first commit
Browse filesmpt-7b-delta-tune-model-div-sal
- README.md +7 -1
- adapter_config.json +20 -0
- adapter_model.bin +3 -0
- all_results.json +9 -0
- best_model_structure.txt +1 -0
- checkpoint-985/README.md +9 -0
- checkpoint-985/adapter_config.json +20 -0
- checkpoint-985/adapter_model.bin +3 -0
- checkpoint-985/optimizer.pt +3 -0
- checkpoint-985/rng_state.pth +3 -0
- checkpoint-985/scheduler.pt +3 -0
- checkpoint-985/special_tokens_map.json +6 -0
- checkpoint-985/tokenizer.json +0 -0
- checkpoint-985/tokenizer_config.json +9 -0
- checkpoint-985/trainer_state.json +70 -0
- checkpoint-985/training_args.bin +3 -0
- eval_results.json +9 -0
- mpt-7b-delta-tune-model-div-sal-arc_challenge +27 -0
- mpt-7b-delta-tune-model-div-sal-hellaswag +27 -0
- mpt-7b-delta-tune-model-div-sal-mmlu +419 -0
- mpt-7b-delta-tune-model-div-sal-truthqa +27 -0
README.md
CHANGED
@@ -1,3 +1,9 @@
|
|
1 |
---
|
2 |
-
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
library_name: peft
|
3 |
---
|
4 |
+
## Training procedure
|
5 |
+
|
6 |
+
### Framework versions
|
7 |
+
|
8 |
+
|
9 |
+
- PEFT 0.4.0
|
adapter_config.json
ADDED
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"auto_mapping": null,
|
3 |
+
"base_model_name_or_path": "/home/vmagent/app/dataset/mpt-7b",
|
4 |
+
"bias": "none",
|
5 |
+
"fan_in_fan_out": false,
|
6 |
+
"inference_mode": true,
|
7 |
+
"init_lora_weights": true,
|
8 |
+
"layers_pattern": null,
|
9 |
+
"layers_to_transform": null,
|
10 |
+
"lora_alpha": 55,
|
11 |
+
"lora_dropout": 0.05,
|
12 |
+
"modules_to_save": null,
|
13 |
+
"peft_type": "LORA",
|
14 |
+
"r": 11,
|
15 |
+
"revision": null,
|
16 |
+
"target_modules": [
|
17 |
+
"Wqkv"
|
18 |
+
],
|
19 |
+
"task_type": "CAUSAL_LM"
|
20 |
+
}
|
adapter_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d538ec07bd538200952bf730d5fd0f674aa684c39af25776553151f551fa35ca
|
3 |
+
size 12467253
|
all_results.json
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 1.0,
|
3 |
+
"eval_loss": 0.9799609184265137,
|
4 |
+
"eval_runtime": 5242.7479,
|
5 |
+
"eval_samples": 15601,
|
6 |
+
"eval_samples_per_second": 2.976,
|
7 |
+
"eval_steps_per_second": 0.372,
|
8 |
+
"eval_tokens": 1722455
|
9 |
+
}
|
best_model_structure.txt
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"num_hidden_layers": [0, 1, 0, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 0, 0, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0], "r": [8, 10, 3, 6, 8, 9, 10, 8, 12, 4, 9, 12, 5, 7, 7, 10, 2, 11, 10, 8, 11, 9, 6, 10, 12, 6, 12, 7, 5, 11, 11, 8], "alpha": [2, 6, 3, 3, 5, 3, 1, 2, 6, 2, 6, 5, 1, 6, 3, 6, 5, 2, 4, 3, 4, 4, 5, 5, 5, 4, 5, 4, 6, 6, 5, 5]}
|
checkpoint-985/README.md
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
library_name: peft
|
3 |
+
---
|
4 |
+
## Training procedure
|
5 |
+
|
6 |
+
### Framework versions
|
7 |
+
|
8 |
+
|
9 |
+
- PEFT 0.4.0
|
checkpoint-985/adapter_config.json
ADDED
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"auto_mapping": null,
|
3 |
+
"base_model_name_or_path": "/home/vmagent/app/dataset/mpt-7b",
|
4 |
+
"bias": "none",
|
5 |
+
"fan_in_fan_out": false,
|
6 |
+
"inference_mode": true,
|
7 |
+
"init_lora_weights": true,
|
8 |
+
"layers_pattern": null,
|
9 |
+
"layers_to_transform": null,
|
10 |
+
"lora_alpha": 55,
|
11 |
+
