
475 |
+
{
|
476 |
+
"epoch": 0.39,
|
477 |
+
"grad_norm": 0.13308674882888374,
|
478 |
+
"learning_rate": 0.00014592972892229778,
|
479 |
+
"loss": 0.4974,
|
480 |
+
"step": 65
|
481 |
+
},
|
482 |
+
{
|
483 |
+
"epoch": 0.39,
|
484 |
+
"grad_norm": 0.12124588853708144,
|
485 |
+
"learning_rate": 0.0001441545498215912,
|
486 |
+
"loss": 0.4463,
|
487 |
+
"step": 66
|
488 |
+
},
|
489 |
+
{
|
490 |
+
"epoch": 0.4,
|
491 |
+
"grad_norm": 0.1183570515369953,
|
492 |
+
"learning_rate": 0.00014236191464085286,
|
493 |
+
"loss": 0.447,
|
494 |
+
"step": 67
|
495 |
+
},
|
496 |
+
{
|
497 |
+
"epoch": 0.4,
|
498 |
+
"grad_norm": 0.13520024884417237,
|
499 |
+
"learning_rate": 0.00014055253208129938,
|
500 |
+
"loss": 0.5309,
|
501 |
+
"step": 68
|
502 |
+
},
|
503 |
+
{
|
504 |
+
"epoch": 0.41,
|
505 |
+
"grad_norm": 0.12184981458813801,
|
506 |
+
"learning_rate": 0.00013872711746506413,
|
507 |
+
"loss": 0.4532,
|
508 |
+
"step": 69
|
509 |
+
},
|
510 |
+
{
|
511 |
+
"epoch": 0.42,
|
512 |
+
"grad_norm": 0.12449299540645078,
|
513 |
+
"learning_rate": 0.00013688639245240078,
|
514 |
+
"loss": 0.5198,
|
515 |
+
"step": 70
|
516 |
+
},
|
517 |
+
{
|
518 |
+
"epoch": 0.42,
|
519 |
+
"grad_norm": 0.1383134750490429,
|
520 |
+
"learning_rate": 0.00013503108475638244,
|
521 |
+
"loss": 0.5629,
|
522 |
+
"step": 71
|
523 |
+
},
|
524 |
+
{
|
525 |
+
"epoch": 0.43,
|
526 |
+
"grad_norm": 0.246237001656926,
|
527 |
+
"learning_rate": 0.0001331619278552068,
|
528 |
+
"loss": 0.4869,
|
529 |
+
"step": 72
|
530 |
+
},
|
531 |
+
{
|
532 |
+
"epoch": 0.43,
|
533 |
+
"grad_norm": 0.13337703940933632,
|
534 |
+
"learning_rate": 0.00013127966070222274,
|
535 |
+
"loss": 0.4792,
|
536 |
+
"step": 73
|
537 |
+
},
|
538 |
+
{
|
539 |
+
"epoch": 0.44,
|
540 |
+
"grad_norm": 0.12428922033806454,
|
541 |
+
"learning_rate": 0.00012938502743379212,
|
542 |
+
"loss": 0.4825,
|
543 |
+
"step": 74
|
544 |
+
},
|
545 |
+
{
|
546 |
+
"epoch": 0.45,
|
547 |
+
"grad_norm": 0.13290774912900208,
|
548 |
+
"learning_rate": 0.00012747877707510252,
|
549 |
+
"loss": 0.5138,
|
550 |
+
"step": 75
|
551 |
+
},
|
552 |
+
{
|
553 |
+
"epoch": 0.45,
|
554 |
+
"grad_norm": 0.11185975046756892,
|
555 |
+
"learning_rate": 0.0001255616632440475,
|
556 |
+
"loss": 0.4815,
|
557 |
+
"step": 76
|
558 |
+
},
|
559 |
+
{
|
560 |
+
"epoch": 0.46,
|
561 |
+
"grad_norm": 0.1130592868215497,
|
562 |
+
"learning_rate": 0.0001236344438532905,
|
563 |
+
"loss": 0.5046,
|
564 |
+
"step": 77
|
565 |
+
},
|
566 |
+
{
|
567 |
+
"epoch": 0.46,
|
568 |
+
"grad_norm": 0.12882943465594857,
|
569 |
+
"learning_rate": 0.0001216978808106318,
|
570 |
+
"loss": 0.5091,
|
571 |
+
"step": 78
|
572 |
+
},
|
573 |
+
{
|
574 |
+
"epoch": 0.47,
|
575 |
+
"grad_norm": 0.14837896297082676,
|
576 |
+
"learning_rate": 0.00011975273971779528,
|
577 |
+
"loss": 0.5158,
|
578 |
+
"step": 79
|
579 |
+
},
|
580 |
+
{
|
581 |
+
"epoch": 0.48,
|
582 |
+
"grad_norm": 0.1265223309856292,
|
583 |
+
"learning_rate": 0.00011779978956775506,
|
584 |
+
"loss": 0.5068,
|
585 |
+
"step": 80
|
586 |
+
},
|
587 |
+
{
|
588 |
+
"epoch": 0.48,
|
589 |
+
"grad_norm": 0.14042502330520407,
|
590 |
+
"learning_rate": 0.0001158398024407215,
|
591 |
+
"loss": 0.5061,
|
592 |
+
"step": 81
|
593 |
+
},
|
594 |
+
{
|
595 |
+
"epoch": 0.49,
|
596 |
+
"grad_norm": 0.1261526695491767,
|
597 |
+
"learning_rate": 0.00011387355319890685,
|
598 |
+
"loss": 0.4691,
|
599 |
+
"step": 82
|
600 |
+
},
|
601 |
+
{
|
602 |
+
"epoch": 0.49,
|
603 |
+
"grad_norm": 0.12007305451001854,
|
604 |
+
"learning_rate": 0.00011190181918019049,
|
605 |
+
"loss": 0.4753,
|
606 |
+
"step": 83
|
607 |
+
},
|
608 |
+
{
|
609 |
+
"epoch": 0.5,
|
610 |
+
"grad_norm": 0.12809956897166885,
|
611 |
+
"learning_rate": 0.00010992537989080618,
|
612 |
+
"loss": 0.4417,
|
613 |
+
"step": 84
|
614 |
+
},
|
615 |
+
{
|
616 |
+
"epoch": 0.5,
|
617 |
+
"eval_loss": 0.9675251841545105,
|
618 |
+
"eval_runtime": 152.4793,
|
619 |
+
"eval_samples_per_second": 1.633,
|
620 |
+
"eval_steps_per_second": 0.82,
|
621 |
+
"step": 84
|
622 |
+
},
|
623 |
+
{
|
624 |
+
"epoch": 0.51,
|
625 |
+
"grad_norm": 0.11858329804793687,
|
626 |
+
"learning_rate": 0.00010794501669717145,
|
627 |
+
"loss": 0.4868,
|
628 |
+
"step": 85
|
629 |
+
},
|
630 |
+
{
|
631 |
+
"epoch": 0.51,
|
632 |
+
"grad_norm": 0.10984649953887334,
|
633 |
+
"learning_rate": 0.00010596151251698199,
|
634 |
+
"loss": 0.4598,
|
635 |
+
"step": 86
|
636 |
+
},
|
637 |
+
{
|
638 |
+
"epoch": 0.52,
|
639 |
+
"grad_norm": 0.10927203986256682,
|
640 |
+
"learning_rate": 0.0001039756515096926,
|
641 |
+
"loss": 0.4693,
|
642 |
+
"step": 87
|
643 |
+
},
|
644 |
+
{
|
645 |
+
"epoch": 0.52,
|
646 |
+
"grad_norm": 0.11205046531522328,
|
647 |
+
"learning_rate": 0.00010198821876650701,
|
648 |
+
"loss": 0.4921,
|
649 |
+
"step": 88
|
650 |
+
},
|
651 |
+
{
|
652 |
+
"epoch": 0.53,
|
653 |
+
"grad_norm": 0.13232347270009215,
|
654 |
+
"learning_rate": 0.0001,
|
655 |
+
"loss": 0.4695,
|
656 |
+
"step": 89
|
657 |
+
},
|
658 |
+
{
|
659 |
+
"epoch": 0.53,
|
660 |
+
"grad_norm": 0.12136881873560385,
|
661 |
+
"learning_rate": 9.801178123349298e-05,
|
662 |
+
"loss": 0.4859,
|
663 |
+
"step": 90
|
664 |
+
},
|
665 |
+
{
|
666 |
+
"epoch": 0.54,
|
667 |
+
"grad_norm": 0.14347476421156694,
|
668 |
+
"learning_rate": 9.602434849030745e-05,
|
669 |
+
"loss": 0.4796,
|
670 |
+
"step": 91
|
671 |
+
},
|
672 |
+
{
|
673 |
+
"epoch": 0.55,
|
674 |
+
"grad_norm": 0.13956845267055204,
|
675 |
+
"learning_rate": 9.403848748301802e-05,
|
676 |
+
"loss": 0.5339,
|
677 |
+
"step": 92
|
678 |
+
},
|
679 |
+
{
|
680 |
+
"epoch": 0.55,
|
681 |
+
"grad_norm": 0.12814010903196785,
|
682 |
+
"learning_rate": 9.205498330282856e-05,
|
683 |
+
"loss": 0.5267,
|
684 |
+
"step": 93
|
685 |
+
},
|
686 |
+
{
|
687 |
+
"epoch": 0.56,
|
688 |
+
"grad_norm": 0.12798850330908082,
|
689 |
+
"learning_rate": 9.007462010919386e-05,
|
690 |
+
"loss": 0.4604,
|
691 |
+
"step": 94
|
692 |
+
},
|
693 |
+
{
|
694 |
+
"epoch": 0.56,
|
695 |
+
"grad_norm": 0.13673366056605873,
|
696 |
+
"learning_rate": 8.809818081980953e-05,
|
697 |
+
"loss": 0.49,
|
698 |
+
"step": 95
|
699 |
+
},
|
700 |
+
{
|
701 |
+
"epoch": 0.57,
|
702 |
+
"grad_norm": 0.12607483394599764,
|
703 |
+
"learning_rate": 8.612644680109319e-05,
|
704 |
+
"loss": 0.4774,
|
705 |
+
"step": 96
|
706 |
+
},
|
707 |
+
{
|
708 |
+
"epoch": 0.58,
|
709 |
+
"grad_norm": 0.1365629261848207,
|
710 |
+
"learning_rate": 8.416019755927851e-05,
|
711 |
+
"loss": 0.4827,
|
712 |
+
"step": 97
|
713 |
+
},
|
714 |
+
{
|
715 |
+
"epoch": 0.58,
|
716 |
+
"grad_norm": 0.12122559291940836,
|
717 |
+
"learning_rate": 8.2200210432245e-05,
|
718 |
+
"loss": 0.5044,
|
719 |
+
"step": 98
|
720 |
+
},
|
721 |
+
{
|
722 |
+
"epoch": 0.59,
|
723 |
+
"grad_norm": 0.11655390642565265,
|
724 |
+
"learning_rate": 8.024726028220474e-05,
|
725 |
+
"loss": 0.503,
|
726 |
+
"step": 99
|
727 |
+
},
|
728 |
+
{
|
729 |
+
"epoch": 0.59,
|
730 |
+
"grad_norm": 0.12394574502796742,
|
731 |
+
"learning_rate": 7.83021191893682e-05,
|
732 |
+
"loss": 0.491,
|
733 |
+
"step": 100
|
734 |
+
},
|
735 |
+
{
|
736 |
+
"epoch": 0.6,
|
737 |
+
"grad_norm": 0.14922841699852962,
|
738 |
+
"learning_rate": 7.636555614670953e-05,
|
739 |
+
"loss": 0.457,
|
740 |
+
"step": 101
|
741 |
+
},
|
742 |
+
{
|
743 |
+
"epoch": 0.61,
|
744 |
+
"grad_norm": 0.11076924096187928,
|
745 |
+
"learning_rate": 7.443833675595255e-05,
|
746 |
+
"loss": 0.4603,
|
747 |
+
"step": 102
|
748 |
+
},
|
749 |
+
{
|
750 |
+
"epoch": 0.61,
|
751 |
+
"grad_norm": 0.1263594611752413,
|
752 |
+
"learning_rate": 7.252122292489747e-05,
|
753 |
+
"loss": 0.4859,
|
754 |
+
"step": 103
|
755 |
+
},
|
756 |
+
{
|
757 |
+
"epoch": 0.62,
|
758 |
+
"grad_norm": 0.11432575178505003,
|
759 |
+
"learning_rate": 7.061497256620793e-05,
|
760 |
+
"loss": 0.4627,
|
761 |
+
"step": 104
|
762 |
+
},
|
763 |
+
{
|
764 |
+
"epoch": 0.62,
|
765 |
+
"grad_norm": 0.1078119391965793,
|
766 |
+
"learning_rate": 6.87203392977773e-05,
|
767 |
+
"loss": 0.4829,
|
768 |
+
"step": 105
|
769 |
+
},
|
770 |
+
{
|
771 |
+
"epoch": 0.63,
|
772 |
+
"grad_norm": 0.12752089816514908,
|
773 |
+
"learning_rate": 6.683807214479323e-05,
|
774 |
+
"loss": 0.46,
|
775 |
+
"step": 106
|
776 |
+
},
|
777 |
+
{
|
778 |
+
"epoch": 0.64,
|
779 |
+
"grad_norm": 0.11421623043902956,
|
780 |
+
"learning_rate": 6.496891524361757e-05,
|
781 |
+
"loss": 0.4429,
|
782 |
+
"step": 107
|
783 |
+
},
|
784 |
+
{
|
785 |
+
"epoch": 0.64,
|
786 |
+
"grad_norm": 0.10432253193399477,
|
787 |
+
"learning_rate": 6.311360754759923e-05,
|
788 |
+
"loss": 0.402,
|
789 |
+
"step": 108
|
790 |
+
},
|
791 |
+
{
|
792 |
+
"epoch": 0.65,
|
793 |
+
"grad_norm": 0.12155248673662734,
|
794 |
+
"learning_rate": 6.127288253493591e-05,
|
795 |
+
"loss": 0.5126,
|
796 |
+
"step": 109
|
797 |
+
},
|
798 |
+
{
|
799 |
+
"epoch": 0.65,
|
800 |
+
"grad_norm": 0.14266947863559803,
|
801 |
+
"learning_rate": 5.9447467918700614e-05,
|
802 |
+
"loss": 0.4821,
|
803 |
+
"step": 110
|
804 |
+
},
|
805 |
+
{
|
806 |
+
"epoch": 0.66,
|
807 |
+
"grad_norm": 0.14851250761112514,
|
808 |
+
"learning_rate": 5.763808535914723e-05,
|
809 |
+
"loss": 0.4891,
|
810 |
+
"step": 111
|
811 |
+
},
|
812 |
+
{
|
813 |
+
"epoch": 0.67,
|
814 |
+
"grad_norm": 0.14264023747361737,
|
815 |
+
"learning_rate": 5.584545017840885e-05,
|
816 |
+
"loss": 0.5181,
|
817 |
+
"step": 112
|
818 |
+
},
|
819 |
+
{
|
820 |
+
"epoch": 0.67,
|
821 |
+
"grad_norm": 0.12837168363458795,
|
822 |
+
"learning_rate": 5.407027107770219e-05,
|
823 |
+
"loss": 0.5599,
|
824 |
+
"step": 113
|
825 |
+
},
|
826 |
+
{
|
827 |
+
"epoch": 0.68,
|
828 |
+
"grad_norm": 0.11874709251257598,
|
829 |
+
"learning_rate": 5.2313249857149414e-05,
|
830 |
+
"loss": 0.4536,
|
831 |
+
"step": 114
|
832 |
+
},
|
833 |
+
{
|
834 |
+
"epoch": 0.68,
|
835 |
+
"grad_norm": 0.12010754957532713,
|
836 |
+
"learning_rate": 5.0575081138327715e-05,
|
837 |
+
"loss": 0.5004,
|
838 |
+
"step": 115
|
839 |
+
},
|
840 |
+
{
|
841 |
+
"epoch": 0.69,
|
842 |
+
"grad_norm": 0.13464124440677885,
|
843 |
+
"learning_rate": 4.885645208965779e-05,
|
844 |
+
"loss": 0.4985,
|
845 |
+
"step": 116
|
846 |
+
},
|
847 |
+
{
|
848 |
+
"epoch": 0.7,
|
849 |
+
"grad_norm": 0.13701854261941088,
|
850 |
+
"learning_rate": 4.715804215473809e-05,
|
851 |
+
"loss": 0.4709,
|
852 |
+
"step": 117
|
853 |
+
},
|
854 |
+
{
|
855 |
+
"epoch": 0.7,
|
856 |
+
"grad_norm": 0.1335483738873249,
|
857 |
+
"learning_rate": 4.548052278373327e-05,
|
858 |
+
"loss": 0.4735,
|
859 |
+
"step": 118
|
860 |
+
},
|
861 |
+
{
|
862 |
+
"epoch": 0.71,
|
863 |
+
"grad_norm": 0.13603172024059101,
|
864 |
+
"learning_rate": 4.382455716792291e-05,
|
865 |
+
"loss": 0.4721,
|
866 |
+
"step": 119
|
867 |
+
},
|
868 |
+
{
|
869 |
+
"epoch": 0.71,
|
870 |
+
"grad_norm": 0.13843339239058639,
|
871 |
+
"learning_rate": 4.219079997751515e-05,
|
872 |
+
"loss": 0.4954,
|
873 |
+
"step": 120
|
874 |
+
},
|
875 |
+
{
|
876 |
+
"epoch": 0.72,
|
877 |
+
"grad_norm": 0.15011169526780793,
|
878 |
+
"learning_rate": 4.0579897102828966e-05,
|
879 |
+
"loss": 0.4648,
|
880 |
+
"step": 121
|
881 |
+
},
|
882 |
+
{
|
883 |
+
"epoch": 0.73,
|
884 |
+
"grad_norm": 0.13061595453081623,
|
885 |
+
"learning_rate": 3.899248539894757e-05,
|
886 |
+
"loss": 0.4801,
|
887 |
+
"step": 122
|
888 |
+
},
|
889 |
+
{
|
890 |
+
"epoch": 0.73,
|
891 |
+
"grad_norm": 0.14067787924603412,
|
892 |
+
"learning_rate": 3.7429192433944014e-05,
|
893 |
+
"loss": 0.4805,
|
894 |
+
"step": 123
|
895 |
+
},
|
896 |
+
{
|
897 |
+
"epoch": 0.74,
|
898 |
+
"grad_norm": 0.13420057703295998,
|
899 |
+
"learning_rate": 3.589063624077802e-05,
|
900 |
+
"loss": 0.4446,
|
901 |
+
"step": 124
|
902 |
+
},
|
903 |
+
{
|
904 |
+
"epoch": 0.74,
|
905 |
+
"grad_norm": 0.14083737654873127,
|
906 |
+
"learning_rate": 3.4377425072962465e-05,
|
907 |
+
"loss": 0.46,
|
908 |
+
"step": 125
|
909 |
+
},
|
910 |
+
{
|
911 |
+
"epoch": 0.75,
|
912 |
+
"grad_norm": 0.13231889777376862,
|
913 |
+
"learning_rate": 3.289015716409631e-05,
|
914 |
+
"loss": 0.4451,
|
915 |
+
"step": 126
|
916 |
+
},
|
917 |
+
{
|
918 |
+
"epoch": 0.75,
|
919 |
+
"eval_loss": 0.9651579260826111,
|
920 |
+
"eval_runtime": 155.5959,
|
921 |
+
"eval_samples_per_second": 1.6,
|
922 |
+
"eval_steps_per_second": 0.803,
|
923 |
+
"step": 126
|
924 |
+
}
|
925 |
+
],
|
926 |
+
"logging_steps": 1,
|
927 |
+
"max_steps": 168,
|
928 |
+
"num_input_tokens_seen": 0,
|
929 |
+
"num_train_epochs": 1,
|
930 |
+
"save_steps": 42,
|
931 |
+
"total_flos": 2.5549258322041897e+18,
|
932 |
+
"train_batch_size": 1,
|
933 |
+
"trial_name": null,
|
934 |
+
"trial_params": null
|
935 |
+
}
|
checkpoint-126/training_args.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:cbfaffe231fe183758cc2353296944fef75710807fa8ea7663978d9e906d00f7
