Minbyul commited on
Commit
258753f
1 Parent(s): fe9c437

End of training

Browse files
Files changed (4) hide show
  1. README.md +16 -10
  2. all_results.json +13 -0
  3. config.json +1 -1
  4. eval_results.json +16 -0
README.md CHANGED
@@ -1,9 +1,15 @@
1
  ---
2
  base_model: Minbyul/llama2-7b-wo-medication_qa-sft
3
  tags:
 
4
  - trl
5
  - dpo
6
  - generated_from_trainer
 
 
 
 
 
7
  model-index:
8
  - name: llama2-7b-dpo-full-sft-wo-medication_qa
9
  results: []
@@ -14,17 +20,17 @@ should probably proofread and complete it, then remove this comment. -->
14
 
15
  # llama2-7b-dpo-full-sft-wo-medication_qa
16
 
17
- This model is a fine-tuned version of [Minbyul/llama2-7b-wo-medication_qa-sft](https://huggingface.co/Minbyul/llama2-7b-wo-medication_qa-sft) on an unknown dataset.
18
  It achieves the following results on the evaluation set:
19
- - Logits/chosen: -0.4292
20
- - Logits/rejected: -0.4708
21
- - Logps/chosen: -476.1255
22
- - Logps/rejected: -635.9033
23
- - Loss: 0.4682
24
- - Rewards/accuracies: 0.9250
25
- - Rewards/chosen: -0.1609
26
- - Rewards/margins: 0.9415
27
- - Rewards/rejected: -1.1024
28
 
29
  ## Model description
30
 
 
1
  ---
2
  base_model: Minbyul/llama2-7b-wo-medication_qa-sft
3
  tags:
4
+ - alignment-handbook
5
  - trl
6
  - dpo
7
  - generated_from_trainer
8
+ - trl
9
+ - dpo
10
+ - generated_from_trainer
11
+ datasets:
12
+ - HuggingFaceH4/ultrafeedback_binarized
13
  model-index:
14
  - name: llama2-7b-dpo-full-sft-wo-medication_qa
15
  results: []
 
20
 
21
  # llama2-7b-dpo-full-sft-wo-medication_qa
22
 
23
+ This model is a fine-tuned version of [Minbyul/llama2-7b-wo-medication_qa-sft](https://huggingface.co/Minbyul/llama2-7b-wo-medication_qa-sft) on the HuggingFaceH4/ultrafeedback_binarized dataset.
24
  It achieves the following results on the evaluation set:
25
+ - Loss: 0.4396
26
+ - Rewards/chosen: -0.1779
27
+ - Rewards/rejected: -1.2468
28
+ - Rewards/accuracies: 0.9500
29
+ - Rewards/margins: 1.0689
30
+ - Logps/rejected: -650.3414
31
+ - Logps/chosen: -477.8221
32
+ - Logits/rejected: -0.4720
33
+ - Logits/chosen: -0.4277
34
 
35
  ## Model description
36
 
all_results.json CHANGED
@@ -1,5 +1,18 @@
1
  {
2
  "epoch": 1.0,
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  "train_loss": 0.05898992068894947,
4
  "train_runtime": 471.0672,
5
  "train_samples": 8410,
 
1
  {
2
  "epoch": 1.0,
3
+ "eval_logits/chosen": -0.42766323685646057,
4
+ "eval_logits/rejected": -0.4719906747341156,
5
+ "eval_logps/chosen": -477.8221130371094,
6
+ "eval_logps/rejected": -650.3414306640625,
7
+ "eval_loss": 0.4395619034767151,
8
+ "eval_rewards/accuracies": 0.949999988079071,
9
+ "eval_rewards/chosen": -0.17786642909049988,
10
+ "eval_rewards/margins": 1.0689051151275635,
11
+ "eval_rewards/rejected": -1.2467713356018066,
12
+ "eval_runtime": 37.8151,
13
+ "eval_samples": 320,
14
+ "eval_samples_per_second": 8.462,
15
+ "eval_steps_per_second": 0.264,
16
  "train_loss": 0.05898992068894947,
17
  "train_runtime": 471.0672,
18
  "train_samples": 8410,
config.json CHANGED
@@ -23,6 +23,6 @@
23
  "tie_word_embeddings": false,
24
  "torch_dtype": "bfloat16",
25
  "transformers_version": "4.39.0.dev0",
26
- "use_cache": false,
27
  "vocab_size": 32000
28
  }
 
23
  "tie_word_embeddings": false,
24
  "torch_dtype": "bfloat16",
25
  "transformers_version": "4.39.0.dev0",
26
+ "use_cache": true,
27
  "vocab_size": 32000
28
  }
eval_results.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "eval_logits/chosen": -0.42766323685646057,
4
+ "eval_logits/rejected": -0.4719906747341156,
5
+ "eval_logps/chosen": -477.8221130371094,
6
+ "eval_logps/rejected": -650.3414306640625,
7
+ "eval_loss": 0.4395619034767151,
8
+ "eval_rewards/accuracies": 0.949999988079071,
9
+ "eval_rewards/chosen": -0.17786642909049988,
10
+ "eval_rewards/margins": 1.0689051151275635,
11
+ "eval_rewards/rejected": -1.2467713356018066,
12
+ "eval_runtime": 37.8151,
13
+ "eval_samples": 320,
14
+ "eval_samples_per_second": 8.462,
15
+ "eval_steps_per_second": 0.264
16
+ }