"lora_dropout": 0.05,
|
12 |
+
"modules_to_save": null,
|
13 |
+
"peft_type": "LORA",
|
14 |
+
"r": 11,
|
15 |
+
"revision": null,
|
16 |
+
"target_modules": [
|
17 |
+
"Wqkv"
|
18 |
+
],
|
19 |
+
"task_type": "CAUSAL_LM"
|
20 |
+
}
|
checkpoint-985/adapter_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d538ec07bd538200952bf730d5fd0f674aa684c39af25776553151f551fa35ca
|
3 |
+
size 12467253
|
checkpoint-985/optimizer.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:0e8dd10e1b991d52db97f9d7e24580bde2f62ecf59226c779994cd26c628c27a
|
3 |
+
size 24939333
|
checkpoint-985/rng_state.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d0d37ec6dfe70dfec2d14a6da6156550726b7ca980de43d4a88d7ba6c925404c
|
3 |
+
size 13553
|
checkpoint-985/scheduler.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:35b9333bfcd7d3fa9837943e8744d19f7572eb49c45fded40340ce3c47b46224
|
3 |
+
size 627
|
checkpoint-985/special_tokens_map.json
ADDED
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"bos_token": "<|endoftext|>",
|
3 |
+
"eos_token": "<|endoftext|>",
|
4 |
+
"pad_token": "<|endoftext|>",
|
5 |
+
"unk_token": "<|endoftext|>"
|
6 |
+
}
|
checkpoint-985/tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
checkpoint-985/tokenizer_config.json
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"add_prefix_space": false,
|
3 |
+
"bos_token": "<|endoftext|>",
|
4 |
+
"clean_up_tokenization_spaces": true,
|
5 |
+
"eos_token": "<|endoftext|>",
|
6 |
+
"model_max_length": 2048,
|
7 |
+
"tokenizer_class": "GPTNeoXTokenizer",
|
8 |
+
"unk_token": "<|endoftext|>"
|
9 |
+
}
|
checkpoint-985/trainer_state.json
ADDED
@@ -0,0 +1,70 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"best_metric": null,
|
3 |
+
"best_model_checkpoint": null,
|
4 |
+
"epoch": 1.0,
|
5 |
+
"global_step": 985,
|
6 |
+
"is_hyper_param_search": false,
|
7 |
+
"is_local_process_zero": true,
|
8 |
+
"is_world_process_zero": true,
|
9 |
+
"log_history": [
|
10 |
+
{
|
11 |
+
"epoch": 0.1,
|
12 |
+
"learning_rate": 8.984771573604062e-05,
|
13 |
+
"loss": 1.1743,
|
14 |
+
"step": 100
|
15 |
+
},
|
16 |
+
{
|
17 |
+
"epoch": 0.2,
|
18 |
+
"learning_rate": 7.969543147208121e-05,
|
19 |
+
"loss": 1.0788,
|
20 |
+
"step": 200
|
21 |
+
},
|
22 |
+
{
|
23 |
+
"epoch": 0.3,
|
24 |
+
"learning_rate": 6.954314720812183e-05,
|
25 |
+
"loss": 1.0732,
|
26 |
+
"step": 300
|
27 |
+
},
|
28 |
+
{
|
29 |
+
"epoch": 0.41,
|
30 |
+
"learning_rate": 5.939086294416244e-05,
|
31 |
+
"loss": 1.0638,
|
32 |
+
"step": 400
|
33 |
+
},
|
34 |
+
{
|
35 |
+
"epoch": 0.51,
|
36 |
+
"learning_rate": 4.9238578680203045e-05,
|
37 |
+
"loss": 1.0558,
|
38 |
+
"step": 500
|
39 |
+
},
|
40 |
+
{
|
41 |
+
"epoch": 0.61,
|
42 |
+
"learning_rate": 3.9086294416243655e-05,
|
43 |
+
"loss": 1.0475,
|
44 |
+
"step": 600
|
45 |
+
},
|
46 |
+
{
|
47 |
+
"epoch": 0.71,
|
48 |
+
"learning_rate": 2.8934010152284264e-05,
|
49 |
+
"loss": 1.0485,
|
50 |
+
"step": 700
|
51 |
+
},
|
52 |
+
{
|
53 |
+
"epoch": 0.81,
|
54 |
+
"learning_rate": 1.8781725888324874e-05,
|
55 |
+
"loss": 1.039,
|
56 |
+
"step": 800
|
57 |
+
},
|
58 |
+
{
|
59 |
+
"epoch": 0.91,
|
60 |
+
"learning_rate": 8.629441624365483e-06,
|
61 |
+
"loss": 1.0402,
|
62 |
+
"step": 900
|
63 |
+
}
|
64 |
+
],
|
65 |
+
"max_steps": 985,
|
66 |
+