|
3 |
+
size 6968
|
checkpoint-126/zero_to_fp32.py
ADDED
@@ -0,0 +1,592 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python
|
2 |
+
|
3 |
+
# Copyright (c) Microsoft Corporation.
|
4 |
+
# SPDX-License-Identifier: Apache-2.0
|
5 |
+
|
6 |
+
# DeepSpeed Team
|
7 |
+
|
8 |
+
# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
|
9 |
+
# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
|
10 |
+
# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
|
11 |
+
# application.
|
12 |
+
#
|
13 |
+
# example: python zero_to_fp32.py . pytorch_model.bin
|
14 |
+
|
15 |
+
import argparse
|
16 |
+
import torch
|
17 |
+
import glob
|
18 |
+
import math
|
19 |
+
import os
|
20 |
+
import re
|
21 |
+
from collections import OrderedDict
|
22 |
+
from dataclasses import dataclass
|
23 |
+
|
24 |
+
# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
|
25 |
+
# DeepSpeed data structures it has to be available in the current python environment.
|
26 |
+
from deepspeed.utils import logger
|
27 |
+
from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
|
28 |
+
FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
|
29 |
+
FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
|
30 |
+
|
31 |
+
|
32 |
+
@dataclass
|
33 |
+
class zero_model_state:
|
34 |
+
buffers: dict()
|
35 |
+
param_shapes: dict()
|
36 |
+
shared_params: list
|
37 |
+
ds_version: int
|
38 |
+
frozen_param_shapes: dict()
|
39 |
+
frozen_param_fragments: dict()
|
40 |
+
|
41 |
+
|
42 |
+
debug = 0
|
43 |
+
|
44 |
+
# load to cpu
|
45 |
+
device = torch.device('cpu')
|
46 |
+
|
47 |
+
|
48 |
+
def atoi(text):
|
49 |
+
return int(text) if text.isdigit() else text
|
50 |
+
|
51 |
+
|
52 |
+
def natural_keys(text):
|
53 |
+
'''
|
54 |
+
alist.sort(key=natural_keys) sorts in human order
|
55 |
+
http://nedbatchelder.com/blog/200712/human_sorting.html
|
56 |
+
(See Toothy's implementation in the comments)
|
57 |
+
'''
|
58 |
+
return [atoi(c) for c in re.split(r'(\d+)', text)]
|
59 |
+
|
60 |
+
|
61 |
+
def get_model_state_file(checkpoint_dir, zero_stage):
|
62 |
+
if not os.path.isdir(checkpoint_dir):
|
63 |
+
raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
|
64 |
+
|
65 |
+
# there should be only one file
|
66 |
+
if zero_stage <= 2:
|
67 |
+
file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
|
68 |
+
elif zero_stage == 3:
|
69 |
+
file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
|
70 |
+
|
71 |
+
if not os.path.exists(file):
|
72 |
+
raise FileNotFoundError(f"can't find model states file at '{file}'")
|
73 |
+
|
74 |
+
return file
|
75 |
+
|
76 |
+
|
77 |
+
def get_checkpoint_files(checkpoint_dir, glob_pattern):
|
78 |
+
# XXX: need to test that this simple glob rule works for multi-node setup too
|
79 |
+
ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
|
80 |
+
|
81 |
+
if len(ckpt_files) == 0:
|
82 |
+
raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
|
83 |
+
|
84 |
+
return ckpt_files
|
85 |
+
|
86 |
+
|
87 |
+
def get_optim_files(checkpoint_dir):
|
88 |
+
return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
|
89 |
+
|
90 |
+
|
91 |
+
def get_model_state_files(checkpoint_dir):
|
92 |
+
return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
|
93 |
+
|
94 |
+
|
95 |
+
def parse_model_states(files):
|
96 |
+
zero_model_states = []
|
97 |
+
for file in files:
|
98 |
+
state_dict = torch.load(file, map_location=device)
|
99 |
+
|
100 |
+
if BUFFER_NAMES not in state_dict:
|
101 |
+
raise ValueError(f"{file} is not a model state checkpoint")
|
102 |
+
buffer_names = state_dict[BUFFER_NAMES]
|
103 |
+
if debug:
|
104 |
+
print("Found buffers:", buffer_names)
|
105 |
+
|
106 |
+
# recover just the buffers while restoring them to fp32 if they were saved in fp16
|
107 |
+
buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
|
108 |
+
param_shapes = state_dict[PARAM_SHAPES]
|
109 |
+
|
110 |
+
# collect parameters that are included in param_shapes
|
111 |
+
param_names = []
|
112 |
+
for s in param_shapes:
|
113 |
+
for name in s.keys():
|
114 |
+
param_names.append(name)
|
115 |
+
|
116 |
+
# update with frozen parameters
|
117 |
+
frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
|
118 |
+
if frozen_param_shapes is not None:
|
119 |
+
if debug:
|
120 |
+
print(f"Found frozen_param_shapes: {frozen_param_shapes}")
|
121 |
+
param_names += list(frozen_param_shapes.keys())
|
122 |
+
|
123 |
+
# handle shared params
|
124 |
+
shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
|
125 |
+
|
126 |
+
ds_version = state_dict.get(DS_VERSION, None)
|
127 |
+
|
128 |
+
frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
|
129 |
+
|
130 |
+
z_model_state = zero_model_state(buffers=buffers,
|
131 |
+
param_shapes=param_shapes,
|
132 |
+
shared_params=shared_params,
|
133 |
+
ds_version=ds_version,
|
134 |
+
frozen_param_shapes=frozen_param_shapes,
|
135 |
+
frozen_param_fragments=frozen_param_fragments)
|
136 |
+
zero_model_states.append(z_model_state)
|
137 |
+
|
138 |
+
return zero_model_states
|
139 |
+
|
140 |
+
|
141 |
+
def parse_optim_states(files, ds_checkpoint_dir):
|
142 |
+
|
143 |
+
total_files = len(files)
|
144 |
+
state_dicts = []
|
145 |
+
for f in files:
|
146 |
+
state_dict = torch.load(f, map_location=device)
|
147 |
+
# immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
|
148 |
+
# and also handle the case where it was already removed by another helper script
|
149 |
+
state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
|
150 |
+
state_dicts.append(state_dict)
|
151 |
+
|
152 |
+
if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
|
153 |
+
raise ValueError(f"{files[0]} is not a zero checkpoint")
|
154 |
+
zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
|
155 |
+
world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
|
156 |
+
|
157 |
+
# For ZeRO-2 each param group can have different partition_count as data parallelism for expert
|
158 |
+
# parameters can be different from data parallelism for non-expert parameters. So we can just
|
159 |
+
# use the max of the partition_count to get the dp world_size.
|
160 |
+
|
161 |
+
if type(world_size) is list:
|
162 |
+
world_size = max(world_size)
|
163 |
+
|
164 |
+
if world_size != total_files:
|
165 |
+
raise ValueError(
|
166 |
+
f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
|
167 |
+
"Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
|
168 |
+
)
|
169 |
+
|
170 |
+
# the groups are named differently in each stage
|
171 |
+
if zero_stage <= 2:
|
172 |
+
fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
|
173 |
+
elif zero_stage == 3:
|
174 |
+
fp32_groups_key = FP32_FLAT_GROUPS
|
175 |
+
else:
|
176 |
+
raise ValueError(f"unknown zero stage {zero_stage}")
|
177 |
+
|
178 |
+
if zero_stage <= 2:
|
179 |
+
fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
|
180 |
+
elif zero_stage == 3:
|
181 |
+
# if there is more than one param group, there will be multiple flattened tensors - one
|
182 |
+
# flattened tensor per group - for simplicity merge them into a single tensor
|
183 |
+
#
|
184 |
+
# XXX: could make the script more memory efficient for when there are multiple groups - it
|
185 |
+
# will require matching the sub-lists of param_shapes for each param group flattened tensor
|
186 |
+
|
187 |
+
fp32_flat_groups = [
|
188 |
+
torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
|
189 |
+
]
|
190 |
+
|
191 |
+
return zero_stage, world_size, fp32_flat_groups
|
192 |
+
|
193 |
+
|
194 |
+
def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
|
195 |
+
"""
|
196 |
+
Returns fp32 state_dict reconstructed from ds checkpoint
|
197 |
+
|
198 |
+
Args:
|
199 |
+
- ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
|
200 |
+
|
201 |
+
"""
|
202 |
+
print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
|
203 |
+
|
204 |
+
optim_files = get_optim_files(ds_checkpoint_dir)
|
205 |
+
zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
|
206 |
+
print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
|
207 |
+
|
208 |
+
model_files = get_model_state_files(ds_checkpoint_dir)
|
209 |
+
|
210 |
+
zero_model_states = parse_model_states(model_files)
|
211 |
+
print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
|
212 |
+
|
213 |
+
if zero_stage <= 2:
|
214 |
+
return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
|
215 |
+
elif zero_stage == 3:
|
216 |
+
return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
|
217 |
+
|
218 |
+
|
219 |
+
def _zero2_merge_frozen_params(state_dict, zero_model_states):
|
220 |
+
if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
|
221 |
+
return
|
222 |
+
|
223 |
+
frozen_param_shapes = zero_model_states[0].frozen_param_shapes
|
224 |
+
frozen_param_fragments = zero_model_states[0].frozen_param_fragments
|
225 |
+
|
226 |
+
if debug:
|
227 |
+
num_elem = sum(s.numel() for s in frozen_param_shapes.values())
|
228 |
+
print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
|
229 |
+
|
230 |
+
wanted_params = len(frozen_param_shapes)
|
231 |
+
wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
|
232 |
+
avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
|
233 |
+
print(f'Frozen params: Have {avail_numel} numels to process.')
|
234 |
+
print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
|
235 |
+
|
236 |
+
total_params = 0
|
237 |
+
total_numel = 0
|
238 |
+
for name, shape in frozen_param_shapes.items():
|
239 |
+
total_params += 1
|
240 |
+
unpartitioned_numel = shape.numel()
|
241 |
+
total_numel += unpartitioned_numel
|
242 |
+
|
243 |
+
state_dict[name] = frozen_param_fragments[name]
|
244 |
+
|
245 |
+
if debug:
|
246 |
+
print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
|
247 |
+
|
248 |
+
print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
|
249 |
+
|
250 |
+
|
251 |
+
def _has_callable(obj, fn):
|
252 |
+
attr = getattr(obj, fn, None)
|
253 |
+
return callable(attr)
|
254 |
+
|
255 |
+
|
256 |
+
def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
|
257 |
+
param_shapes = zero_model_states[0].param_shapes
|
258 |
+
|
259 |
+
# Reconstruction protocol:
|
260 |
+
#
|
261 |
+
# XXX: document this
|
262 |
+
|
263 |
+
if debug:
|
264 |
+
for i in range(world_size):
|
265 |
+
for j in range(len(fp32_flat_groups[0])):
|
266 |
+
print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
|
267 |
+
|
268 |
+
# XXX: memory usage doubles here (zero2)
|
269 |
+
num_param_groups = len(fp32_flat_groups[0])
|
270 |
+
merged_single_partition_of_fp32_groups = []
|
271 |
+
for i in range(num_param_groups):
|
272 |
+
merged_partitions = [sd[i] for sd in fp32_flat_groups]
|
273 |
+
full_single_fp32_vector = torch.cat(merged_partitions, 0)
|
274 |
+
merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
|
275 |
+
avail_numel = sum(
|
276 |
+
[full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
|
277 |
+
|
278 |
+
if debug:
|
279 |
+
wanted_params = sum([len(shapes) for shapes in param_shapes])
|
280 |
+
wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
|
281 |
+
# not asserting if there is a mismatch due to possible padding
|
282 |
+
print(f"Have {avail_numel} numels to process.")
|
283 |
+
print(f"Need {wanted_numel} numels in {wanted_params} params.")
|
284 |
+
|
285 |
+
# params
|
286 |
+
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
287 |
+
# out-of-core computing solution
|
288 |
+
total_numel = 0
|
289 |
+
total_params = 0
|
290 |
+
for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
|
291 |
+
offset = 0
|
292 |
+
avail_numel = full_single_fp32_vector.numel()
|
293 |
+
for name, shape in shapes.items():
|
294 |
+
|
295 |
+
unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
|
296 |
+
total_numel += unpartitioned_numel
|
297 |
+
total_params += 1
|
298 |
+
|
299 |
+
if debug:
|
300 |
+
print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
|
301 |
+
state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
|
302 |
+
offset += unpartitioned_numel
|
303 |
+
|
304 |
+
# Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
|
305 |
+
# avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
|
306 |
+
# paddings performed in the code it's almost impossible to predict the exact numbers w/o the
|
307 |
+
# live optimizer object, so we are checking that the numbers are within the right range
|
308 |
+
align_to = 2 * world_size
|
309 |
+
|
310 |
+
def zero2_align(x):
|
311 |
+
return align_to * math.ceil(x / align_to)
|
312 |
+
|
313 |
+
if debug:
|
314 |
+
print(f"original offset={offset}, avail_numel={avail_numel}")
|
315 |
+
|
316 |
+
offset = zero2_align(offset)
|
317 |
+
avail_numel = zero2_align(avail_numel)
|
318 |
+
|
319 |
+
if debug:
|
320 |
+
print(f"aligned offset={offset}, avail_numel={avail_numel}")
|
321 |
+
|
322 |
+
# Sanity check
|
323 |
+
if offset != avail_numel:
|
324 |
+
raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
325 |
+
|
326 |
+
print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
|
327 |
+
|
328 |
+
|
329 |
+
def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
|
330 |
+
state_dict = OrderedDict()
|
331 |
+
|
332 |
+
# buffers
|
333 |
+
buffers = zero_model_states[0].buffers
|
334 |
+
state_dict.update(buffers)
|
335 |
+
if debug:
|
336 |
+
print(f"added {len(buffers)} buffers")
|
337 |
+
|
338 |
+
_zero2_merge_frozen_params(state_dict, zero_model_states)
|
339 |
+
|
340 |
+
_zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
|
341 |
+
|
342 |
+
# recover shared parameters
|
343 |
+
for pair in zero_model_states[0].shared_params:
|
344 |
+
if pair[1] in state_dict:
|
345 |
+
state_dict[pair[0]] = state_dict[pair[1]]
|
346 |
+
|
347 |
+
return state_dict
|
348 |
+
|
349 |
+
|
350 |
+
def zero3_partitioned_param_info(unpartitioned_numel, world_size):
|
351 |
+
remainder = unpartitioned_numel % world_size
|
352 |
+
padding_numel = (world_size - remainder) if remainder else 0
|
353 |
+
partitioned_numel = math.ceil(unpartitioned_numel / world_size)
|
354 |
+
return partitioned_numel, padding_numel
|
355 |
+
|
356 |
+
|
357 |
+
def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
|
358 |
+
if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
|
359 |
+
return
|
360 |
+
|
361 |
+
if debug:
|
362 |
+
for i in range(world_size):
|
363 |
+
num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
|
364 |
+
print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
|
365 |
+
|
366 |
+
frozen_param_shapes = zero_model_states[0].frozen_param_shapes
|
367 |
+
wanted_params = len(frozen_param_shapes)
|
368 |
+
wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
|
369 |
+
avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
|
370 |
+
print(f'Frozen params: Have {avail_numel} numels to process.')
