End of training
Browse files
README.md
ADDED
@@ -0,0 +1,164 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
base_model: microsoft/Phi-3-mini-4k-instruct
|
3 |
+
library_name: peft
|
4 |
+
license: mit
|
5 |
+
tags:
|
6 |
+
- axolotl
|
7 |
+
- generated_from_trainer
|
8 |
+
model-index:
|
9 |
+
- name: phi3-nosys-gpt4ominiplans-27k-512rank-long
|
10 |
+
results: []
|
11 |
+
---
|
12 |
+
|
13 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
14 |
+
should probably proofread and complete it, then remove this comment. -->
|
15 |
+
|
16 |
+
[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
|
17 |
+
<details><summary>See axolotl config</summary>
|
18 |
+
|
19 |
+
axolotl version: `0.4.1`
|
20 |
+
```yaml
|
21 |
+
# model and tokenizer
|
22 |
+
base_model: microsoft/Phi-3-mini-4k-instruct # change for model
|
23 |
+
trust_remote_code: true
|
24 |
+
sequence_len: 2048
|
25 |
+
|
26 |
+
strict: false
|
27 |
+
|
28 |
+
model_type: AutoModelForCausalLM
|
29 |
+
tokenizer_type: AutoTokenizer
|
30 |
+
bf16: auto
|
31 |
+
pad_to_sequence_len: true
|
32 |
+
save_safetensors: true
|
33 |
+
|
34 |
+
|
35 |
+
datasets:
|
36 |
+
- path: verifiers-for-code/sampled_10k_from_27k
|
37 |
+
type: completion
|
38 |
+
field: text_nosys_phi
|
39 |
+
train_on_split: train
|
40 |
+
|
41 |
+
val_set_size: 0.05
|
42 |
+
|
43 |
+
# lora
|
44 |
+
adapter: lora
|
45 |
+
lora_r: 512
|
46 |
+
lora_alpha: 32
|
47 |
+
lora_dropout: 0.05
|
48 |
+
lora_target_linear: true
|
49 |
+
lora_modules_to_save:
|
50 |
+
- embed_tokens
|
51 |
+
- lm_head
|
52 |
+
use_rslora: true
|
53 |
+
|
54 |
+
# logging
|
55 |
+
wandb_project: valeris
|
56 |
+
wandb_name: phi3-nosys-gpt4ominiplans-27k-512rank-long
|
57 |
+
|
58 |
+
output_dir: ./outputs/phi3-nosys-gpt4ominiplans-27k-512rank-long
|
59 |
+
|
60 |
+
gradient_accumulation_steps: 2
|
61 |
+
gradient_checkpointing: true
|
62 |
+
gradient_checkpointing_kwargs:
|
63 |
+
use_reentrant: true
|
64 |
+
micro_batch_size: 2
|
65 |
+
num_epochs: 3
|
66 |
+
eval_batch_size: 2
|
67 |
+
warmup_ratio: 0.05
|
68 |
+
learning_rate: 1e-5
|
69 |
+
lr_scheduler: cosine
|
70 |
+
optimizer: adamw_torch
|
71 |
+
|
72 |
+
hub_model_id: verifiers-for-code/phi3-nosys-gpt4ominiplans-27k-512rank-long
|
73 |
+
push_to_hub: true
|
74 |
+
hub_strategy: all_checkpoints
|
75 |
+
hub_always_push: true
|
76 |
+
evals_per_epoch: 8
|
77 |
+
saves_per_epoch: 4
|
78 |
+
logging_steps: 1
|
79 |
+
# eval_table_size: 10
|
80 |
+
# eval_max_new_tokens: 512
|
81 |
+
|
82 |
+
tokens: ["<thinking>", "</thinking>", "<plan>", "</plan>"]
|
83 |
+
|
84 |
+
special_tokens:
|
85 |
+
pad_token: "<|endoftext|>"
|
86 |
+
|
87 |
+
```
|
88 |
+
|
89 |
+
</details><br>
|
90 |
+
|
91 |
+
# phi3-nosys-gpt4ominiplans-27k-512rank-long
|
92 |
+
|
93 |
+
This model is a fine-tuned version of [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct) on the None dataset.