"num_train_epochs": 1,
|
67 |
+
"total_flos": 1.5601672920445747e+17,
|
68 |
+
"trial_name": null,
|
69 |
+
"trial_params": null
|
70 |
+
}
|
checkpoint-985/training_args.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f08761150839d8b4837eeaeeb20dfb3e2028628b007b9758c4101c7b1ab0d120
|
3 |
+
size 4091
|
eval_results.json
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 1.0,
|
3 |
+
"eval_loss": 0.9799609184265137,
|
4 |
+
"eval_runtime": 5242.7479,
|
5 |
+
"eval_samples": 15601,
|
6 |
+
"eval_samples_per_second": 2.976,
|
7 |
+
"eval_steps_per_second": 0.372,
|
8 |
+
"eval_tokens": 1722455
|
9 |
+
}
|
mpt-7b-delta-tune-model-div-sal-arc_challenge
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"results": {
|
3 |
+
"arc_challenge": {
|
4 |
+
"acc": 0.454778156996587,
|
5 |
+
"acc_stderr": 0.014551507060836353,
|
6 |
+
"acc_norm": 0.5017064846416383,
|
7 |
+
"acc_norm_stderr": 0.014611305705056995
|
8 |
+
}
|
9 |
+
},
|
10 |
+
"versions": {
|
11 |
+
"arc_challenge": 0
|
12 |
+
},
|
13 |
+
"config": {
|
14 |
+
"model": "hf-causal-experimental",
|
15 |
+
"model_args": "pretrained=/home/vmagent/app/data/mpt-7b-delta-tune-model-div-sal/merged_model,use_accelerate=True,trust_remote_code=True,dtype=float16",
|
16 |
+
"num_fewshot": 25,
|
17 |
+
"batch_size": "auto",
|
18 |
+
"batch_sizes": [
|
19 |
+
32
|
20 |
+
],
|
21 |
+
"device": null,
|
22 |
+
"no_cache": false,
|
23 |
+
"limit": null,
|
24 |
+
"bootstrap_iters": 100000,
|
25 |
+
"description_dict": {}
|
26 |
+
}
|
27 |
+
}
|
mpt-7b-delta-tune-model-div-sal-hellaswag
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"results": {
|
3 |
+
"hellaswag": {
|
4 |
+
"acc": 0.5733917546305517,
|
5 |
+
"acc_stderr": 0.004935735300348862,
|
6 |
+
"acc_norm": 0.7630950009958176,
|
7 |
+
"acc_norm_stderr": 0.004243145587737571
|
8 |
+
}
|
9 |
+
},
|
10 |
+
"versions": {
|
11 |
+
"hellaswag": 0
|
12 |
+
},
|
13 |
+
"config": {
|
14 |
+
"model": "hf-causal-experimental",
|
15 |
+
"model_args": "pretrained=/home/vmagent/app/data/mpt-7b-delta-tune-model-div-sal/merged_model,use_accelerate=True,trust_remote_code=True,dtype=float16",
|
16 |
+
"num_fewshot": 10,
|
17 |
+
"batch_size": "auto",
|
18 |
+
"batch_sizes": [
|
19 |
+
32
|
20 |
+
],
|
21 |
+
"device": null,
|
22 |
+
"no_cache": false,
|
23 |
+
"limit": null,
|
24 |
+
"bootstrap_iters": 100000,
|
25 |
+
"description_dict": {}
|
26 |
+
}
|
27 |
+
}
|
mpt-7b-delta-tune-model-div-sal-mmlu
ADDED
@@ -0,0 +1,419 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"results": {
|
3 |
+
"hendrycksTest-abstract_algebra": {
|
4 |
+
"acc": 0.27,
|
5 |
+
"acc_stderr": 0.0446196043338474,
|
6 |
+
"acc_norm": 0.27,
|
7 |
+
"acc_norm_stderr": 0.0446196043338474
|
8 |
+
},
|
9 |
+
"hendrycksTest-anatomy": {
|
10 |
+
"acc": 0.3111111111111111,
|
11 |
+
"acc_stderr": 0.039992628766177214,
|
12 |
+
"acc_norm": 0.3111111111111111,
|
13 |
+
"acc_norm_stderr": 0.039992628766177214
|
14 |
+
},
|
15 |
+
"hendrycksTest-astronomy": {
|
16 |
+
"acc": 0.24342105263157895,
|
17 |
+
"acc_stderr": 0.034923496688842384,
|
18 |
+
"acc_norm": 0.24342105263157895,
|
19 |
+
"acc_norm_stderr": 0.034923496688842384
|
20 |
+
},
|
21 |
+
"hendrycksTest-business_ethics": {
|
22 |
+
"acc": 0.28,
|
23 |
+
"acc_stderr": 0.04512608598542127,
|
24 |
+
"acc_norm": 0.28,