|
371 |
+
print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
|
372 |
+
|
373 |
+
total_params = 0
|
374 |
+
total_numel = 0
|
375 |
+
for name, shape in zero_model_states[0].frozen_param_shapes.items():
|
376 |
+
total_params += 1
|
377 |
+
unpartitioned_numel = shape.numel()
|
378 |
+
total_numel += unpartitioned_numel
|
379 |
+
|
380 |
+
param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
|
381 |
+
state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
|
382 |
+
|
383 |
+
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
384 |
+
|
385 |
+
if debug:
|
386 |
+
print(
|
387 |
+
f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
388 |
+
)
|
389 |
+
|
390 |
+
print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
|
391 |
+
|
392 |
+
|
393 |
+
def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
|
394 |
+
param_shapes = zero_model_states[0].param_shapes
|
395 |
+
avail_numel = fp32_flat_groups[0].numel() * world_size
|
396 |
+
# Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
|
397 |
+
# param, re-consolidating each param, while dealing with padding if any
|
398 |
+
|
399 |
+
# merge list of dicts, preserving order
|
400 |
+
param_shapes = {k: v for d in param_shapes for k, v in d.items()}
|
401 |
+
|
402 |
+
if debug:
|
403 |
+
for i in range(world_size):
|
404 |
+
print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
|
405 |
+
|
406 |
+
wanted_params = len(param_shapes)
|
407 |
+
wanted_numel = sum(shape.numel() for shape in param_shapes.values())
|
408 |
+
# not asserting if there is a mismatch due to possible padding
|
409 |
+
avail_numel = fp32_flat_groups[0].numel() * world_size
|
410 |
+
print(f"Trainable params: Have {avail_numel} numels to process.")
|
411 |
+
print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
|
412 |
+
|
413 |
+
# params
|
414 |
+
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
415 |
+
# out-of-core computing solution
|
416 |
+
offset = 0
|
417 |
+
total_numel = 0
|
418 |
+
total_params = 0
|
419 |
+
for name, shape in param_shapes.items():
|
420 |
+
|
421 |
+
unpartitioned_numel = shape.numel()
|
422 |
+
total_numel += unpartitioned_numel
|
423 |
+
total_params += 1
|
424 |
+
|
425 |
+
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
426 |
+
|
427 |
+
if debug:
|
428 |
+
print(
|
429 |
+
f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
430 |
+
)
|
431 |
+
|
432 |
+
# XXX: memory usage doubles here
|
433 |
+
state_dict[name] = torch.cat(
|
434 |
+
tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
|
435 |
+
0).narrow(0, 0, unpartitioned_numel).view(shape)
|
436 |
+
offset += partitioned_numel
|
437 |
+
|
438 |
+
offset *= world_size
|
439 |
+
|
440 |
+
# Sanity check
|
441 |
+
if offset != avail_numel:
|
442 |
+
raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
443 |
+
|
444 |
+
print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
|
445 |
+
|
446 |
+
|
447 |
+
def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
|
448 |
+
state_dict = OrderedDict()
|
449 |
+
|
450 |
+
# buffers
|
451 |
+
buffers = zero_model_states[0].buffers
|
452 |
+
state_dict.update(buffers)
|
453 |
+
if debug:
|
454 |
+
print(f"added {len(buffers)} buffers")
|
455 |
+
|
456 |
+
_zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
|
457 |
+
|
458 |
+
_zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
|
459 |
+
|
460 |
+
# recover shared parameters
|
461 |
+
for pair in zero_model_states[0].shared_params:
|
462 |
+
if pair[1] in state_dict:
|
463 |
+
state_dict[pair[0]] = state_dict[pair[1]]
|
464 |
+
|
465 |
+
return state_dict
|
466 |
+
|
467 |
+
|
468 |
+
def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
|
469 |
+
"""
|
470 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
|
471 |
+
``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
|
472 |
+
via a model hub.
|
473 |
+
|
474 |
+
Args:
|
475 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder
|
476 |
+
- ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
|
477 |
+
|
478 |
+
Returns:
|
479 |
+
- pytorch ``state_dict``
|
480 |
+
|
481 |
+
Note: this approach may not work if your application doesn't have sufficient free CPU memory and
|
482 |
+
you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
|
483 |
+
the checkpoint.
|
484 |
+
|
485 |
+
A typical usage might be ::
|
486 |
+
|
487 |
+
from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
|
488 |
+
# do the training and checkpoint saving
|
489 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
|
490 |
+
model = model.cpu() # move to cpu
|
491 |
+
model.load_state_dict(state_dict)
|
492 |
+
# submit to model hub or save the model to share with others
|
493 |
+
|
494 |
+
In this example the ``model`` will no longer be usable in the deepspeed context of the same
|
495 |
+
application. i.e. you will need to re-initialize the deepspeed engine, since
|
496 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
497 |
+
|
498 |
+
If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
|
499 |
+
|
500 |
+
"""
|
501 |
+
if tag is None:
|
502 |
+
latest_path = os.path.join(checkpoint_dir, 'latest')
|
503 |
+
if os.path.isfile(latest_path):
|
504 |
+
with open(latest_path, 'r') as fd:
|
505 |
+
tag = fd.read().strip()
|
506 |
+
else:
|
507 |
+
raise ValueError(f"Unable to find 'latest' file at {latest_path}")
|
508 |
+
|
509 |
+
ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
|
510 |
+
|
511 |
+
if not os.path.isdir(ds_checkpoint_dir):
|
512 |
+
raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
|
513 |
+
|
514 |
+
return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
|
515 |
+
|
516 |
+
|
517 |
+
def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
|
518 |
+
"""
|
519 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
|
520 |
+
loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
|
521 |
+
|
522 |
+
Args:
|
523 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
524 |
+
- ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
|
525 |
+
- ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
|
526 |
+
"""
|
527 |
+
|
528 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
|
529 |
+
print(f"Saving fp32 state dict to {output_file}")
|
530 |
+
torch.save(state_dict, output_file)
|
531 |
+
|
532 |
+
|
533 |
+
def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
|
534 |
+
"""
|
535 |
+
1. Put the provided model to cpu
|
536 |
+
2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
|
537 |
+
3. Load it into the provided model
|
538 |
+
|
539 |
+
Args:
|
540 |
+
- ``model``: the model object to update
|
541 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
542 |
+
- ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
|
543 |
+
|
544 |
+
Returns:
|
545 |
+
- ``model`: modified model
|
546 |
+
|
547 |
+
Make sure you have plenty of CPU memory available before you call this function. If you don't
|
548 |
+
have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
|
549 |
+
conveniently placed for you in the checkpoint folder.
|
550 |
+
|
551 |
+
A typical usage might be ::
|
552 |
+
|
553 |
+
from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
|
554 |
+
model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
|
555 |
+
# submit to model hub or save the model to share with others
|
556 |
+
|
557 |
+
Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
|
558 |
+
of the same application. i.e. you will need to re-initialize the deepspeed engine, since
|
559 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
560 |
+
|
561 |
+
"""
|
562 |
+
logger.info(f"Extracting fp32 weights")
|
563 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
|
564 |
+
|
565 |
+
logger.info(f"Overwriting model with fp32 weights")
|
566 |
+
model = model.cpu()
|
567 |
+
model.load_state_dict(state_dict, strict=False)
|
568 |
+
|
569 |
+
return model
|
570 |
+
|
571 |
+
|
572 |
+
if __name__ == "__main__":
|
573 |
+
|
574 |
+
parser = argparse.ArgumentParser()
|
575 |
+
parser.add_argument("checkpoint_dir",
|
576 |
+
type=str,
|
577 |
+
help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
|
578 |
+
parser.add_argument(
|
579 |
+
"output_file",
|
580 |
+
type=str,
|
581 |
+
help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
|
582 |
+
parser.add_argument("-t",
|
583 |
+
"--tag",
|
584 |
+
type=str,
|
585 |
+
default=None,
|
586 |
+
help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
|
587 |
+
parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
|
588 |
+
args = parser.parse_args()
|
589 |
+
|
590 |
+
debug = args.debug
|
591 |
+
|
592 |
+
convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)
|
checkpoint-168/README.md
ADDED
@@ -0,0 +1,202 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
library_name: peft
|
3 |
+
base_model: ai21labs/Jamba-v0.1
|
4 |
+
---
|
5 |
+
|
6 |
+
# Model Card for Model ID
|
7 |
+
|
8 |
+
<!-- Provide a quick summary of what the model is/does. -->
|
9 |
+
|
10 |
+
|
11 |
+
|
12 |
+
## Model Details
|
13 |
+
|
14 |
+
### Model Description
|
15 |
+
|
16 |
+
<!-- Provide a longer summary of what this model is. -->
|
17 |
+
|
18 |
+
|
19 |
+
|
20 |
+
- **Developed by:** [More Information Needed]
|
21 |
+
- **Funded by [optional]:** [More Information Needed]
|
22 |
+
- **Shared by [optional]:** [More Information Needed]
|
23 |
+
- **Model type:** [More Information Needed]
|
24 |
+
- **Language(s) (NLP):** [More Information Needed]
|
25 |
+
- **License:** [More Information Needed]
|
26 |
+
- **Finetuned from model [optional]:** [More Information Needed]
|
27 |
+
|
28 |
+
### Model Sources [optional]
|
29 |
+
|
30 |
+
<!-- Provide the basic links for the model. -->
|
31 |
+
|
32 |
+
- **Repository:** [More Information Needed]
|
33 |
+
- **Paper [optional]:** [More Information Needed]
|
34 |
+
- **Demo [optional]:** [More Information Needed]
|
35 |
+
|
36 |
+
## Uses
|
37 |
+
|
38 |
+
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
|
39 |
+
|
40 |
+
### Direct Use
|
41 |
+
|
42 |
+
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
|
43 |
+
|
44 |
+
[More Information Needed]
|
45 |
+
|
46 |
+
### Downstream Use [optional]
|
47 |
+
|
48 |
+
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
|
49 |
+
|
50 |
+
[More Information Needed]
|
51 |
+
|
52 |
+
### Out-of-Scope Use
|
53 |
+
|
54 |
+
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
|
55 |
+
|
56 |
+
[More Information Needed]
|
57 |
+
|
58 |
+
## Bias, Risks, and Limitations
|
59 |
+
|
60 |
+
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
|
61 |
+
|
62 |
+
[More Information Needed]
|
63 |
+
|
64 |
+
### Recommendations
|
65 |
+
|
66 |
+
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
|
67 |
+
|
68 |
+
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
|
69 |
+
|
70 |
+
## How to Get Started with the Model
|
71 |
+
|
72 |
+
Use the code below to get started with the model.
|
73 |
+
|
74 |
+
[More Information Needed]
|
75 |
+
|
76 |
+
## Training Details
|
77 |
+
|
78 |
+
### Training Data
|
79 |
+
|
80 |
+
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
|
81 |
+
|
82 |
+
[More Information Needed]
|
83 |
+
|
84 |
+
### Training Procedure
|
85 |
+
|
86 |
+
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
|
87 |
+
|
88 |
+
#### Preprocessing [optional]
|
89 |
+
|
90 |
+
[More Information Needed]
|
91 |
+
|
92 |
+
|
93 |
+
#### Training Hyperparameters
|
94 |
+
|
95 |
+
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
|
96 |
+
|
97 |
+
#### Speeds, Sizes, Times [optional]
|
98 |
+
|
99 |
+
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
|
100 |
+
|
101 |
+
[More Information Needed]
|
102 |
+
|
103 |
+
## Evaluation
|
104 |
+
|
105 |
+
<!-- This section describes the evaluation protocols and provides the results. -->
|
106 |
+
|
107 |
+
### Testing Data, Factors & Metrics
|
108 |
+
|
109 |
+
#### Testing Data
|
110 |
+
|
111 |
+
<!-- This should link to a Dataset Card if possible. -->
|
112 |
+
|
113 |
+
[More Information Needed]
|
114 |
+
|
115 |
+
#### Factors
|
116 |
+
|
117 |
+
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
|
118 |
+
|
119 |
+
[More Information Needed]
|
120 |
+
|
121 |
+
#### Metrics
|
122 |
+
|
123 |
+
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
|
124 |
+
|
125 |
+
[More Information Needed]
|
126 |
+
|
127 |
+
### Results
|
128 |
+
|
129 |
+
[More Information Needed]
|
130 |
+
|
131 |
+
#### Summary
|
132 |
+
|
133 |
+
|
134 |
+
|
135 |
+
## Model Examination [optional]
|
136 |
+
|
137 |
+
<!-- Relevant interpretability work for the model goes here -->