|
94 |
+
It achieves the following results on the evaluation set:
|
95 |
+
- Loss: 0.6378
|
96 |
+
|
97 |
+
## Model description
|
98 |
+
|
99 |
+
More information needed
|
100 |
+
|
101 |
+
## Intended uses & limitations
|
102 |
+
|
103 |
+
More information needed
|
104 |
+
|
105 |
+
## Training and evaluation data
|
106 |
+
|
107 |
+
More information needed
|
108 |
+
|
109 |
+
## Training procedure
|
110 |
+
|
111 |
+
### Training hyperparameters
|
112 |
+
|
113 |
+
The following hyperparameters were used during training:
|
114 |
+
- learning_rate: 1e-05
|
115 |
+
- train_batch_size: 2
|
116 |
+
- eval_batch_size: 2
|
117 |
+
- seed: 42
|
118 |
+
- distributed_type: multi-GPU
|
119 |
+
- num_devices: 8
|
120 |
+
- gradient_accumulation_steps: 2
|
121 |
+
- total_train_batch_size: 32
|
122 |
+
- total_eval_batch_size: 16
|
123 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
124 |
+
- lr_scheduler_type: cosine
|
125 |
+
- lr_scheduler_warmup_steps: 44
|
126 |
+
- num_epochs: 3
|
127 |
+
|
128 |
+
### Training results
|
129 |
+
|
130 |
+
| Training Loss | Epoch | Step | Validation Loss |
|
131 |
+
|:-------------:|:------:|:----:|:---------------:|
|
132 |
+
| 1.0833 | 0.0034 | 1 | 1.0330 |
|
133 |
+
| 1.0093 | 0.1279 | 38 | 0.9910 |
|
134 |
+
| 0.9169 | 0.2559 | 76 | 0.8668 |
|
135 |
+
| 0.795 | 0.3838 | 114 | 0.7676 |
|
136 |
+
| 0.6999 | 0.5118 | 152 | 0.7243 |
|
137 |
+
| 0.7246 | 0.6397 | 190 | 0.6989 |
|
138 |
+
| 0.6873 | 0.7677 | 228 | 0.6816 |
|
139 |
+
| 0.7014 | 0.8956 | 266 | 0.6687 |
|
140 |
+
| 0.6586 | 1.0236 | 304 | 0.6585 |
|
141 |
+
| 0.6532 | 1.1515 | 342 | 0.6511 |
|
142 |
+
| 0.6334 | 1.2795 | 380 | 0.6463 |
|
143 |
+
| 0.5968 | 1.4074 | 418 | 0.6434 |
|
144 |
+
| 0.6366 | 1.5354 | 456 | 0.6414 |
|
145 |
+
| 0.6126 | 1.6633 | 494 | 0.6400 |
|
146 |
+
| 0.6564 | 1.7912 | 532 | 0.6391 |
|
147 |
+
| 0.6296 | 1.9192 | 570 | 0.6387 |
|
148 |
+
| 0.6225 | 2.0471 | 608 | 0.6383 |
|
149 |
+
| 0.6354 | 2.1751 | 646 | 0.6381 |
|
150 |
+
| 0.6111 | 2.3030 | 684 | 0.6379 |
|
151 |
+
| 0.5899 | 2.4310 | 722 | 0.6378 |
|
152 |
+
| 0.6415 | 2.5589 | 760 | 0.6378 |
|
153 |
+
| 0.6443 | 2.6869 | 798 | 0.6377 |
|
154 |
+
| 0.6103 | 2.8148 | 836 | 0.6377 |
|
155 |
+
| 0.6451 | 2.9428 | 874 | 0.6378 |
|
156 |
+
|
157 |
+
|
158 |
+
### Framework versions
|
159 |
+
|
160 |
+
- PEFT 0.11.1
|
161 |
+
- Transformers 4.44.0.dev0
|
162 |
+
- Pytorch 2.4.0
|
163 |
+
- Datasets 2.19.1
|
164 |
+
- Tokenizers 0.19.1
|