|
25 |
+
"acc_norm_stderr": 0.04512608598542127
|
26 |
+
},
|
27 |
+
"hendrycksTest-clinical_knowledge": {
|
28 |
+
"acc": 0.3018867924528302,
|
29 |
+
"acc_stderr": 0.028254200344438655,
|
30 |
+
"acc_norm": 0.3018867924528302,
|
31 |
+
"acc_norm_stderr": 0.028254200344438655
|
32 |
+
},
|
33 |
+
"hendrycksTest-college_biology": {
|
34 |
+
"acc": 0.2916666666666667,
|
35 |
+
"acc_stderr": 0.03800968060554857,
|
36 |
+
"acc_norm": 0.2916666666666667,
|
37 |
+
"acc_norm_stderr": 0.03800968060554857
|
38 |
+
},
|
39 |
+
"hendrycksTest-college_chemistry": {
|
40 |
+
"acc": 0.26,
|
41 |
+
"acc_stderr": 0.0440844002276808,
|
42 |
+
"acc_norm": 0.26,
|
43 |
+
"acc_norm_stderr": 0.0440844002276808
|
44 |
+
},
|
45 |
+
"hendrycksTest-college_computer_science": {
|
46 |
+
"acc": 0.37,
|
47 |
+
"acc_stderr": 0.048523658709391,
|
48 |
+
"acc_norm": 0.37,
|
49 |
+
"acc_norm_stderr": 0.048523658709391
|
50 |
+
},
|
51 |
+
"hendrycksTest-college_mathematics": {
|
52 |
+
"acc": 0.27,
|
53 |
+
"acc_stderr": 0.044619604333847394,
|
54 |
+
"acc_norm": 0.27,
|
55 |
+
"acc_norm_stderr": 0.044619604333847394
|
56 |
+
},
|
57 |
+
"hendrycksTest-college_medicine": {
|
58 |
+
"acc": 0.2543352601156069,
|
59 |
+
"acc_stderr": 0.0332055644308557,
|
60 |
+
"acc_norm": 0.2543352601156069,
|
61 |
+
"acc_norm_stderr": 0.0332055644308557
|
62 |
+
},
|
63 |
+
"hendrycksTest-college_physics": {
|
64 |
+
"acc": 0.23529411764705882,
|
65 |
+
"acc_stderr": 0.04220773659171452,
|
66 |
+
"acc_norm": 0.23529411764705882,
|
67 |
+
"acc_norm_stderr": 0.04220773659171452
|
68 |
+
},
|
69 |
+
"hendrycksTest-computer_security": {
|
70 |
+
"acc": 0.32,
|
71 |
+
"acc_stderr": 0.046882617226215034,
|
72 |
+
"acc_norm": 0.32,
|
73 |
+
"acc_norm_stderr": 0.046882617226215034
|
74 |
+
},
|
75 |
+
"hendrycksTest-conceptual_physics": {
|
76 |
+
"acc": 0.31063829787234043,
|
77 |
+
"acc_stderr": 0.03025123757921317,
|
78 |
+
"acc_norm": 0.31063829787234043,
|
79 |
+
"acc_norm_stderr": 0.03025123757921317
|
80 |
+
},
|
81 |
+
"hendrycksTest-econometrics": {
|
82 |
+
"acc": 0.2894736842105263,
|
83 |
+
"acc_stderr": 0.04266339443159394,
|
84 |
+
"acc_norm": 0.2894736842105263,
|
85 |
+
"acc_norm_stderr": 0.04266339443159394
|
86 |
+
},
|
87 |
+
"hendrycksTest-electrical_engineering": {
|
88 |
+
"acc": 0.3103448275862069,
|
89 |
+
"acc_stderr": 0.038552896163789485,
|
90 |
+
"acc_norm": 0.3103448275862069,
|
91 |
+
"acc_norm_stderr": 0.038552896163789485
|
92 |
+
},
|
93 |
+
"hendrycksTest-elementary_mathematics": {
|
94 |
+
"acc": 0.2804232804232804,
|
95 |
+
"acc_stderr": 0.023135287974325628,
|
96 |
+
"acc_norm": 0.2804232804232804,
|
97 |
+
"acc_norm_stderr": 0.023135287974325628
|
98 |
+
},
|
99 |
+
"hendrycksTest-formal_logic": {
|
100 |
+
"acc": 0.18253968253968253,
|
101 |
+
"acc_stderr": 0.03455071019102146,
|
102 |
+
"acc_norm": 0.18253968253968253,
|
103 |
+
"acc_norm_stderr": 0.03455071019102146
|
104 |
+
},
|
105 |
+
"hendrycksTest-global_facts": {
|
106 |
+
"acc": 0.33,
|
107 |
+
"acc_stderr": 0.047258156262526045,
|
108 |
+
"acc_norm": 0.33,
|
109 |
+
"acc_norm_stderr": 0.047258156262526045
|
110 |
+
},
|
111 |
+
"hendrycksTest-high_school_biology": {
|
112 |
+
"acc": 0.31290322580645163,
|
113 |
+
"acc_stderr": 0.02637756702864586,
|
114 |
+