|
138 |
+
|
139 |
+
[More Information Needed]
|
140 |
+
|
141 |
+
## Environmental Impact
|
142 |
+
|
143 |
+
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
|
144 |
+
|
145 |
+
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
|
146 |
+
|
147 |
+
- **Hardware Type:** [More Information Needed]
|
148 |
+
- **Hours used:** [More Information Needed]
|
149 |
+
- **Cloud Provider:** [More Information Needed]
|
150 |
+
- **Compute Region:** [More Information Needed]
|
151 |
+
- **Carbon Emitted:** [More Information Needed]
|
152 |
+
|
153 |
+
## Technical Specifications [optional]
|
154 |
+
|
155 |
+
### Model Architecture and Objective
|
156 |
+
|
157 |
+
[More Information Needed]
|
158 |
+
|
159 |
+
### Compute Infrastructure
|
160 |
+
|
161 |
+
[More Information Needed]
|
162 |
+
|
163 |
+
#### Hardware
|
164 |
+
|
165 |
+
[More Information Needed]
|
166 |
+
|
167 |
+
#### Software
|
168 |
+
|
169 |
+
[More Information Needed]
|
170 |
+
|
171 |
+
## Citation [optional]
|
172 |
+
|
173 |
+
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
174 |
+
|
175 |
+
**BibTeX:**
|
176 |
+
|
177 |
+
[More Information Needed]
|
178 |
+
|
179 |
+
**APA:**
|
180 |
+
|
181 |
+
[More Information Needed]
|
182 |
+
|
183 |
+
## Glossary [optional]
|
184 |
+
|
185 |
+
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
|
186 |
+
|
187 |
+
[More Information Needed]
|
188 |
+
|
189 |
+
## More Information [optional]
|
190 |
+
|
191 |
+
[More Information Needed]
|
192 |
+
|
193 |
+
## Model Card Authors [optional]
|
194 |
+
|
195 |
+
[More Information Needed]
|
196 |
+
|
197 |
+
## Model Card Contact
|
198 |
+
|
199 |
+
[More Information Needed]
|
200 |
+
### Framework versions
|
201 |
+
|
202 |
+
- PEFT 0.10.0
|
checkpoint-168/adapter_config.json
ADDED
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"alpha_pattern": {},
|
3 |
+
"auto_mapping": null,
|
4 |
+
"base_model_name_or_path": "ai21labs/Jamba-v0.1",
|
5 |
+
"bias": "none",
|
6 |
+
"fan_in_fan_out": null,
|
7 |
+
"inference_mode": true,
|
8 |
+
"init_lora_weights": true,
|
9 |
+
"layer_replication": null,
|
10 |
+
"layers_pattern": null,
|
11 |
+
"layers_to_transform": null,
|
12 |
+
"loftq_config": {},
|
13 |
+
"lora_alpha": 32,
|
14 |
+
"lora_dropout": 0.05,
|
15 |
+
"megatron_config": null,
|
16 |
+
"megatron_core": "megatron.core",
|
17 |
+
"modules_to_save": null,
|
18 |
+
"peft_type": "LORA",
|
19 |
+
"r": 16,
|
20 |
+
"rank_pattern": {},
|
21 |
+
"revision": null,
|
22 |
+
"target_modules": [
|
23 |
+
"x_proj",
|
24 |
+
"o_proj",
|
25 |
+
"up_proj",
|
26 |
+
"v_proj",
|
27 |
+
"down_proj",
|
28 |
+
"dt_proj",
|
29 |
+
"k_proj",
|
30 |
+
"q_proj",
|
31 |
+
"in_proj",
|
32 |
+
"gate_proj",
|
33 |
+
"out_proj",
|
34 |
+
"router"
|
35 |
+
],
|
36 |
+
"task_type": "CAUSAL_LM",
|
37 |
+
"use_dora": false,
|
38 |
+
"use_rslora": false
|
39 |
+
}
|
checkpoint-168/adapter_model.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e965e9e667252ca3c021de51e081f6a3831a14d282903458d11e44f878a73597
|
3 |
+
size 531611600
|
checkpoint-168/global_step168/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b66d8b3841b807a595a50e795a52d67de68eb94015081f06580ca93cd533b645
|
3 |
+
size 797643792
|
checkpoint-168/global_step168/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b94caa551e67c01642f6b55c16bf11b71089c2f0fbd09d4215bd26e34ed7a530
|
3 |
+
size 797644432
|
checkpoint-168/global_step168/mp_rank_00_model_states.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:8130cd37e9434fdc6e150b642bbebca3eeebac77da37779b2183486786db25a5
|
3 |
+
size 1345660121
|
checkpoint-168/latest
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
global_step168
|
checkpoint-168/rng_state_0.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:753bf22647e1e0f236cfa8df696cc134d9e1c0684712d00bf72503af01afa161
|
3 |
+
size 14512
|
checkpoint-168/rng_state_1.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:8c15347a58866f49cdc458b966fb754bbfbe4244548a2a26de0617bb3969cd63
|
3 |
+
size 14512
|
checkpoint-168/scheduler.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:35ea2e1439244ec6e596a083a4657a3e879cc60c5a80ec3ad3f705a12a8fa66a
|
3 |
+
size 1064
|
checkpoint-168/special_tokens_map.json
ADDED
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"bos_token": {
|
3 |
+
"content": "<|startoftext|>",
|
4 |
+
"lstrip": false,
|
5 |
+
"normalized": false,
|
6 |
+
"rstrip": false,
|
7 |
+
"single_word": false
|
8 |
+
},
|
9 |
+
"eos_token": {
|
10 |
+
"content": "<|endoftext|>",
|
11 |
+
"lstrip": false,
|
12 |
+
"normalized": false,
|
13 |
+
"rstrip": false,
|
14 |
+
"single_word": false
|
15 |
+
},
|
16 |
+
"pad_token": {
|
17 |
+
"content": "<|pad|>",
|
18 |
+
"lstrip": false,
|
19 |
+
"normalized": false,
|
20 |
+
"rstrip": false,
|
21 |
+
"single_word": false
|
22 |
+
},
|
23 |
+
"unk_token": {
|
24 |
+
"content": "<|unk|>",
|
25 |
+
"lstrip": false,
|
26 |
+
"normalized": false,
|
27 |
+
"rstrip": false,
|
28 |
+
"single_word": false
|
29 |
+
}
|
30 |
+
}
|
checkpoint-168/tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
checkpoint-168/tokenizer.model
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:02fd6530b8ede0eedd8e509fcab32da7b1dd04c8119f8498c787100f13112713
|
3 |
+
size 1124742
|
checkpoint-168/tokenizer_config.json
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"add_bos_token": true,
|
3 |
+
"add_eos_token": false,
|
4 |
+
"added_tokens_decoder": {
|
5 |
+
"0": {
|
6 |
+
"content": "<|pad|>",
|
7 |
+
"lstrip": false,
|
8 |
+
"normalized": false,
|
9 |
+
"rstrip": false,
|
10 |
+
"single_word": false,
|
11 |
+
"special": true
|
12 |
+
},
|
13 |
+
"1": {
|
14 |
+
"content": "<|startoftext|>",
|
15 |
+
"lstrip": false,
|
16 |
+
"normalized": false,
|
17 |
+
"rstrip": false,
|
18 |
+
"single_word": false,
|
19 |
+
"special": true
|
20 |
+
},
|
21 |
+
"2": {
|
22 |
+
"content": "<|endoftext|>",
|
23 |
+
"lstrip": false,
|
24 |
+
"normalized": false,
|
25 |
+
"rstrip": false,
|
26 |
+
"single_word": false,
|
27 |
+
"special": true
|
28 |
+
},
|
29 |
+
"3": {
|
30 |
+
"content": "<|unk|>",
|
31 |
+
"lstrip": false,
|
32 |
+
"normalized": false,
|
33 |
+
"rstrip": false,
|
34 |
+
"single_word": false,
|
35 |
+
"special": true
|
36 |
+
}
|
37 |
+
},
|
38 |
+
"bos_token": "<|startoftext|>",
|
39 |
+
"chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
|
40 |
+
"clean_up_tokenization_spaces": false,
|
41 |
+
"eos_token": "<|endoftext|>",
|
42 |
+
"model_max_length": 1000000000000000019884624838656,
|
43 |
+
"pad_token": "<|pad|>",
|
44 |
+
"spaces_between_special_tokens": false,
|
45 |
+
"tokenizer_class": "LlamaTokenizer",
|
46 |
+
"unk_token": "<|unk|>",
|
47 |
+
"use_default_system_prompt": false
|
48 |
+
}
|
checkpoint-168/trainer_state.json
ADDED
@@ -0,0 +1,1237 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"best_metric": null,
|
3 |
+
"best_model_checkpoint": null,
|
4 |
+
"epoch": 0.9985141158989599,
|
5 |
+
"eval_steps": 42,
|
6 |
+
"global_step": 168,
|
7 |
+
"is_hyper_param_search": false,
|
8 |
+
"is_local_process_zero": true,
|
9 |
+
"is_world_process_zero": true,
|
10 |
+
"log_history": [
|
11 |
+
{
|
12 |
+
"epoch": 0.01,
|
13 |
+
"grad_norm": 0.30475762367257353,
|
14 |
+
"learning_rate": 2e-05,
|
15 |
+
"loss": 0.6274,
|
16 |
+
"step": 1
|
17 |
+
},
|
18 |
+
{
|
19 |
+
"epoch": 0.01,
|
20 |
+
"eval_loss": 1.0297596454620361,
|
21 |
+
"eval_runtime": 150.6014,
|
22 |
+
"eval_samples_per_second": 1.653,
|
23 |
+
"eval_steps_per_second": 0.83,
|
24 |
+
"step": 1
|
25 |
+
},
|
26 |
+
{
|
27 |
+
"epoch": 0.01,
|
28 |
+
"grad_norm": 0.2798137150913395,
|
29 |
+
"learning_rate": 4e-05,
|
30 |
+
"loss": 0.6362,
|
31 |
+
"step": 2
|
32 |
+
},
|
33 |
+
{
|
34 |
+
"epoch": 0.02,
|
35 |
+
"grad_norm": 0.3178684451319545,
|
36 |
+
"learning_rate": 6e-05,
|
37 |
+
"loss": 0.6299,
|
38 |
+
"step": 3
|
39 |
+
},
|
40 |
+
{
|
41 |
+
"epoch": 0.02,
|
42 |
+
"grad_norm": 0.33651284916847835,
|
43 |
+
"learning_rate": 8e-05,
|
44 |
+
"loss": 0.6391,
|
45 |
+
"step": 4
|
46 |
+
},
|
47 |
+
{
|
48 |
+
"epoch": 0.03,
|
49 |
+
"grad_norm": 0.23984915388648712,
|
50 |
+
"learning_rate": 0.0001,
|
51 |
+
"loss": 0.6071,
|
52 |
+
"step": 5
|
53 |
+
},
|
54 |
+
{
|
55 |
+
"epoch": 0.04,
|
56 |
+
"grad_norm": 0.20514040410017348,
|
57 |
+
"learning_rate": 0.00012,
|
58 |
+
"loss": 0.5996,
|
59 |
+
"step": 6
|
60 |
+
},
|
61 |
+
{
|
62 |
+
"epoch": 0.04,
|
63 |
+
"grad_norm": 0.1950018128286362,
|
64 |
+
"learning_rate": 0.00014,
|
65 |
+
"loss": 0.6298,
|
66 |
+
"step": 7
|
67 |
+
},
|
68 |
+
{
|
69 |
+
"epoch": 0.05,
|
70 |
+
"grad_norm": 0.14246019947393238,
|
71 |
+
"learning_rate": 0.00016,
|
72 |
+
"loss": 0.5108,
|
73 |
+
"step": 8
|
74 |
+
},
|
75 |
+
{
|
76 |
+
"epoch": 0.05,
|
77 |
+
"grad_norm": 0.15792014279750227,
|
78 |
+
"learning_rate": 0.00018,
|
79 |
+
"loss": 0.5529,
|
80 |
+
"step": 9
|
81 |
+
},
|
82 |
+
{
|
83 |
+
"epoch": 0.06,
|
84 |
+
"grad_norm": 0.1517889177511264,
|
85 |
+
"learning_rate": 0.0002,
|
86 |
+
"loss": 0.5433,
|
87 |
+
"step": 10
|
88 |
+
},
|
89 |
+
{
|
90 |
+
"epoch": 0.07,
|
91 |
+
"grad_norm": 0.1372121219437277,
|
92 |
+
"learning_rate": 0.00019998023297700658,
|
93 |
+
"loss": 0.5856,
|
94 |
+
"step": 11
|
95 |
+
},
|
96 |
+
{
|
97 |
+
"epoch": 0.07,
|
98 |
+
"grad_norm": 0.16740807394942855,
|
99 |
+
"learning_rate": 0.00019992093972273018,
|
100 |
+
"loss": 0.5546,
|
101 |
+
"step": 12
|
102 |
+
},
|
103 |
+
{
|
104 |
+
"epoch": 0.08,
|
105 |
+
"grad_norm": 0.13512320693394078,
|
106 |
+
"learning_rate": 0.00019982214367819328,
|
107 |
+
"loss": 0.6193,
|
108 |
+
"step": 13
|
109 |
+
},
|
110 |
+
{
|
111 |
+
"epoch": 0.08,
|
112 |
+
"grad_norm": 0.16169796294070152,
|
113 |
+
"learning_rate": 0.0001996838839014696,
|
114 |
+
"loss": 0.5495,
|
115 |
+
"step": 14
|
116 |
+
},
|
117 |
+
{
|
118 |
+
"epoch": 0.09,
|
119 |
+
"grad_norm": 0.16796913812281988,
|
120 |
+
"learning_rate": 0.00019950621505224273,
|
121 |
+
"loss": 0.5035,
|
122 |
+
"step": 15
|
123 |
+
},
|
124 |
+
{
|
125 |
+
"epoch": 0.1,
|
126 |
+
"grad_norm": 0.1800514764162192,
|
127 |
+
"learning_rate": 0.00019928920737019733,
|
128 |
+
"loss": 0.5083,
|
129 |
+
"step": 16
|
130 |
+
},
|
131 |
+
{
|
132 |
+
"epoch": 0.1,
|
133 |
+
"grad_norm": 0.170432124866908,
|
134 |
+
"learning_rate": 0.0001990329466472502,
|
135 |
+
"loss": 0.632,
|
136 |
+
"step": 17
|
137 |
+
},
|
138 |
+
{
|
139 |
+
"epoch": 0.11,
|
140 |
+
"grad_norm": 0.19129325489749488,
|
141 |
+
"learning_rate": 0.00019873753419363336,
|
142 |
+
"loss": 0.4813,
|
143 |
+
"step": 18
|
144 |
+
},
|
145 |
+
{
|
146 |
+
"epoch": 0.11,
|
147 |
+
"grad_norm": 0.1459357988760762,
|
148 |
+
"learning_rate": 0.00019840308679784207,
|
149 |
+
"loss": 0.4973,
|
150 |
+
"step": 19
|
151 |
+
},
|
152 |
+
{
|
153 |
+
"epoch": 0.12,
|
154 |
+
"grad_norm": 0.192594730984382,
|
155 |
+
"learning_rate": 0.00019802973668046363,
|
156 |
+
"loss": 0.5291,
|
157 |
+
"step": 20
|
158 |
+
},
|
159 |
+
{
|
160 |
+
"epoch": 0.12,
|
161 |
+
"grad_norm": 0.859025467969139,
|
162 |
+
"learning_rate": 0.0001976176314419051,
|
163 |
+
"loss": 0.5296,
|
164 |
+
"step": 21
|
165 |
+
},
|
166 |
+
{
|
167 |
+
"epoch": 0.13,
|
168 |
+
"grad_norm": 0.13366297885670222,
|
169 |
+
"learning_rate": 0.000197166934004041,
|
170 |
+
"loss": 0.4819,
|
171 |
+
"step": 22
|
172 |
+
},
|
173 |
+
{
|
174 |
+
"epoch": 0.14,
|
175 |
+
"grad_norm": 0.15698714419747645,
|
176 |
+
"learning_rate": 0.00019667782254580374,
|
177 |
+
"loss": 0.5409,
|
178 |
+
"step": 23
|
179 |
+
},
|
180 |
+
{
|
181 |
+
"epoch": 0.14,
|
182 |
+
"grad_norm": 0.10995943735837355,
|
183 |
+
"learning_rate": 0.00019615049043274205,
|
184 |
+
"loss": 0.5108,
|
185 |
+
"step": 24
|
186 |
+
},
|
187 |
+
{
|
188 |
+
"epoch": 0.15,
|
189 |
+
"grad_norm": 0.10796742192788925,
|
190 |
+
"learning_rate": 0.00019558514614057609,
|
191 |
+
"loss": 0.5215,
|
192 |
+
"step": 25
|
193 |
+
},
|
194 |
+
{
|
195 |
+
"epoch": 0.15,
|
196 |
+
"grad_norm": 0.11641740089490231,
|
197 |
+
"learning_rate": 0.00019498201317277828,
|
198 |
+
"loss": 0.5012,
|
199 |
+
"step": 26
|
200 |
+
},
|
201 |
+
{
|
202 |
+
"epoch": 0.16,
|
203 |
+
"grad_norm": 0.1120175962893241,
|
204 |
+
"learning_rate": 0.00019434132997221345,
|
205 |
+
"loss": 0.474,
|
206 |
+
"step": 27
|
207 |
+
},
|
208 |
+
{
|
209 |
+
"epoch": 0.17,
|
210 |
+
"grad_norm": 0.1218171278782483,
|
211 |
+
"learning_rate": 0.0001936633498268728,
|
212 |
+
"loss": 0.5216,
|
213 |
+
"step": 28
|
214 |
+
},
|
215 |
+
{
|
216 |
+
"epoch": 0.17,
|
217 |
+
"grad_norm": 0.11718521115928844,
|
218 |
+
"learning_rate": 0.0001929483407697387,
|
219 |
+
"loss": 0.4856,
|
220 |
+
"step": 29
|
221 |
+
},
|
222 |
+
{
|
223 |
+
"epoch": 0.18,
|
224 |
+
"grad_norm": 0.12611471038571026,
|
225 |
+
"learning_rate": 0.00019219658547282067,
|
226 |
+
"loss": 0.4823,
|
227 |
+
"step": 30
|
228 |
+
},
|
229 |
+
{
|
230 |
+
"epoch": 0.18,
|
231 |
+
"grad_norm": 0.11106871615269753,
|
232 |
+
"learning_rate": 0.00019140838113540346,
|
233 |
+
"loss": 0.4869,
|
234 |
+
"step": 31
|
235 |
+
},
|
236 |
+
{
|
237 |
+
"epoch": 0.19,
|
238 |
+
"grad_norm": 0.1416503230360699,
|
239 |
+
"learning_rate": 0.00019058403936655233,
|
240 |
+
"loss": 0.5341,
|
241 |
+
"step": 32
|
242 |
+
},
|
243 |
+
{
|
244 |
+
"epoch": 0.2,
|
245 |
+
"grad_norm": 0.10761396399791698,
|
246 |
+
"learning_rate": 0.00018972388606192125,
|
247 |
+
"loss": 0.4304,
|
248 |
+
"step": 33
|
249 |
+
},
|
250 |
+
{
|
251 |
+
"epoch": 0.2,
|
252 |
+
"grad_norm": 0.10975376180434356,
|
253 |
+
"learning_rate": 0.0001888282612749132,
|
254 |
+
"loss": 0.4646,
|
255 |
+
"step": 34
|
256 |
+
},
|
257 |
+
{
|
258 |
+
"epoch": 0.21,
|
259 |
+
"grad_norm": 0.12848879670359908,
|
260 |
+
"learning_rate": 0.00018789751908224338,
|
261 |
+
"loss": 0.4972,
|
262 |
+
"step": 35
|
263 |
+
},
|
264 |
+
{
|
265 |
+
"epoch": 0.21,
|
266 |
+
"grad_norm": 0.11904721819683833,
|
267 |
+
"learning_rate": 0.00018693202744395827,
|
268 |
+
"loss": 0.505,
|
269 |
+
"step": 36
|
270 |
+
},
|
271 |
+
{
|
272 |
+
"epoch": 0.22,
|
273 |
+
"grad_norm": 0.12249852034224981,
|
274 |
+
"learning_rate": 0.00018593216805796612,
|
275 |
+
"loss": 0.5396,
|
276 |
+
"step": 37
|
277 |
+
},
|
278 |
+
{
|
279 |
+
"epoch": 0.23,
|
280 |
+
"grad_norm": 0.12453395046646995,
|
281 |
+
"learning_rate": 0.00018489833620913642,
|
282 |
+
"loss": 0.4917,
|
283 |
+
"step": 38
|
284 |
+
},
|
285 |
+
{
|
286 |
+
"epoch": 0.23,
|
287 |
+
"grad_norm": 0.12585770374422164,
|
288 |
+
"learning_rate": 0.00018383094061302766,
|
289 |
+
"loss": 0.5079,
|
290 |
+
"step": 39
|
291 |
+
},
|
292 |
+
{
|
293 |
+
"epoch": 0.24,
|
294 |
+
"grad_norm": 0.11095271476322731,
|
295 |
+
"learning_rate": 0.00018273040325430574,
|
296 |
+
"loss": 0.4812,
|
297 |
+
"step": 40
|
298 |
+
},
|
299 |
+
{
|
300 |
+
"epoch": 0.24,
|
301 |
+
"grad_norm": 0.12968115101635422,
|
302 |
+
"learning_rate": 0.00018159715921991612,
|
303 |
+
"loss": 0.5106,
|
304 |
+
"step": 41
|
305 |
+
},
|
306 |
+
{
|
307 |
+
"epoch": 0.25,
|
308 |
+
"grad_norm": 0.10933018515590627,
|
309 |
+
"learning_rate": 0.00018043165652707649,
|
310 |
+
"loss": 0.4403,
|
311 |
+
"step": 42
|
312 |
+
},
|
313 |
+
{
|
314 |
+
"epoch": 0.25,
|
315 |
+
"eval_loss": 0.9767947196960449,
|
316 |
+
"eval_runtime": 152.0343,
|
317 |
+
"eval_samples_per_second": 1.638,
|
318 |
+
"eval_steps_per_second": 0.822,
|
319 |
+
"step": 42
|
320 |
+
},
|
321 |
+
{
|
322 |
+
"epoch": 0.26,
|
323 |
+
"grad_norm": 0.11525719626792096,
|
324 |
+
"learning_rate": 0.00017923435594615744,
|
325 |
+
"loss": 0.482,
|
326 |
+
"step": 43
|
327 |
+
},
|
328 |
+
{
|
329 |
+
"epoch": 0.26,
|
330 |
+
"grad_norm": 0.12962154411778218,
|
331 |
+
"learning_rate": 0.00017800573081852122,
|
332 |
+
"loss": 0.5452,
|
333 |
+
"step": 44
|
334 |
+
},
|
335 |
+
{
|
336 |
+
"epoch": 0.27,
|
337 |
+
"grad_norm": 0.12555700120045588,
|
338 |
+
"learning_rate": 0.0001767462668693908,
|
339 |
+
"loss": 0.5084,
|
340 |
+
"step": 45
|
341 |
+
},
|
342 |
+
{
|
343 |
+
"epoch": 0.27,
|
344 |
+
"grad_norm": 0.11427565378293324,
|
345 |
+
"learning_rate": 0.00017545646201582303,
|
346 |
+
"loss": 0.5191,
|
347 |
+
"step": 46
|
348 |
+
},
|
349 |
+
{
|
350 |
+
"epoch": 0.28,
|
351 |
+
"grad_norm": 0.10974901402857151,
|
352 |
+
"learning_rate": 0.00017413682616986185,
|
353 |
+
"loss": 0.4703,
|
354 |
+
"step": 47
|
355 |
+
},
|
356 |
+
{
|
357 |
+
"epoch": 0.29,
|
358 |
+
"grad_norm": 0.11781465084480325,
|
359 |
+
"learning_rate": 0.00017278788103694943,
|
360 |
+
"loss": 0.4548,
|
361 |
+
"step": 48
|
362 |
+
},
|
363 |
+
{
|
364 |
+
"epoch": 0.29,
|
365 |
+
"grad_norm": 0.10781807228559999,
|
366 |
+
"learning_rate": 0.000171410159909675,
|
367 |
+
"loss": 0.476,
|
368 |
+
"step": 49
|
369 |
+
},
|
370 |
+
{
|
371 |
+
"epoch": 0.3,
|
372 |
+
"grad_norm": 0.12502639462035098,
|
373 |
+
"learning_rate": 0.00017000420745694254,
|
374 |
+
"loss": 0.5084,
|
375 |
+
"step": 50
|
376 |
+
},
|
377 |
+
{
|
378 |
+
"epoch": 0.3,
|
379 |
+
"grad_norm": 0.10718920826593327,
|
380 |
+
"learning_rate": 0.00016857057950864132,
|
381 |
+
"loss": 0.5093,
|
382 |
+
"step": 51
|
383 |
+
},
|
384 |
+
{
|
385 |
+
"epoch": 0.31,
|
386 |
+
"grad_norm": 0.10040549880547282,
|
387 |
+
"learning_rate": 0.0001671098428359037,
|
388 |
+
"loss": 0.4644,
|
389 |
+
"step": 52
|
390 |
+
},
|
391 |
+
{
|
392 |
+
"epoch": 0.32,
|
393 |
+
"grad_norm": 0.11778478994740472,
|
394 |
+
"learning_rate": 0.00016562257492703757,
|
395 |
+
"loss": 0.4725,
|
396 |
+
"step": 53
|
397 |
+
},
|
398 |
+
{
|
399 |
+
"epoch": 0.32,
|
400 |
+
"grad_norm": 0.1008386031049932,
|
401 |
+
"learning_rate": 0.000164109363759222,
|
402 |
+
"loss": 0.5121,
|
403 |
+
"step": 54
|
404 |
+
},
|
405 |
+
{
|
406 |
+
"epoch": 0.33,
|
407 |
+
"grad_norm": 0.1170302528140235,
|
408 |
+
"learning_rate": 0.000162570807566056,
|
409 |
+
"loss": 0.4766,
|
410 |
+
"step": 55
|
411 |
+
},
|
412 |
+
{
|
413 |
+
"epoch": 0.33,
|
414 |
+
"grad_norm": 0.1104526773884303,
|
415 |
+
"learning_rate": 0.00016100751460105243,
|
416 |
+
"loss": 0.4886,
|
417 |
+
"step": 56
|
418 |
+
},
|
419 |
+
{
|
420 |
+
"epoch": 0.34,
|
421 |
+
"grad_norm": 0.10467920768691032,
|
422 |
+
"learning_rate": 0.00015942010289717105,
|
423 |
+
"loss": 0.4703,
|
424 |
+
"step": 57
|
425 |
+
},
|
426 |
+
{
|
427 |
+
"epoch": 0.34,
|
428 |
+
"grad_norm": 0.11551406829220555,
|
429 |
+
"learning_rate": 0.00015780920002248484,
|
430 |
+
"loss": 0.4837,
|
431 |
+
"step": 58
|
432 |
+
},
|
433 |
+
{
|
434 |
+
"epoch": 0.35,
|
435 |
+
"grad_norm": 0.11133818831887894,
|
436 |
+
"learning_rate": 0.0001561754428320771,
|
437 |
+
"loss": 0.5148,
|
438 |
+
"step": 59
|
439 |
+
},
|
440 |
+
{
|
441 |
+
"epoch": 0.36,
|
442 |
+
"grad_norm": 0.11281448423273216,
|
443 |
+
"learning_rate": 0.00015451947721626676,
|
444 |
+
"loss": 0.4561,
|
445 |
+
"step": 60
|
446 |
+
},
|
447 |
+
{
|
448 |
+
"epoch": 0.36,
|
449 |
+
"grad_norm": 0.13934126997471205,
|
450 |
+
"learning_rate": 0.00015284195784526195,
|
451 |
+
"loss": 0.5069,
|
452 |
+
"step": 61
|
453 |
+
},
|
454 |
+
{
|
455 |
+
"epoch": 0.37,
|
456 |
+
"grad_norm": 0.11851655387640142,
|
457 |
+
"learning_rate": 0.00015114354791034225,
|
458 |
+
"loss": 0.5094,
|
459 |
+
"step": 62
|
460 |
+
},
|
461 |
+
{
|
462 |
+
"epoch": 0.37,
|
463 |
+
"grad_norm": 0.12909148374566123,
|
464 |
+
"learning_rate": 0.0001494249188616723,
|
465 |
+
"loss": 0.581,
|
466 |
+
"step": 63
|
467 |
+
},
|
468 |
+
{
|
469 |
+
"epoch": 0.38,
|
470 |
+
"grad_norm": 0.11070161341925377,
|
471 |
+
"learning_rate": 0.00014768675014285062,
|
472 |
+
"loss": 0.4585,
|
473 |
+
"step": 64
|
474 |
+
},
|
475 |
+
{
|
476 |
+
"epoch": 0.39,
|
477 |
+
"grad_norm": 0.13308674882888374,
|
478 |
+
"learning_rate": 0.00014592972892229778,
|
479 |
+
"loss": 0.4974,
|
480 |
+
"step": 65
|
481 |
+
},
|
482 |
+
{
|
483 |
+
"epoch": 0.39,
|
484 |
+
"grad_norm": 0.12124588853708144,
|
485 |
+
"learning_rate": 0.0001441545498215912,
|
486 |
+
"loss": 0.4463,
|
487 |
+
"step": 66
|
488 |
+
},
|
489 |
+
{
|
490 |
+
"epoch": 0.4,
|
491 |
+
"grad_norm": 0.1183570515369953,
|
492 |
+
"learning_rate": 0.00014236191464085286,
|
493 |
+
"loss": 0.447,
|
494 |
+
"step": 67
|
495 |
+
},
|
496 |
+
{
|
497 |
+
"epoch": 0.4,
|
498 |
+
"grad_norm": 0.13520024884417237,
|
499 |
+
"learning_rate": 0.00014055253208129938,
|
500 |
+
"loss": 0.5309,
|
501 |
+
"step": 68
|
502 |
+
},
|
503 |
+
{
|
504 |
+
"epoch": 0.41,
|
505 |
+
"grad_norm": 0.12184981458813801,
|
506 |
+
"learning_rate": 0.00013872711746506413,
|
507 |
+
"loss": 0.4532,
|
508 |
+
"step": 69
|
509 |
+
},
|
510 |
+
{
|
511 |
+
"epoch": 0.42,
|
512 |
+
"grad_norm": 0.12449299540645078,
|
513 |
+
"learning_rate": 0.00013688639245240078,
|
514 |
+
"loss": 0.5198,
|
515 |
+
"step": 70
|
516 |
+
},
|
517 |
+
{
|
518 |
+
"epoch": 0.42,
|
519 |
+
"grad_norm": 0.1383134750490429,
|
520 |
+
"learning_rate": 0.00013503108475638244,
|
521 |
+
"loss": 0.5629,
|
522 |
+
"step": 71
|
523 |
+
},
|
524 |
+
{
|
525 |
+
"epoch": 0.43,
|
526 |
+
"grad_norm": 0.246237001656926,
|
527 |
+
"learning_rate": 0.0001331619278552068,
|
528 |
+
"loss": 0.4869,
|
529 |
+
"step": 72
|
530 |
+
},
|
531 |
+
{
|
532 |
+
"epoch": 0.43,
|
533 |
+
"grad_norm": 0.13337703940933632,
|
534 |
+
"learning_rate": 0.00013127966070222274,
|
535 |
+
"loss": 0.4792,
|
536 |
+
"step": 73
|
537 |
+
},
|
538 |
+
{
|
539 |
+
"epoch": 0.44,
|
540 |
+
"grad_norm": 0.12428922033806454,
|
541 |
+
"learning_rate": 0.00012938502743379212,
|
542 |
+
"loss": 0.4825,
|
543 |
+
"step": 74
|
544 |
+
},
|
545 |
+
{
|
546 |
+
"epoch": 0.45,
|
547 |
+
"grad_norm": 0.13290774912900208,
|
548 |
+
"learning_rate": 0.00012747877707510252,
|
549 |
+
"loss": 0.5138,
|
550 |
+
"step": 75
|
551 |
+
},
|
552 |
+
{
|
553 |
+
"epoch": 0.45,
|
554 |
+
"grad_norm": 0.11185975046756892,
|
555 |
+
"learning_rate": 0.0001255616632440475,
|
556 |
+
"loss": 0.4815,
|
557 |
+
"step": 76
|
558 |
+
},
|
559 |
+
{
|
560 |
+
"epoch": 0.46,
|
561 |
+
"grad_norm": 0.1130592868215497,
|
562 |
+
"learning_rate": 0.0001236344438532905,
|
563 |
+
"loss": 0.5046,
|
564 |
+
"step": 77
|
565 |
+
},
|
566 |
+
{
|
567 |
+
"epoch": 0.46,
|
568 |
+
"grad_norm": 0.12882943465594857,
|
569 |
+
"learning_rate": 0.0001216978808106318,
|
570 |
+
"loss": 0.5091,
|
571 |
+
"step": 78
|
572 |
+
},
|
573 |
+
{
|
574 |
+
"epoch": 0.47,
|
575 |
+
"grad_norm": 0.14837896297082676,
|
576 |
+
"learning_rate": 0.00011975273971779528,
|
577 |
+
"loss": 0.5158,
|
578 |
+
"step": 79
|
579 |
+
},
|
580 |
+
{
|
581 |
+
"epoch": 0.48,
|
582 |
+
"grad_norm": 0.1265223309856292,
|
583 |
+
"learning_rate": 0.00011779978956775506,
|
584 |
+
"loss": 0.5068,
|
585 |
+
"step": 80
|
586 |
+
},
|
587 |
+
{
|
588 |
+
"epoch": 0.48,
|
589 |
+
"grad_norm": 0.14042502330520407,
|
590 |
+
"learning_rate": 0.0001158398024407215,
|
591 |
+
"loss": 0.5061,
|
592 |
+
"step": 81
|
593 |
+
},
|
594 |
+
{
|
595 |
+
"epoch": 0.49,
|
596 |
+
"grad_norm": 0.1261526695491767,
|
597 |
+
"learning_rate": 0.00011387355319890685,
|
598 |
+
"loss": 0.4691,
|
599 |
+
"step": 82
|
600 |
+
},
|
601 |
+
{
|
602 |
+
"epoch": 0.49,
|
603 |
+
"grad_norm": 0.12007305451001854,
|
604 |
+
"learning_rate": 0.00011190181918019049,
|
605 |
+
"loss": 0.4753,
|
606 |
+
"step": 83
|
607 |
+
},
|
608 |
+
{
|
609 |
+
"epoch": 0.5,
|
610 |
+
"grad_norm": 0.12809956897166885,
|
611 |
+
"learning_rate": 0.00010992537989080618,
|
612 |
+
"loss": 0.4417,
|
613 |
+
"step": 84
|
614 |
+
},
|
615 |
+
{
|
616 |
+
"epoch": 0.5,
|
617 |
+
"eval_loss": 0.9675251841545105,
|
618 |
+
"eval_runtime": 152.4793,
|
619 |
+
"eval_samples_per_second": 1.633,
|
620 |
+
"eval_steps_per_second": 0.82,
|
621 |
+
"step": 84
|
622 |
+
},
|
623 |
+
{
|
624 |
+
"epoch": 0.51,
|
625 |
+
"grad_norm": 0.11858329804793687,
|
626 |
+
"learning_rate": 0.00010794501669717145,
|
627 |
+
"loss": 0.4868,
|
628 |
+
"step": 85
|
629 |
+
},
|
630 |
+
{
|
631 |
+
"epoch": 0.51,
|
632 |
+
"grad_norm": 0.10984649953887334,
|
633 |
+
"learning_rate": 0.00010596151251698199,
|
634 |
+
"loss": 0.4598,
|
635 |
+
"step": 86
|
636 |
+
},
|
637 |
+
{
|
638 |
+
"epoch": 0.52,
|
639 |
+
"grad_norm": 0.10927203986256682,
|
640 |
+
"learning_rate": 0.0001039756515096926,
|
641 |
+
"loss": 0.4693,
|
642 |
+
"step": 87
|
643 |
+
},
|
644 |
+
{
|
645 |
+
"epoch": 0.52,
|
646 |
+
"grad_norm": 0.11205046531522328,
|
647 |
+
"learning_rate": 0.00010198821876650701,
|
648 |
+
"loss": 0.4921,
|
649 |
+
"step": 88
|
650 |
+
},
|
651 |
+
{
|
652 |
+
"epoch": 0.53,
|
653 |
+
"grad_norm": 0.13232347270009215,
|
654 |
+
"learning_rate": 0.0001,
|
655 |
+
"loss": 0.4695,
|
656 |
+
"step": 89
|
657 |
+
},
|
658 |
+
{
|
659 |
+
"epoch": 0.53,
|
660 |
+
"grad_norm": 0.12136881873560385,
|
661 |
+
"learning_rate": 9.801178123349298e-05,
|
662 |
+
"loss": 0.4859,
|
663 |
+
"step": 90
|
664 |
+
},
|
665 |
+
{
|
666 |
+
"epoch": 0.54,
|
667 |
+
"grad_norm": 0.14347476421156694,
|
668 |
+
"learning_rate": 9.602434849030745e-05,
|
669 |
+
"loss": 0.4796,
|
670 |
+
"step": 91
|
671 |
+
},
|
672 |
+
{
|
673 |
+
"epoch": 0.55,
|
674 |
+
"grad_norm": 0.13956845267055204,
|
675 |