"acc_norm": 0.31290322580645163,
|
115 |
+
"acc_norm_stderr": 0.02637756702864586
|
116 |
+
},
|
117 |
+
"hendrycksTest-high_school_chemistry": {
|
118 |
+
"acc": 0.2512315270935961,
|
119 |
+
"acc_stderr": 0.030516530732694433,
|
120 |
+
"acc_norm": 0.2512315270935961,
|
121 |
+
"acc_norm_stderr": 0.030516530732694433
|
122 |
+
},
|
123 |
+
"hendrycksTest-high_school_computer_science": {
|
124 |
+
"acc": 0.3,
|
125 |
+
"acc_stderr": 0.046056618647183814,
|
126 |
+
"acc_norm": 0.3,
|
127 |
+
"acc_norm_stderr": 0.046056618647183814
|
128 |
+
},
|
129 |
+
"hendrycksTest-high_school_european_history": {
|
130 |
+
"acc": 0.3090909090909091,
|
131 |
+
"acc_stderr": 0.036085410115739666,
|
132 |
+
"acc_norm": 0.3090909090909091,
|
133 |
+
"acc_norm_stderr": 0.036085410115739666
|
134 |
+
},
|
135 |
+
"hendrycksTest-high_school_geography": {
|
136 |
+
"acc": 0.32323232323232326,
|
137 |
+
"acc_stderr": 0.033322999210706444,
|
138 |
+
"acc_norm": 0.32323232323232326,
|
139 |
+
"acc_norm_stderr": 0.033322999210706444
|
140 |
+
},
|
141 |
+
"hendrycksTest-high_school_government_and_politics": {
|
142 |
+
"acc": 0.32124352331606215,
|
143 |
+
"acc_stderr": 0.033699508685490674,
|
144 |
+
"acc_norm": 0.32124352331606215,
|
145 |
+
"acc_norm_stderr": 0.033699508685490674
|
146 |
+
},
|
147 |
+
"hendrycksTest-high_school_macroeconomics": {
|
148 |
+
"acc": 0.3076923076923077,
|
149 |
+
"acc_stderr": 0.023400928918310502,
|
150 |
+
"acc_norm": 0.3076923076923077,
|
151 |
+
"acc_norm_stderr": 0.023400928918310502
|
152 |
+
},
|
153 |
+
"hendrycksTest-high_school_mathematics": {
|
154 |
+
"acc": 0.25925925925925924,
|
155 |
+
"acc_stderr": 0.026719240783712177,
|
156 |
+
"acc_norm": 0.25925925925925924,
|
157 |
+
"acc_norm_stderr": 0.026719240783712177
|
158 |
+
},
|
159 |
+
"hendrycksTest-high_school_microeconomics": {
|
160 |
+
"acc": 0.25630252100840334,
|
161 |
+
"acc_stderr": 0.02835962087053395,
|
162 |
+
"acc_norm": 0.25630252100840334,
|
163 |
+
"acc_norm_stderr": 0.02835962087053395
|
164 |
+
},
|
165 |
+
"hendrycksTest-high_school_physics": {
|
166 |
+
"acc": 0.271523178807947,
|
167 |
+
"acc_stderr": 0.036313298039696545,
|
168 |
+
"acc_norm": 0.271523178807947,
|
169 |
+
"acc_norm_stderr": 0.036313298039696545
|
170 |
+
},
|
171 |
+
"hendrycksTest-high_school_psychology": {
|
172 |
+
"acc": 0.27339449541284405,
|
173 |
+
"acc_stderr": 0.01910929984609828,
|
174 |
+
"acc_norm": 0.27339449541284405,
|
175 |
+
"acc_norm_stderr": 0.01910929984609828
|
176 |
+
},
|
177 |
+
"hendrycksTest-high_school_statistics": {
|
178 |
+
"acc": 0.21296296296296297,
|
179 |
+
"acc_stderr": 0.027920963147993666,
|
180 |
+
"acc_norm": 0.21296296296296297,
|
181 |
+
"acc_norm_stderr": 0.027920963147993666
|
182 |
+
},
|
183 |
+
"hendrycksTest-high_school_us_history": {
|
184 |
+
"acc": 0.25,
|
185 |
+
"acc_stderr": 0.03039153369274154,
|
186 |
+
"acc_norm": 0.25,
|
187 |
+
"acc_norm_stderr": 0.03039153369274154
|
188 |
+
},
|
189 |
+
"hendrycksTest-high_school_world_history": {
|
190 |
+
"acc": 0.2742616033755274,
|
191 |
+
"acc_stderr": 0.029041333510598025,
|
192 |
+
"acc_norm": 0.2742616033755274,
|
193 |
+
"acc_norm_stderr": 0.029041333510598025
|
194 |
+
},
|
195 |
+
"hendrycksTest-human_aging": {
|
196 |
+
"acc": 0.36771300448430494,
|
197 |
+