+
"learning_rate": 9.403848748301802e-05,
|
676 |
+
"loss": 0.5339,
|
677 |
+
"step": 92
|
678 |
+
},
|
679 |
+
{
|
680 |
+
"epoch": 0.55,
|
681 |
+
"grad_norm": 0.12814010903196785,
|
682 |
+
"learning_rate": 9.205498330282856e-05,
|
683 |
+
"loss": 0.5267,
|
684 |
+
"step": 93
|
685 |
+
},
|
686 |
+
{
|
687 |
+
"epoch": 0.56,
|
688 |
+
"grad_norm": 0.12798850330908082,
|
689 |
+
"learning_rate": 9.007462010919386e-05,
|
690 |
+
"loss": 0.4604,
|
691 |
+
"step": 94
|
692 |
+
},
|
693 |
+
{
|
694 |
+
"epoch": 0.56,
|
695 |
+
"grad_norm": 0.13673366056605873,
|
696 |
+
"learning_rate": 8.809818081980953e-05,
|
697 |
+
"loss": 0.49,
|
698 |
+
"step": 95
|
699 |
+
},
|
700 |
+
{
|
701 |
+
"epoch": 0.57,
|
702 |
+
"grad_norm": 0.12607483394599764,
|
703 |
+
"learning_rate": 8.612644680109319e-05,
|
704 |
+
"loss": 0.4774,
|
705 |
+
"step": 96
|
706 |
+
},
|
707 |
+
{
|
708 |
+
"epoch": 0.58,
|
709 |
+
"grad_norm": 0.1365629261848207,
|
710 |
+
"learning_rate": 8.416019755927851e-05,
|
711 |
+
"loss": 0.4827,
|
712 |
+
"step": 97
|
713 |
+
},
|
714 |
+
{
|
715 |
+
"epoch": 0.58,
|
716 |
+
"grad_norm": 0.12122559291940836,
|
717 |
+
"learning_rate": 8.2200210432245e-05,
|
718 |
+
"loss": 0.5044,
|
719 |
+
"step": 98
|
720 |
+
},
|
721 |
+
{
|
722 |
+
"epoch": 0.59,
|
723 |
+
"grad_norm": 0.11655390642565265,
|
724 |
+
"learning_rate": 8.024726028220474e-05,
|
725 |
+
"loss": 0.503,
|
726 |
+
"step": 99
|
727 |
+
},
|
728 |
+
{
|
729 |
+
"epoch": 0.59,
|
730 |
+
"grad_norm": 0.12394574502796742,
|
731 |
+
"learning_rate": 7.83021191893682e-05,
|
732 |
+
"loss": 0.491,
|
733 |
+
"step": 100
|
734 |
+
},
|
735 |
+
{
|
736 |
+
"epoch": 0.6,
|
737 |
+
"grad_norm": 0.14922841699852962,
|
738 |
+
"learning_rate": 7.636555614670953e-05,
|
739 |
+
"loss": 0.457,
|
740 |
+
"step": 101
|
741 |
+
},
|
742 |
+
{
|
743 |
+
"epoch": 0.61,
|
744 |
+
"grad_norm": 0.11076924096187928,
|
745 |
+
"learning_rate": 7.443833675595255e-05,
|
746 |
+
"loss": 0.4603,
|
747 |
+
"step": 102
|
748 |
+
},
|
749 |
+
{
|
750 |
+
"epoch": 0.61,
|
751 |
+
"grad_norm": 0.1263594611752413,
|
752 |
+
"learning_rate": 7.252122292489747e-05,
|
753 |
+
"loss": 0.4859,
|
754 |
+
"step": 103
|
755 |
+
},
|
756 |
+
{
|
757 |
+
"epoch": 0.62,
|
758 |
+
"grad_norm": 0.11432575178505003,
|
759 |
+
"learning_rate": 7.061497256620793e-05,
|
760 |
+
"loss": 0.4627,
|
761 |
+
"step": 104
|
762 |
+
},
|
763 |
+
{
|
764 |
+
"epoch": 0.62,
|
765 |
+
"grad_norm": 0.1078119391965793,
|
766 |
+
"learning_rate": 6.87203392977773e-05,
|
767 |
+
"loss": 0.4829,
|
768 |
+
"step": 105
|
769 |
+
},
|
770 |
+
{
|
771 |
+
"epoch": 0.63,
|
772 |
+
"grad_norm": 0.12752089816514908,
|
773 |
+
"learning_rate": 6.683807214479323e-05,
|
774 |
+
"loss": 0.46,
|
775 |
+
"step": 106
|
776 |
+
},
|
777 |
+
{
|
778 |
+
"epoch": 0.64,
|
779 |
+
"grad_norm": 0.11421623043902956,
|
780 |
+
"learning_rate": 6.496891524361757e-05,
|
781 |
+
"loss": 0.4429,
|
782 |
+
"step": 107
|
783 |
+
},
|
784 |
+
{
|
785 |
+
"epoch": 0.64,
|
786 |
+
"grad_norm": 0.10432253193399477,
|
787 |
+
"learning_rate": 6.311360754759923e-05,
|
788 |
+
"loss": 0.402,
|
789 |
+
"step": 108
|
790 |
+
},
|
791 |
+
{
|
792 |
+
"epoch": 0.65,
|
793 |
+
"grad_norm": 0.12155248673662734,
|
794 |
+
"learning_rate": 6.127288253493591e-05,
|
795 |
+
"loss": 0.5126,
|
796 |
+
"step": 109
|
797 |
+
},
|
798 |
+
{
|
799 |
+
"epoch": 0.65,
|
800 |
+
"grad_norm": 0.14266947863559803,
|
801 |
+
"learning_rate": 5.9447467918700614e-05,
|
802 |
+
"loss": 0.4821,
|
803 |
+
"step": 110
|
804 |
+
},
|
805 |
+
{
|
806 |
+
"epoch": 0.66,
|
807 |
+
"grad_norm": 0.14851250761112514,
|
808 |
+
"learning_rate": 5.763808535914723e-05,
|
809 |
+
"loss": 0.4891,
|
810 |
+
"step": 111
|
811 |
+
},
|
812 |
+
{
|
813 |
+
"epoch": 0.67,
|
814 |
+
"grad_norm": 0.14264023747361737,
|
815 |
+
"learning_rate": 5.584545017840885e-05,
|
816 |
+
"loss": 0.5181,
|
817 |
+
"step": 112
|
818 |
+
},
|
819 |
+
{
|
820 |
+
"epoch": 0.67,
|
821 |
+
"grad_norm": 0.12837168363458795,
|
822 |
+
"learning_rate": 5.407027107770219e-05,
|
823 |
+
"loss": 0.5599,
|
824 |
+
"step": 113
|
825 |
+
},
|
826 |
+
{
|
827 |
+
"epoch": 0.68,
|
828 |
+
"grad_norm": 0.11874709251257598,
|
829 |
+
"learning_rate": 5.2313249857149414e-05,
|
830 |
+
"loss": 0.4536,
|
831 |
+
"step": 114
|
832 |
+
},
|
833 |
+
{
|
834 |
+
"epoch": 0.68,
|
835 |
+
"grad_norm": 0.12010754957532713,
|
836 |
+
"learning_rate": 5.0575081138327715e-05,
|
837 |
+
"loss": 0.5004,
|
838 |
+
"step": 115
|
839 |
+
},
|
840 |
+
{
|
841 |
+
"epoch": 0.69,
|
842 |
+
"grad_norm": 0.13464124440677885,
|
843 |
+
"learning_rate": 4.885645208965779e-05,
|
844 |
+
"loss": 0.4985,
|
845 |
+
"step": 116
|
846 |
+
},
|
847 |
+
{
|
848 |
+
"epoch": 0.7,
|
849 |
+
"grad_norm": 0.13701854261941088,
|
850 |
+
"learning_rate": 4.715804215473809e-05,
|
851 |
+
"loss": 0.4709,
|
852 |
+
"step": 117
|
853 |
+
},
|
854 |
+
{
|
855 |
+
"epoch": 0.7,
|
856 |
+
"grad_norm": 0.1335483738873249,
|
857 |
+
"learning_rate": 4.548052278373327e-05,
|
858 |
+
"loss": 0.4735,
|
859 |
+
"step": 118
|
860 |
+
},
|
861 |
+
{
|
862 |
+
"epoch": 0.71,
|
863 |
+
"grad_norm": 0.13603172024059101,
|
864 |
+
"learning_rate": 4.382455716792291e-05,
|
865 |
+
"loss": 0.4721,
|
866 |
+
"step": 119
|
867 |
+
},
|
868 |
+
{
|
869 |
+
"epoch": 0.71,
|
870 |
+
"grad_norm": 0.13843339239058639,
|
871 |
+
"learning_rate": 4.219079997751515e-05,
|
872 |
+
"loss": 0.4954,
|
873 |
+
"step": 120
|
874 |
+
},
|
875 |
+
{
|
876 |
+
"epoch": 0.72,
|
877 |
+
"grad_norm": 0.15011169526780793,
|
878 |
+
"learning_rate": 4.0579897102828966e-05,
|
879 |
+
"loss": 0.4648,
|
880 |
+
"step": 121
|
881 |
+
},
|
882 |
+
{
|
883 |
+
"epoch": 0.73,
|
884 |
+
"grad_norm": 0.13061595453081623,
|
885 |
+
"learning_rate": 3.899248539894757e-05,
|
886 |
+
"loss": 0.4801,
|
887 |
+
"step": 122
|
888 |
+
},
|
889 |
+
{
|
890 |
+
"epoch": 0.73,
|
891 |
+
"grad_norm": 0.14067787924603412,
|
892 |
+
"learning_rate": 3.7429192433944014e-05,
|
893 |
+
"loss": 0.4805,
|
894 |
+
"step": 123
|
895 |
+
},
|
896 |
+
{
|
897 |
+
"epoch": 0.74,
|
898 |
+
"grad_norm": 0.13420057703295998,
|
899 |
+
"learning_rate": 3.589063624077802e-05,
|
900 |
+
"loss": 0.4446,
|
901 |
+
"step": 124
|
902 |
+
},
|
903 |
+
{
|
904 |
+
"epoch": 0.74,
|
905 |
+
"grad_norm": 0.14083737654873127,
|
906 |
+
"learning_rate": 3.4377425072962465e-05,
|
907 |
+
"loss": 0.46,
|
908 |
+
"step": 125
|
909 |
+
},
|
910 |
+
{
|
911 |
+
"epoch": 0.75,
|
912 |
+
"grad_norm": 0.13231889777376862,
|
913 |
+
"learning_rate": 3.289015716409631e-05,
|
914 |
+
"loss": 0.4451,
|
915 |
+
"step": 126
|
916 |
+
},
|
917 |
+
{
|
918 |
+
"epoch": 0.75,
|
919 |
+
"eval_loss": 0.9651579260826111,
|
920 |
+
"eval_runtime": 155.5959,
|
921 |
+
"eval_samples_per_second": 1.6,
|
922 |
+
"eval_steps_per_second": 0.803,
|
923 |
+
"step": 126
|
924 |
+
},
|
925 |
+
{
|
926 |
+
"epoch": 0.75,
|
927 |
+
"grad_norm": 0.1119782513388846,
|
928 |
+
"learning_rate": 3.14294204913587e-05,
|
929 |
+
"loss": 0.4728,
|
930 |
+
"step": 127
|
931 |
+
},
|
932 |
+
{
|
933 |
+
"epoch": 0.76,
|
934 |
+
"grad_norm": 0.1258610262132018,
|
935 |
+
"learning_rate": 2.9995792543057478e-05,
|
936 |
+
"loss": 0.4793,
|
937 |
+
"step": 128
|
938 |
+
},
|
939 |
+
{
|
940 |
+
"epoch": 0.77,
|
941 |
+
"grad_norm": 0.1332956040603169,
|
942 |
+
"learning_rate": 2.8589840090325027e-05,
|
943 |
+
"loss": 0.4919,
|
944 |
+
"step": 129
|
945 |
+
},
|
946 |
+
{
|
947 |
+
"epoch": 0.77,
|
948 |
+
"grad_norm": 0.13669237836174272,
|
949 |
+
"learning_rate": 2.7212118963050592e-05,
|
950 |
+
"loss": 0.4863,
|
951 |
+
"step": 130
|
952 |
+
},
|
953 |
+
{
|
954 |
+
"epoch": 0.78,
|
955 |
+
"grad_norm": 0.13718073990096244,
|
956 |
+
"learning_rate": 2.586317383013821e-05,
|
957 |
+
"loss": 0.4652,
|
958 |
+
"step": 131
|
959 |
+
},
|
960 |
+
{
|
961 |
+
"epoch": 0.78,
|
962 |
+
"grad_norm": 0.1328715725965393,
|
963 |
+
"learning_rate": 2.4543537984176978e-05,
|
964 |
+
"loss": 0.4464,
|
965 |
+
"step": 132
|
966 |
+
},
|
967 |
+
{
|
968 |
+
"epoch": 0.79,
|
969 |
+
"grad_norm": 0.12474829320490133,
|
970 |
+
"learning_rate": 2.325373313060919e-05,
|
971 |
+
"loss": 0.501,
|
972 |
+
"step": 133
|
973 |
+
},
|
974 |
+
{
|
975 |
+
"epoch": 0.8,
|
976 |
+
"grad_norm": 0.13832785688421897,
|
977 |
+
"learning_rate": 2.19942691814788e-05,
|
978 |
+
"loss": 0.5113,
|
979 |
+
"step": 134
|
980 |
+
},
|
981 |
+
{
|
982 |
+
"epoch": 0.8,
|
983 |
+
"grad_norm": 0.13446374612957607,
|
984 |
+
"learning_rate": 2.076564405384258e-05,
|
985 |
+
"loss": 0.452,
|
986 |
+
"step": 135
|
987 |
+
},
|
988 |
+
{
|
989 |
+
"epoch": 0.81,
|
990 |
+
"grad_norm": 0.15087962227561943,
|
991 |
+
"learning_rate": 1.9568343472923524e-05,
|
992 |
+
"loss": 0.5114,
|
993 |
+
"step": 136
|
994 |
+
},
|
995 |
+
{
|
996 |
+
"epoch": 0.81,
|
997 |
+
"grad_norm": 0.13471199433579678,
|
998 |
+
"learning_rate": 1.840284078008393e-05,
|
999 |
+
"loss": 0.5074,
|
1000 |
+
"step": 137
|
1001 |
+
},
|
1002 |
+
{
|
1003 |
+
"epoch": 0.82,
|
1004 |
+
"grad_norm": 0.14471602871304384,
|
1005 |
+
"learning_rate": 1.7269596745694295e-05,
|
1006 |
+
"loss": 0.4812,
|
1007 |
+
"step": 138
|
1008 |
+
},
|
1009 |
+
{
|
1010 |
+
"epoch": 0.83,
|
1011 |
+
"grad_norm": 0.12791095480012135,
|
1012 |
+
"learning_rate": 1.616905938697234e-05,
|
1013 |
+
"loss": 0.439,
|
1014 |
+
"step": 139
|
1015 |
+
},
|
1016 |
+
{
|
1017 |
+
"epoch": 0.83,
|
1018 |
+
"grad_norm": 0.13395748535894433,
|
1019 |
+
"learning_rate": 1.5101663790863596e-05,
|
1020 |
+
"loss": 0.4942,
|
1021 |
+
"step": 140
|
1022 |
+
},
|
1023 |
+
{
|
1024 |
+
"epoch": 0.84,
|
1025 |
+
"grad_norm": 0.1216157843375959,
|
1026 |
+
"learning_rate": 1.4067831942033904e-05,
|
1027 |
+
"loss": 0.4904,
|
1028 |
+
"step": 141
|
1029 |
+
},
|
1030 |
+
{
|
1031 |
+
"epoch": 0.84,
|
1032 |
+
"grad_norm": 0.12513029490165878,
|
1033 |
+
"learning_rate": 1.3067972556041752e-05,
|
1034 |
+
"loss": 0.4583,
|
1035 |
+
"step": 142
|
1036 |
+
},
|
1037 |
+
{
|
1038 |
+
"epoch": 0.85,
|
1039 |
+
"grad_norm": 0.11835922329329535,
|
1040 |
+
"learning_rate": 1.210248091775663e-05,
|
1041 |
+
"loss": 0.5281,
|
1042 |
+
"step": 143
|
1043 |
+
},
|
1044 |
+
{
|
1045 |
+
"epoch": 0.86,
|
1046 |
+
"grad_norm": 0.14513302866752326,
|
1047 |
+
"learning_rate": 1.1171738725086833e-05,
|
1048 |
+
"loss": 0.4432,
|
1049 |
+
"step": 144
|
1050 |
+
},
|
1051 |
+
{
|
1052 |
+
"epoch": 0.86,
|
1053 |
+
"grad_norm": 0.11934584390756656,
|
1054 |
+
"learning_rate": 1.0276113938078769e-05,
|
1055 |
+
"loss": 0.4664,
|
1056 |
+
"step": 145
|
1057 |
+
},
|
1058 |
+
{
|
1059 |
+
"epoch": 0.87,
|
1060 |
+
"grad_norm": 0.11746932949614314,
|
1061 |
+
"learning_rate": 9.415960633447674e-06,
|
1062 |
+
"loss": 0.4452,
|
1063 |
+
"step": 146
|
1064 |
+
},
|
1065 |
+
{
|
1066 |
+
"epoch": 0.87,
|
1067 |
+
"grad_norm": 0.1183312470782195,
|
1068 |
+
"learning_rate": 8.59161886459654e-06,
|
1069 |
+
"loss": 0.4858,
|
1070 |
+
"step": 147
|
1071 |
+
},
|
1072 |
+
{
|
1073 |
+
"epoch": 0.88,
|
1074 |
+
"grad_norm": 0.1226342260426643,
|
1075 |
+
"learning_rate": 7.803414527179343e-06,
|
1076 |
+
"loss": 0.4609,
|
1077 |
+
"step": 148
|
1078 |
+
},
|
1079 |
+
{
|
1080 |
+
"epoch": 0.89,
|
1081 |
+
"grad_norm": 0.12849903086517628,
|
1082 |
+
"learning_rate": 7.051659230261298e-06,
|
1083 |
+
"loss": 0.4992,
|
1084 |
+
"step": 149
|
1085 |
+
},
|
1086 |
+
{
|
1087 |
+
"epoch": 0.89,
|
1088 |
+
"grad_norm": 0.12558798780888547,
|
1089 |
+
"learning_rate": 6.336650173127223e-06,
|
1090 |
+
"loss": 0.5055,
|
1091 |
+
"step": 150
|
1092 |
+
},
|
1093 |
+
{
|
1094 |
+
"epoch": 0.9,
|
1095 |
+
"grad_norm": 0.11991457404362381,
|
1096 |
+
"learning_rate": 5.658670027786561e-06,
|
1097 |
+
"loss": 0.4711,
|
1098 |
+
"step": 151
|
1099 |
+
},
|
1100 |
+
{
|
1101 |
+
"epoch": 0.9,
|
1102 |
+
"grad_norm": 0.14486189343828018,
|
1103 |
+
"learning_rate": 5.017986827221733e-06,
|
1104 |
+
"loss": 0.4984,
|
1105 |
+
"step": 152
|
1106 |
+
},
|
1107 |
+
{
|
1108 |
+
"epoch": 0.91,
|
1109 |
+
"grad_norm": 0.14292272311516058,
|
1110 |
+
"learning_rate": 4.4148538594239174e-06,
|
1111 |
+
"loss": 0.518,
|
1112 |
+
"step": 153
|
1113 |
+
},
|
1114 |
+
{
|
1115 |
+
"epoch": 0.92,
|
1116 |
+
"grad_norm": 0.12669351883956026,
|
1117 |
+
"learning_rate": 3.849509567257959e-06,
|
1118 |
+
"loss": 0.509,
|
1119 |
+
"step": 154
|
1120 |
+
},
|
1121 |
+
{
|
1122 |
+
"epoch": 0.92,
|
1123 |
+
"grad_norm": 0.14745027617256945,
|
1124 |
+
"learning_rate": 3.3221774541962845e-06,
|
1125 |
+
"loss": 0.4655,
|
1126 |
+
"step": 155
|
1127 |
+
},
|
1128 |
+
{
|
1129 |
+
"epoch": 0.93,
|
1130 |
+
"grad_norm": 0.13904171638337365,
|
1131 |
+
"learning_rate": 2.8330659959589946e-06,
|
1132 |
+
"loss": 0.445,
|
1133 |
+
"step": 156
|
1134 |
+
},
|
1135 |
+
{
|
1136 |
+
"epoch": 0.93,
|
1137 |
+
"grad_norm": 0.12971202836694892,
|
1138 |
+
"learning_rate": 2.3823685580949273e-06,
|
1139 |
+
"loss": 0.4534,
|
1140 |
+
"step": 157
|
1141 |
+
},
|
1142 |
+
{
|
1143 |
+
"epoch": 0.94,
|
1144 |
+
"grad_norm": 0.13437185949159883,
|
1145 |
+
"learning_rate": 1.9702633195363917e-06,
|
1146 |
+
"loss": 0.4785,
|
1147 |
+
"step": 158
|
1148 |
+
},
|
1149 |
+
{
|
1150 |
+
"epoch": 0.95,
|
1151 |
+
"grad_norm": 0.12319276879857373,
|
1152 |
+
"learning_rate": 1.5969132021579347e-06,
|
1153 |
+
"loss": 0.5011,
|
1154 |
+
"step": 159
|
1155 |
+
},
|
1156 |
+
{
|
1157 |
+
"epoch": 0.95,
|
1158 |
+
"grad_norm": 0.12918661435233847,
|
1159 |
+
"learning_rate": 1.2624658063666639e-06,
|
1160 |
+
"loss": 0.5236,
|
1161 |
+
"step": 160
|
1162 |
+
},
|
1163 |
+
{
|
1164 |
+
"epoch": 0.96,
|
1165 |
+
"grad_norm": 0.12874241011432666,
|
1166 |
+
"learning_rate": 9.670533527498137e-07,
|
1167 |
+
"loss": 0.4439,
|
1168 |
+
"step": 161
|
1169 |
+
},
|
1170 |
+
{
|
1171 |
+
"epoch": 0.96,
|
1172 |
+
"grad_norm": 0.1315361924610259,
|
1173 |
+
"learning_rate": 7.10792629802659e-07,
|
1174 |
+
"loss": 0.4929,
|
1175 |
+
"step": 162
|
1176 |
+
},
|
1177 |
+
{
|
1178 |
+
"epoch": 0.97,
|
1179 |
+
"grad_norm": 0.1336404295007477,
|
1180 |
+
"learning_rate": 4.937849477572587e-07,
|
1181 |
+
"loss": 0.433,
|
1182 |
+
"step": 163
|
1183 |
+
},
|
1184 |
+
{
|
1185 |
+
"epoch": 0.97,
|
1186 |
+
"grad_norm": 0.11865993758025514,
|
1187 |
+
"learning_rate": 3.161160985304168e-07,
|
1188 |
+
"loss": 0.4266,
|
1189 |
+
"step": 164
|
1190 |
+
},
|
1191 |
+
{
|
1192 |
+
"epoch": 0.98,
|
1193 |
+
"grad_norm": 0.12559739797694514,
|
1194 |
+
"learning_rate": 1.7785632180670198e-07,
|
1195 |
+
"loss": 0.436,
|
1196 |
+
"step": 165
|
1197 |
+
},
|
1198 |
+
{
|
1199 |
+
"epoch": 0.99,
|
1200 |
+
"grad_norm": 0.12409278477769135,
|
1201 |
+
"learning_rate": 7.906027726981568e-08,
|
1202 |
+
"loss": 0.5017,
|
1203 |
+
"step": 166
|
1204 |
+
},
|
1205 |
+
{
|
1206 |
+
"epoch": 0.99,
|
1207 |
+
"grad_norm": 0.1398337808177678,
|
1208 |
+
"learning_rate": 1.976702299344435e-08,
|
1209 |
+
"loss": 0.5112,
|
1210 |
+
"step": 167
|
1211 |
+
},
|
1212 |
+
{
|
1213 |
+
"epoch": 1.0,
|
1214 |
+
"grad_norm": 0.11434132222175684,
|
1215 |
+
"learning_rate": 0.0,
|
1216 |
+
"loss": 0.4616,
|
1217 |
+
"step": 168
|
1218 |
+
},
|
1219 |
+
{
|
1220 |
+
"epoch": 1.0,
|
1221 |
+
"eval_loss": 0.9651336073875427,
|
1222 |
+
"eval_runtime": 154.7455,
|
1223 |
+
"eval_samples_per_second": 1.609,
|
1224 |
+
"eval_steps_per_second": 0.808,
|
1225 |
+
"step": 168
|
1226 |
+
}
|
1227 |
+
],
|
1228 |
+
"logging_steps": 1,
|
1229 |
+
"max_steps": 168,
|
1230 |
+
"num_input_tokens_seen": 0,
|
1231 |
+
"num_train_epochs": 1,
|
1232 |
+
"save_steps": 42,
|
1233 |
+
"total_flos": 3.406567776272253e+18,
|
1234 |
+
"train_batch_size": 1,
|
1235 |
+
"trial_name": null,
|
1236 |
+
"trial_params": null
|
1237 |
+
}
|
checkpoint-168/training_args.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:cbfaffe231fe183758cc2353296944fef75710807fa8ea7663978d9e906d00f7
|
3 |
+
size 6968
|
checkpoint-168/zero_to_fp32.py
ADDED
@@ -0,0 +1,592 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python
|
2 |
+
|
3 |
+
# Copyright (c) Microsoft Corporation.
|
4 |
+
# SPDX-License-Identifier: Apache-2.0
|
5 |
+
|
6 |
+
# DeepSpeed Team
|
7 |
+
|
8 |
+
# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
|
9 |
+
# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
|
10 |
+
# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
|
11 |
+
# application.
|
12 |
+
#
|
13 |
+
# example: python zero_to_fp32.py . pytorch_model.bin
|
14 |
+
|
15 |
+
import argparse
|
16 |
+
import torch
|
17 |
+
import glob
|
18 |
+
import math
|
19 |
+
import os
|
20 |
+
import re
|
21 |
+
from collections import OrderedDict
|
22 |
+
from dataclasses import dataclass
|
23 |
+
|
24 |
+
# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
|
25 |
+
# DeepSpeed data structures it has to be available in the current python environment.
|
26 |
+
from deepspeed.utils import logger
|
27 |
+
from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
|
28 |
+
FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
|
29 |
+
FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
|
30 |
+
|
31 |
+
|
32 |
+
@dataclass
|
33 |
+
class zero_model_state:
|
34 |
+
buffers: dict()
|
35 |
+
param_shapes: dict()
|
36 |
+
shared_params: list
|
37 |
+
ds_version: int
|
38 |
+
frozen_param_shapes: dict()
|
39 |
+
frozen_param_fragments: dict()
|
40 |
+
|
41 |
+
|
42 |
+
debug = 0
|
43 |
+
|
44 |
+
# load to cpu
|
45 |
+
device = torch.device('cpu')
|
46 |
+
|
47 |
+
|
48 |
+
def atoi(text):
|
49 |
+
return int(text) if text.isdigit() else text
|
50 |
+
|
51 |
+
|
52 |
+
def natural_keys(text):
|
53 |
+
'''
|
54 |
+
alist.sort(key=natural_keys) sorts in human order
|
55 |
+
http://nedbatchelder.com/blog/200712/human_sorting.html
|
56 |
+
(See Toothy's implementation in the comments)
|
57 |
+
'''
|
58 |
+
return [atoi(c) for c in re.split(r'(\d+)', text)]
|
59 |
+
|
60 |
+
|
61 |
+
def get_model_state_file(checkpoint_dir, zero_stage):
|
62 |
+
if not os.path.isdir(checkpoint_dir):
|
63 |
+
raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
|
64 |
+
|
65 |
+
# there should be only one file
|
66 |
+
if zero_stage <= 2:
|
67 |
+
file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
|
68 |
+
elif zero_stage == 3:
|
69 |
+
file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
|
70 |
+
|
71 |
+
if not os.path.exists(file):
|
72 |
+
raise FileNotFoundError(f"can't find model states file at '{file}'")
|
73 |
+
|
74 |
+
return file
|
75 |
+
|
76 |
+
|
77 |
+
def get_checkpoint_files(checkpoint_dir, glob_pattern):
|
78 |
+
# XXX: need to test that this simple glob rule works for multi-node setup too
|
79 |
+
ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
|
80 |
+
|
81 |
+
if len(ckpt_files) == 0:
|
82 |
+
raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
|
83 |
+
|
84 |
+
return ckpt_files
|
85 |
+
|
86 |
+
|
87 |
+
def get_optim_files(checkpoint_dir):
|
88 |
+
return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
|
89 |
+
|
90 |
+
|
91 |
+
def get_model_state_files(checkpoint_dir):
|
92 |
+
return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
|
93 |
+
|
94 |
+
|
95 |
+
def parse_model_states(files):
|
96 |
+
zero_model_states = []
|
97 |
+
for file in files:
|
98 |
+
state_dict = torch.load(file, map_location=device)
|
99 |
+
|
100 |
+
if BUFFER_NAMES not in state_dict:
|
101 |
+
raise ValueError(f"{file} is not a model state checkpoint")
|
102 |
+
buffer_names = state_dict[BUFFER_NAMES]
|
103 |
+
if debug:
|
104 |
+
print("Found buffers:", buffer_names)
|
105 |
+
|
106 |
+
# recover just the buffers while restoring them to fp32 if they were saved in fp16
|
107 |
+
buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
|
108 |
+
param_shapes = state_dict[PARAM_SHAPES]
|
109 |
+
|
110 |
+
# collect parameters that are included in param_shapes
|
111 |
+
param_names = []
|
112 |
+
for s in param_shapes:
|
113 |
+
for name in s.keys():
|
114 |
+
param_names.append(name)
|
115 |
+
|
116 |
+
# update with frozen parameters
|
117 |
+
frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
|
118 |
+
if frozen_param_shapes is not None:
|
119 |
+
if debug:
|
120 |
+
print(f"Found frozen_param_shapes: {frozen_param_shapes}")
|
121 |
+
param_names += list(frozen_param_shapes.keys())
|
122 |
+
|
123 |
+
# handle shared params
|
124 |
+
shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
|
125 |
+
|
126 |
+
ds_version = state_dict.get(DS_VERSION, None)
|
127 |
+
|
128 |
+
frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
|
129 |
+
|
130 |
+
z_model_state = zero_model_state(buffers=buffers,
|
131 |
+
param_shapes=param_shapes,
|
132 |
+
shared_params=shared_params,
|
133 |
+
ds_version=ds_version,
|
134 |
+
frozen_param_shapes=frozen_param_shapes,
|
135 |
+
frozen_param_fragments=frozen_param_fragments)
|
136 |
+
zero_model_states.append(z_model_state)
|
137 |
+
|
138 |
+
return zero_model_states
|
139 |
+
|
140 |
+
|
141 |
+
def parse_optim_states(files, ds_checkpoint_dir):
|
142 |
+
|
143 |
+
total_files = len(files)
|
144 |
+
state_dicts = []
|
145 |
+
for f in files:
|
146 |
+
state_dict = torch.load(f, map_location=device)
|
147 |
+
# immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
|
148 |
+
# and also handle the case where it was already removed by another helper script
|
149 |
+
state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
|
150 |
+
state_dicts.append(state_dict)
|
151 |
+
|
152 |
+
if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
|
153 |
+
raise ValueError(f"{files[0]} is not a zero checkpoint")
|
154 |
+
zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
|
155 |
+
world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
|
156 |
+
|
157 |
+
# For ZeRO-2 each param group can have different partition_count as data parallelism for expert
|
158 |
+
# parameters can be different from data parallelism for non-expert parameters. So we can just
|
159 |
+
# use the max of the partition_count to get the dp world_size.
|
160 |
+
|
161 |
+
if type(world_size) is list:
|
162 |
+
world_size = max(world_size)
|
163 |
+
|
164 |
+
if world_size != total_files:
|
165 |
+
raise ValueError(
|
166 |
+
f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
|
167 |
+
"Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
|
168 |
+
)
|
169 |
+
|
170 |
+
# the groups are named differently in each stage
|
171 |
+
if zero_stage <= 2:
|
172 |
+
fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
|
173 |
+
elif zero_stage == 3:
|
174 |
+
fp32_groups_key = FP32_FLAT_GROUPS
|
175 |
+
else:
|
176 |
+
raise ValueError(f"unknown zero stage {zero_stage}")
|
177 |
+
|
178 |
+
if zero_stage <= 2:
|
179 |
+
fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
|
180 |
+
elif zero_stage == 3:
|
181 |
+
# if there is more than one param group, there will be multiple flattened tensors - one
|
182 |
+
# flattened tensor per group - for simplicity merge them into a single tensor
|
183 |
+
#
|
184 |
+
# XXX: could make the script more memory efficient for when there are multiple groups - it
|
185 |
+
# will require matching the sub-lists of param_shapes for each param group flattened tensor
|
186 |
+
|
187 |
+
fp32_flat_groups = [
|
188 |
+
torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
|
189 |
+
]
|
190 |
+
|
191 |
+
return zero_stage, world_size, fp32_flat_groups
|
192 |
+
|
193 |
+
|
194 |
+
def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
|
195 |
+
"""
|
196 |
+
Returns fp32 state_dict reconstructed from ds checkpoint
|
197 |
+
|
198 |
+
Args:
|
199 |
+
- ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
|
200 |
+
|
201 |
+
"""
|
202 |
+
print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
|
203 |
+
|
204 |
+
optim_files = get_optim_files(ds_checkpoint_dir)
|
205 |
+
zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
|
206 |
+
print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
|
207 |
+
|
208 |
+
model_files = get_model_state_files(ds_checkpoint_dir)
|
209 |
+
|
210 |
+
zero_model_states = parse_model_states(model_files)
|
211 |
+
print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
|
212 |
+
|
213 |
+
if zero_stage <= 2:
|
214 |
+
return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
|
215 |
+
elif zero_stage == 3:
|
216 |
+
return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
|
217 |
+
|
218 |
+
|
219 |
+
def _zero2_merge_frozen_params(state_dict, zero_model_states):
|
220 |
+
if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
|
221 |
+
return
|
222 |
+
|
223 |
+
frozen_param_shapes = zero_model_states[0].frozen_param_shapes
|
224 |
+
frozen_param_fragments = zero_model_states[0].frozen_param_fragments
|
225 |
+
|
226 |
+
if debug:
|
227 |
+
num_elem = sum(s.numel() for s in frozen_param_shapes.values())
|
228 |
+
print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
|
229 |
+
|
230 |
+
wanted_params = len(frozen_param_shapes)
|
231 |
+
wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
|
232 |
+
avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
|
233 |
+
print(f'Frozen params: Have {avail_numel} numels to process.')