"acc_stderr": 0.03236198350928276,
|
198 |
+
"acc_norm": 0.36771300448430494,
|
199 |
+
"acc_norm_stderr": 0.03236198350928276
|
200 |
+
},
|
201 |
+
"hendrycksTest-human_sexuality": {
|
202 |
+
"acc": 0.33587786259541985,
|
203 |
+
"acc_stderr": 0.04142313771996665,
|
204 |
+
"acc_norm": 0.33587786259541985,
|
205 |
+
"acc_norm_stderr": 0.04142313771996665
|
206 |
+
},
|
207 |
+
"hendrycksTest-international_law": {
|
208 |
+
"acc": 0.4214876033057851,
|
209 |
+
"acc_stderr": 0.045077322787750944,
|
210 |
+
"acc_norm": 0.4214876033057851,
|
211 |
+
"acc_norm_stderr": 0.045077322787750944
|
212 |
+
},
|
213 |
+
"hendrycksTest-jurisprudence": {
|
214 |
+
"acc": 0.3333333333333333,
|
215 |
+
"acc_stderr": 0.04557239513497752,
|
216 |
+
"acc_norm": 0.3333333333333333,
|
217 |
+
"acc_norm_stderr": 0.04557239513497752
|
218 |
+
},
|
219 |
+
"hendrycksTest-logical_fallacies": {
|
220 |
+
"acc": 0.27607361963190186,
|
221 |
+
"acc_stderr": 0.0351238528370505,
|
222 |
+
"acc_norm": 0.27607361963190186,
|
223 |
+
"acc_norm_stderr": 0.0351238528370505
|
224 |
+
},
|
225 |
+
"hendrycksTest-machine_learning": {
|
226 |
+
"acc": 0.35714285714285715,
|
227 |
+
"acc_stderr": 0.04547960999764376,
|
228 |
+
"acc_norm": 0.35714285714285715,
|
229 |
+
"acc_norm_stderr": 0.04547960999764376
|
230 |
+
},
|
231 |
+
"hendrycksTest-management": {
|
232 |
+
"acc": 0.30097087378640774,
|
233 |
+
"acc_stderr": 0.045416094465039476,
|
234 |
+
"acc_norm": 0.30097087378640774,
|
235 |
+
"acc_norm_stderr": 0.045416094465039476
|
236 |
+
},
|
237 |
+
"hendrycksTest-marketing": {
|
238 |
+
"acc": 0.2863247863247863,
|
239 |
+
"acc_stderr": 0.029614323690456648,
|
240 |
+
"acc_norm": 0.2863247863247863,
|
241 |
+
"acc_norm_stderr": 0.029614323690456648
|
242 |
+
},
|
243 |
+
"hendrycksTest-medical_genetics": {
|
244 |
+
"acc": 0.28,
|
245 |
+
"acc_stderr": 0.045126085985421255,
|
246 |
+
"acc_norm": 0.28,
|
247 |
+
"acc_norm_stderr": 0.045126085985421255
|
248 |
+
},
|
249 |
+
"hendrycksTest-miscellaneous": {
|
250 |
+
"acc": 0.30395913154533843,
|
251 |
+
"acc_stderr": 0.016448321686769043,
|
252 |
+
"acc_norm": 0.30395913154533843,
|
253 |
+
"acc_norm_stderr": 0.016448321686769043
|
254 |
+
},
|
255 |
+
"hendrycksTest-moral_disputes": {
|
256 |
+
"acc": 0.2832369942196532,
|
257 |
+
"acc_stderr": 0.02425790170532337,
|
258 |
+
"acc_norm": 0.2832369942196532,
|
259 |
+
"acc_norm_stderr": 0.02425790170532337
|
260 |
+
},
|
261 |
+
"hendrycksTest-moral_scenarios": {
|
262 |
+
"acc": 0.2424581005586592,
|
263 |
+
"acc_stderr": 0.014333522059217889,
|
264 |
+
"acc_norm": 0.2424581005586592,
|
265 |
+
"acc_norm_stderr": 0.014333522059217889
|
266 |
+
},
|
267 |
+
"hendrycksTest-nutrition": {
|
268 |
+
"acc": 0.2908496732026144,
|
269 |
+
"acc_stderr": 0.026004800363952113,
|
270 |
+
"acc_norm": 0.2908496732026144,
|
271 |
+
"acc_norm_stderr": 0.026004800363952113
|
272 |
+
},
|
273 |
+
"hendrycksTest-philosophy": {
|
274 |
+
"acc": 0.28938906752411575,
|
275 |
+
"acc_stderr": 0.02575586592263294,
|
276 |
+
"acc_norm": 0.28938906752411575,
|
277 |
+
"acc_norm_stderr": 0.02575586592263294
|
278 |
+
},
|
279 |
+
"hendrycksTest-prehistory": {
|
280 |
+
"acc": 0.2808641975308642,
|
281 |
+
"acc_stderr": 0.025006469755799204,
|
282 |
+
"acc_norm": 0.2808641975308642,
|
283 |
+
"acc_norm_stderr": 0.025006469755799204
|
284 |
+
},
|
285 |
+
"hendrycksTest-professional_accounting": {
|
286 |
+
"acc": 0.2872340425531915,
|
287 |
+
"acc_stderr": 0.026992199173064356,
|
288 |
+
"acc_norm": 0.2872340425531915,
|
289 |
+
"acc_norm_stderr": 0.026992199173064356
|
290 |
+
},
|
291 |
+
"hendrycksTest-professional_law": {
|
292 |
+
"acc": 0.26401564537157757,
|
293 |
+
"acc_stderr": 0.011258435537723821,
|
294 |
+
"acc_norm": 0.26401564537157757,
|
295 |
+
"acc_norm_stderr": 0.011258435537723821
|
296 |
+
},
|
297 |
+
"hendrycksTest-professional_medicine": {
|
298 |
+
"acc": 0.20588235294117646,
|
299 |
+
"acc_stderr": 0.024562204314142317,
|
300 |
+
"acc_norm": 0.20588235294117646,
|
301 |
+
"acc_norm_stderr": 0.024562204314142317
|
302 |
+
},
|
303 |
+
"hendrycksTest-professional_psychology": {
|
304 |
+
"acc": 0.2826797385620915,
|
305 |
+
"acc_stderr": 0.018217269552053432,
|
306 |
+
"acc_norm": 0.2826797385620915,
|
307 |
+
"acc_norm_stderr": 0.018217269552053432
|
308 |
+
},
|
309 |
+
"hendrycksTest-public_relations": {
|
310 |
+
"acc": 0.32727272727272727,
|
311 |
+
"acc_stderr": 0.04494290866252089,
|
312 |
+
"acc_norm": 0.32727272727272727,
|
313 |
+
"acc_norm_stderr": 0.04494290866252089
|
314 |
+
},
|
315 |
+
"hendrycksTest-security_studies": {
|
316 |
+
"acc": 0.3020408163265306,
|
317 |
+
"acc_stderr": 0.02939360931987981,
|
318 |
+
"acc_norm": 0.3020408163265306,
|
319 |
+
"acc_norm_stderr": 0.02939360931987981
|
320 |
+
},
|
321 |
+
"hendrycksTest-sociology": {
|
322 |
+
"acc": 0.27860696517412936,
|
323 |
+
"acc_stderr": 0.03170056183497308,
|
324 |
+
"acc_norm": 0.27860696517412936,
|
325 |
+
"acc_norm_stderr": 0.03170056183497308
|
326 |
+
},
|
327 |
+
"hendrycksTest-us_foreign_policy": {
|
328 |
+
"acc": 0.31,
|
329 |
+
"acc_stderr": 0.04648231987117316,
|
330 |
+
"acc_norm": 0.31,
|
331 |
+
"acc_norm_stderr": 0.04648231987117316
|
332 |
+
},
|
333 |
+
"hendrycksTest-virology": {
|
334 |
+
"acc": 0.3313253012048193,
|
335 |
+
"acc_stderr": 0.036643147772880864,
|
336 |
+
"acc_norm": 0.3313253012048193,
|
337 |
+
"acc_norm_stderr": 0.036643147772880864
|
338 |
+
},
|
339 |
+
"hendrycksTest-world_religions": {
|
340 |
+
"acc": 0.2573099415204678,
|
341 |
+
"acc_stderr": 0.03352799844161865,
|
342 |
+
"acc_norm": 0.2573099415204678,
|
343 |
+
"acc_norm_stderr": 0.03352799844161865
|
344 |
+
}
|
345 |
+
},
|
346 |
+
"versions": {
|
347 |
+
"hendrycksTest-abstract_algebra": 1,
|
348 |
+
"hendrycksTest-anatomy": 1,
|
349 |
+
"hendrycksTest-astronomy": 1,
|
350 |
+
"hendrycksTest-business_ethics": 1,
|
351 |
+
"hendrycksTest-clinical_knowledge": 1,
|
352 |
+
"hendrycksTest-college_biology": 1,
|
353 |
+
"hendrycksTest-college_chemistry": 1,
|
354 |
+
"hendrycksTest-college_computer_science": 1,
|
355 |
+
"hendrycksTest-college_mathematics": 1,
|
356 |
+
"hendrycksTest-college_medicine": 1,
|
357 |
+
"hendrycksTest-college_physics": 1,
|
358 |
+
"hendrycksTest-computer_security": 1,
|
359 |
+
"hendrycksTest-conceptual_physics": 1,
|
360 |
+
"hendrycksTest-econometrics": 1,
|
361 |
+
"hendrycksTest-electrical_engineering": 1,
|
362 |
+
"hendrycksTest-elementary_mathematics": 1,
|
363 |
+
"hendrycksTest-formal_logic": 1,
|
364 |
+
"hendrycksTest-global_facts": 1,
|
365 |
+
"hendrycksTest-high_school_biology": 1,