|
234 |
+
print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
|
235 |
+
|
236 |
+
total_params = 0
|
237 |
+
total_numel = 0
|
238 |
+
for name, shape in frozen_param_shapes.items():
|
239 |
+
total_params += 1
|
240 |
+
unpartitioned_numel = shape.numel()
|
241 |
+
total_numel += unpartitioned_numel
|
242 |
+
|
243 |
+
state_dict[name] = frozen_param_fragments[name]
|
244 |
+
|
245 |
+
if debug:
|
246 |
+
print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
|
247 |
+
|
248 |
+
print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
|
249 |
+
|
250 |
+
|
251 |
+
def _has_callable(obj, fn):
|
252 |
+
attr = getattr(obj, fn, None)
|
253 |
+
return callable(attr)
|
254 |
+
|
255 |
+
|
256 |
+
def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
|
257 |
+
param_shapes = zero_model_states[0].param_shapes
|
258 |
+
|
259 |
+
# Reconstruction protocol:
|
260 |
+
#
|
261 |
+
# XXX: document this
|
262 |
+
|
263 |
+
if debug:
|
264 |
+
for i in range(world_size):
|
265 |
+
for j in range(len(fp32_flat_groups[0])):
|
266 |
+
print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
|
267 |
+
|
268 |
+
# XXX: memory usage doubles here (zero2)
|
269 |
+
num_param_groups = len(fp32_flat_groups[0])
|
270 |
+
merged_single_partition_of_fp32_groups = []
|
271 |
+
for i in range(num_param_groups):
|
272 |
+
merged_partitions = [sd[i] for sd in fp32_flat_groups]
|
273 |
+
full_single_fp32_vector = torch.cat(merged_partitions, 0)
|
274 |
+
merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
|
275 |
+
avail_numel = sum(
|
276 |
+
[full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
|
277 |
+
|
278 |
+
if debug:
|
279 |
+
wanted_params = sum([len(shapes) for shapes in param_shapes])
|
280 |
+
wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
|
281 |
+
# not asserting if there is a mismatch due to possible padding
|
282 |
+
print(f"Have {avail_numel} numels to process.")
|
283 |
+
print(f"Need {wanted_numel} numels in {wanted_params} params.")
|
284 |
+
|
285 |
+
# params
|
286 |
+
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
287 |
+
# out-of-core computing solution
|
288 |
+
total_numel = 0
|
289 |
+
total_params = 0
|
290 |
+
for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
|
291 |
+
offset = 0
|
292 |
+
avail_numel = full_single_fp32_vector.numel()
|
293 |
+
for name, shape in shapes.items():
|
294 |
+
|
295 |
+
unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
|
296 |
+
total_numel += unpartitioned_numel
|
297 |
+
total_params += 1
|
298 |
+
|
299 |
+
if debug:
|
300 |
+
print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
|
301 |
+
state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
|
302 |
+
offset += unpartitioned_numel
|
303 |
+
|
304 |
+
# Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
|
305 |
+
# avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
|
306 |
+
# paddings performed in the code it's almost impossible to predict the exact numbers w/o the
|
307 |
+
# live optimizer object, so we are checking that the numbers are within the right range
|
308 |
+
align_to = 2 * world_size
|
309 |
+
|
310 |
+
def zero2_align(x):
|
311 |
+
return align_to * math.ceil(x / align_to)
|
312 |
+
|
313 |
+
if debug:
|
314 |
+
print(f"original offset={offset}, avail_numel={avail_numel}")
|
315 |
+
|
316 |
+
offset = zero2_align(offset)
|
317 |
+
avail_numel = zero2_align(avail_numel)
|
318 |
+
|
319 |
+
if debug:
|
320 |
+
print(f"aligned offset={offset}, avail_numel={avail_numel}")
|
321 |
+
|
322 |
+
# Sanity check
|
323 |
+
if offset != avail_numel:
|
324 |
+
raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
325 |
+
|
326 |
+
print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
|
327 |
+
|
328 |
+
|
329 |
+
def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
|
330 |
+
state_dict = OrderedDict()
|
331 |
+
|
332 |
+
# buffers
|
333 |
+
buffers = zero_model_states[0].buffers
|
334 |
+
state_dict.update(buffers)
|
335 |
+
if debug:
|
336 |
+
print(f"added {len(buffers)} buffers")
|
337 |
+
|
338 |
+
_zero2_merge_frozen_params(state_dict, zero_model_states)
|
339 |
+
|
340 |
+
_zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
|
341 |
+
|
342 |
+
# recover shared parameters
|
343 |
+
for pair in zero_model_states[0].shared_params:
|
344 |
+
if pair[1] in state_dict:
|
345 |
+
state_dict[pair[0]] = state_dict[pair[1]]
|
346 |
+
|
347 |
+
return state_dict
|
348 |
+
|
349 |
+
|
350 |
+
def zero3_partitioned_param_info(unpartitioned_numel, world_size):
|
351 |
+
remainder = unpartitioned_numel % world_size
|
352 |
+
padding_numel = (world_size - remainder) if remainder else 0
|
353 |
+
partitioned_numel = math.ceil(unpartitioned_numel / world_size)
|
354 |
+
return partitioned_numel, padding_numel
|
355 |
+
|
356 |
+
|
357 |
+
def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
|
358 |
+
if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
|
359 |
+
return
|
360 |
+
|
361 |
+
if debug:
|
362 |
+
for i in range(world_size):
|
363 |
+
num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
|
364 |
+
print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
|
365 |
+
|
366 |
+
frozen_param_shapes = zero_model_states[0].frozen_param_shapes
|
367 |
+
wanted_params = len(frozen_param_shapes)
|
368 |
+
wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
|
369 |
+
avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
|
370 |
+
print(f'Frozen params: Have {avail_numel} numels to process.')
|
371 |
+
print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
|
372 |
+
|
373 |
+
total_params = 0
|
374 |
+
total_numel = 0
|
375 |
+
for name, shape in zero_model_states[0].frozen_param_shapes.items():
|
376 |
+
total_params += 1
|
377 |
+
unpartitioned_numel = shape.numel()
|
378 |
+
total_numel += unpartitioned_numel
|
379 |
+
|
380 |
+
param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
|
381 |
+
state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
|
382 |
+
|
383 |
+
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
384 |
+
|
385 |
+
if debug:
|
386 |
+
print(
|
387 |
+
f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
388 |
+
)
|
389 |
+
|
390 |
+
print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
|
391 |
+
|
392 |
+
|
393 |
+
def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
|
394 |
+
param_shapes = zero_model_states[0].param_shapes
|
395 |
+
avail_numel = fp32_flat_groups[0].numel() * world_size
|
396 |
+
# Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
|
397 |
+
# param, re-consolidating each param, while dealing with padding if any
|
398 |
+
|
399 |
+
# merge list of dicts, preserving order
|
400 |
+
param_shapes = {k: v for d in param_shapes for k, v in d.items()}
|
401 |
+
|
402 |
+
if debug:
|
403 |
+
for i in range(world_size):
|
404 |
+
print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
|
405 |
+
|
406 |
+
wanted_params = len(param_shapes)
|
407 |
+
wanted_numel = sum(shape.numel() for shape in param_shapes.values())
|
408 |
+
# not asserting if there is a mismatch due to possible padding
|
409 |
+
avail_numel = fp32_flat_groups[0].numel() * world_size
|
410 |
+
print(f"Trainable params: Have {avail_numel} numels to process.")
|
411 |
+
print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
|
412 |
+
|
413 |
+
# params
|
414 |
+
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
415 |
+
# out-of-core computing solution
|
416 |
+
offset = 0
|
417 |
+
total_numel = 0
|
418 |
+
total_params = 0
|
419 |
+
for name, shape in param_shapes.items():
|
420 |
+
|
421 |
+
unpartitioned_numel = shape.numel()
|
422 |
+
total_numel += unpartitioned_numel
|
423 |
+
total_params += 1
|
424 |
+
|
425 |
+
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
426 |
+
|
427 |
+
if debug:
|
428 |
+
print(
|
429 |
+
f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
430 |
+
)
|
431 |
+
|
432 |
+
# XXX: memory usage doubles here
|
433 |
+
state_dict[name] = torch.cat(
|
434 |
+
tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
|
435 |
+
0).narrow(0, 0, unpartitioned_numel).view(shape)
|
436 |
+
offset += partitioned_numel
|
437 |
+
|
438 |
+
offset *= world_size
|
439 |
+
|
440 |
+
# Sanity check
|
441 |
+
if offset != avail_numel:
|
442 |
+
raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
443 |
+
|
444 |
+
print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
|
445 |
+
|
446 |
+
|
447 |
+
def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
|
448 |
+
state_dict = OrderedDict()
|
449 |
+
|
450 |
+
# buffers
|
451 |
+
buffers = zero_model_states[0].buffers
|
452 |
+
state_dict.update(buffers)
|
453 |
+
if debug:
|
454 |
+
print(f"added {len(buffers)} buffers")
|
455 |
+
|
456 |
+
_zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
|
457 |
+
|
458 |
+
_zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
|
459 |
+
|
460 |
+
# recover shared parameters
|
461 |
+
for pair in zero_model_states[0].shared_params:
|
462 |
+
if pair[1] in state_dict:
|
463 |
+
state_dict[pair[0]] = state_dict[pair[1]]
|
464 |
+
|
465 |
+
return state_dict
|
466 |
+
|
467 |
+
|
468 |
+
def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
|
469 |
+
"""
|
470 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
|
471 |
+
``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
|
472 |
+
via a model hub.
|
473 |
+
|
474 |
+
Args:
|
475 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder
|
476 |
+
- ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
|
477 |
+
|
478 |
+
Returns:
|
479 |
+
- pytorch ``state_dict``
|
480 |
+
|
481 |
+
Note: this approach may not work if your application doesn't have sufficient free CPU memory and
|
482 |
+
you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
|
483 |
+
the checkpoint.
|
484 |
+
|
485 |
+
A typical usage might be ::
|
486 |
+
|
487 |
+
from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
|
488 |
+
# do the training and checkpoint saving
|
489 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
|
490 |
+
model = model.cpu() # move to cpu
|
491 |
+
model.load_state_dict(state_dict)
|
492 |
+
# submit to model hub or save the model to share with others
|
493 |
+
|
494 |
+
In this example the ``model`` will no longer be usable in the deepspeed context of the same
|
495 |
+
application. i.e. you will need to re-initialize the deepspeed engine, since
|
496 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
497 |
+
|
498 |
+
If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
|
499 |
+
|
500 |
+
"""
|
501 |
+
if tag is None:
|
502 |
+
latest_path = os.path.join(checkpoint_dir, 'latest')
|
503 |
+
if os.path.isfile(latest_path):
|
504 |
+
with open(latest_path, 'r') as fd:
|
505 |
+
tag = fd.read().strip()
|
506 |
+
else:
|
507 |
+
raise ValueError(f"Unable to find 'latest' file at {latest_path}")
|
508 |
+
|
509 |
+
ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
|
510 |
+
|
511 |
+
if not os.path.isdir(ds_checkpoint_dir):
|
512 |
+
raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
|
513 |
+
|
514 |
+
return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
|
515 |
+
|
516 |
+
|
517 |
+
def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
|
518 |
+
"""
|
519 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
|
520 |
+
loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
|
521 |
+
|
522 |
+
Args:
|
523 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
524 |
+
- ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
|
525 |
+
- ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
|
526 |
+
"""
|
527 |
+
|
528 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
|
529 |
+
print(f"Saving fp32 state dict to {output_file}")
|
530 |
+
torch.save(state_dict, output_file)
|
531 |
+
|
532 |
+
|
533 |
+
def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
|
534 |
+
"""
|
535 |
+
1. Put the provided model to cpu
|
536 |
+
2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
|
537 |
+
3. Load it into the provided model
|
538 |
+
|
539 |
+
Args:
|
540 |
+
- ``model``: the model object to update
|
541 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
542 |
+
- ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
|
543 |
+
|
544 |
+
Returns:
|
545 |
+
- ``model`: modified model
|
546 |
+
|
547 |
+
Make sure you have plenty of CPU memory available before you call this function. If you don't
|
548 |
+
have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
|
549 |
+
conveniently placed for you in the checkpoint folder.
|
550 |
+
|
551 |
+
A typical usage might be ::
|
552 |
+
|
553 |
+
from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
|
554 |
+
model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
|
555 |
+
# submit to model hub or save the model to share with others
|
556 |
+
|
557 |
+
Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
|
558 |
+
of the same application. i.e. you will need to re-initialize the deepspeed engine, since
|
559 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
560 |
+
|
561 |
+
"""
|
562 |
+
logger.info(f"Extracting fp32 weights")
|
563 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
|
564 |
+
|
565 |
+
logger.info(f"Overwriting model with fp32 weights")
|
566 |
+
model = model.cpu()
|
567 |
+
model.load_state_dict(state_dict, strict=False)
|
568 |
+
|
569 |
+
return model
|
570 |
+
|
571 |
+
|
572 |
+
if __name__ == "__main__":
|
573 |
+
|
574 |
+
parser = argparse.ArgumentParser()
|
575 |
+
parser.add_argument("checkpoint_dir",
|
576 |
+
type=str,
|
577 |
+
help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
|
578 |
+
parser.add_argument(
|
579 |
+
"output_file",
|
580 |
+
type=str,
|
581 |
+
help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
|
582 |
+
parser.add_argument("-t",
|
583 |
+
"--tag",
|
584 |
+
type=str,
|
585 |
+
default=None,
|
586 |
+
help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
|
587 |
+
parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
|
588 |
+
args = parser.parse_args()
|
589 |
+
|
590 |
+
debug = args.debug
|
591 |
+
|
592 |
+
convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)
|
config.json
CHANGED
@@ -40,15 +40,13 @@
|
|
40 |
"quantization_config": {
|
41 |
"_load_in_4bit": true,
|
42 |
"_load_in_8bit": false,
|
43 |
-
"bnb_4bit_compute_dtype": "
|
44 |
-
"bnb_4bit_quant_storage": "
|
45 |
-
"bnb_4bit_quant_type": "
|
46 |
-
"bnb_4bit_use_double_quant":
|
47 |
"llm_int8_enable_fp32_cpu_offload": false,
|
48 |
"llm_int8_has_fp16_weight": false,
|
49 |
-
"llm_int8_skip_modules":
|
50 |
-
"mamba"
|
51 |
-
],
|
52 |
"llm_int8_threshold": 6.0,
|
53 |
"load_in_4bit": true,
|
54 |
"load_in_8bit": false,
|
@@ -58,9 +56,9 @@
|
|
58 |
"router_aux_loss_coef": 0.001,
|
59 |
"sliding_window": null,
|
60 |
"tie_word_embeddings": false,
|
61 |
-
"torch_dtype": "
|
62 |
"transformers_version": "4.40.0.dev0",
|
63 |
-
"use_cache":
|
64 |
"use_mamba_kernels": true,
|
65 |
"vocab_size": 65536
|
66 |
}
|
|
|
40 |
"quantization_config": {
|
41 |
"_load_in_4bit": true,
|
42 |
"_load_in_8bit": false,
|
43 |
+
"bnb_4bit_compute_dtype": "bfloat16",
|
44 |
+
"bnb_4bit_quant_storage": "bfloat16",
|
45 |
+
"bnb_4bit_quant_type": "nf4",
|
46 |
+
"bnb_4bit_use_double_quant": true,
|
47 |
"llm_int8_enable_fp32_cpu_offload": false,
|
48 |
"llm_int8_has_fp16_weight": false,
|
49 |
+
"llm_int8_skip_modules": null,
|
|
|
|
|
50 |
"llm_int8_threshold": 6.0,
|
51 |
"load_in_4bit": true,
|
52 |
"load_in_8bit": false,
|
|
|
56 |
"router_aux_loss_coef": 0.001,
|
57 |
"sliding_window": null,
|
58 |
"tie_word_embeddings": false,
|
59 |
+
"torch_dtype": "bfloat16",
|
60 |
"transformers_version": "4.40.0.dev0",
|
61 |
+
"use_cache": false,
|
62 |
"use_mamba_kernels": true,
|
63 |
"vocab_size": 65536
|
64 |
}
|
tokenizer_config.json
CHANGED
@@ -36,6 +36,7 @@
|
|
36 |
}
|
37 |
},
|
38 |
"bos_token": "<|startoftext|>",
|
|
|
39 |
"clean_up_tokenization_spaces": false,
|
40 |
"eos_token": "<|endoftext|>",
|
41 |
"model_max_length": 1000000000000000019884624838656,
|
|
|
36 |
}
|
37 |
},
|
38 |
"bos_token": "<|startoftext|>",
|
39 |
+
"chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
|
40 |
"clean_up_tokenization_spaces": false,
|
41 |
"eos_token": "<|endoftext|>",
|
42 |
"model_max_length": 1000000000000000019884624838656,
|