|
366 |
+
"hendrycksTest-high_school_chemistry": 1,
|
367 |
+
"hendrycksTest-high_school_computer_science": 1,
|
368 |
+
"hendrycksTest-high_school_european_history": 1,
|
369 |
+
"hendrycksTest-high_school_geography": 1,
|
370 |
+
"hendrycksTest-high_school_government_and_politics": 1,
|
371 |
+
"hendrycksTest-high_school_macroeconomics": 1,
|
372 |
+
"hendrycksTest-high_school_mathematics": 1,
|
373 |
+
"hendrycksTest-high_school_microeconomics": 1,
|
374 |
+
"hendrycksTest-high_school_physics": 1,
|
375 |
+
"hendrycksTest-high_school_psychology": 1,
|
376 |
+
"hendrycksTest-high_school_statistics": 1,
|
377 |
+
"hendrycksTest-high_school_us_history": 1,
|
378 |
+
"hendrycksTest-high_school_world_history": 1,
|
379 |
+
"hendrycksTest-human_aging": 1,
|
380 |
+
"hendrycksTest-human_sexuality": 1,
|
381 |
+
"hendrycksTest-international_law": 1,
|
382 |
+
"hendrycksTest-jurisprudence": 1,
|
383 |
+
"hendrycksTest-logical_fallacies": 1,
|
384 |
+
"hendrycksTest-machine_learning": 1,
|
385 |
+
"hendrycksTest-management": 1,
|
386 |
+
"hendrycksTest-marketing": 1,
|
387 |
+
"hendrycksTest-medical_genetics": 1,
|
388 |
+
"hendrycksTest-miscellaneous": 1,
|
389 |
+
"hendrycksTest-moral_disputes": 1,
|
390 |
+
"hendrycksTest-moral_scenarios": 1,
|
391 |
+
"hendrycksTest-nutrition": 1,
|
392 |
+
"hendrycksTest-philosophy": 1,
|
393 |
+
"hendrycksTest-prehistory": 1,
|
394 |
+
"hendrycksTest-professional_accounting": 1,
|
395 |
+
"hendrycksTest-professional_law": 1,
|
396 |
+
"hendrycksTest-professional_medicine": 1,
|
397 |
+
"hendrycksTest-professional_psychology": 1,
|
398 |
+
"hendrycksTest-public_relations": 1,
|
399 |
+
"hendrycksTest-security_studies": 1,
|
400 |
+
"hendrycksTest-sociology": 1,
|
401 |
+
"hendrycksTest-us_foreign_policy": 1,
|
402 |
+
"hendrycksTest-virology": 1,
|
403 |
+
"hendrycksTest-world_religions": 1
|
404 |
+
},
|
405 |
+
"config": {
|
406 |
+
"model": "hf-causal-experimental",
|
407 |
+
"model_args": "pretrained=/home/vmagent/app/data/mpt-7b-delta-tune-model-div-sal/merged_model,use_accelerate=True,trust_remote_code=True,dtype=float16",
|
408 |
+
"num_fewshot": 5,
|
409 |
+
"batch_size": "auto",
|
410 |
+
"batch_sizes": [
|
411 |
+
16
|
412 |
+
],
|
413 |
+
"device": null,
|
414 |
+
"no_cache": false,
|
415 |
+
"limit": null,
|
416 |
+
"bootstrap_iters": 100000,
|
417 |
+
"description_dict": {}
|
418 |
+
}
|
419 |
+
}
|
mpt-7b-delta-tune-model-div-sal-truthqa
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"results": {
|
3 |
+
"truthfulqa_mc": {
|
4 |
+
"mc1": 0.22643818849449204,
|
5 |
+
"mc1_stderr": 0.01465133732460258,
|
6 |
+
"mc2": 0.3448308279815954,
|
7 |
+
"mc2_stderr": 0.0135712184614352
|
8 |
+
}
|
9 |
+
},
|
10 |
+
"versions": {
|
11 |
+
"truthfulqa_mc": 1
|
12 |
+
},
|
13 |
+
"config": {
|
14 |
+
"model": "hf-causal-experimental",
|
15 |
+
"model_args": "pretrained=/home/vmagent/app/data/mpt-7b-delta-tune-model-div-sal/merged_model,use_accelerate=True,trust_remote_code=True,dtype=float16",
|
16 |
+
"num_fewshot": 0,
|
17 |
+
"batch_size": "auto",
|
18 |
+
"batch_sizes": [
|
19 |
+
32
|
20 |
+
],
|
21 |
+
"device": null,
|
22 |
+
"no_cache": false,
|
23 |
+
"limit": null,
|
24 |
+
"bootstrap_iters": 100000,
|
25 |
+
"description_dict": {}
|
26 |
+
}
|
27 |
+
}
|