Hugging Face's logo

yakazimir
/
qwen_cCPO_entropy

Text Generation
Transformers
Safetensors
qwen2
alignment-handbook
trl
simpo
Generated from Trainer
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
yakazimir commited on
Commit
c1af23d
·
verified ·
1 Parent(s): 13d5763

Model save

Browse files
Files changed (4) hide show
  1. README.md +26 -33
  2. all_results.json +4 -18
  3. train_results.json +4 -4
  4. trainer_state.json +0 -0
README.md CHANGED
@@ -3,15 +3,9 @@ library_name: transformers
3
  license: other
4
  base_model: trl-lib/qwen1.5-0.5b-sft
5
  tags:
6
- - alignment-handbook
7
  - trl
8
  - simpo
9
  - generated_from_trainer
10
- - trl
11
- - simpo
12
- - generated_from_trainer
13
- datasets:
14
- - yakazimir/ultrafeedback_binarized
15
  model-index:
16
  - name: qwen_cCPO_entropy
17
  results: []
@@ -22,18 +16,17 @@ should probably proofread and complete it, then remove this comment. -->
22
 
23
  # qwen_cCPO_entropy
24
 
25
- This model is a fine-tuned version of [trl-lib/qwen1.5-0.5b-sft](https://huggingface.co/trl-lib/qwen1.5-0.5b-sft) on the yakazimir/ultrafeedback_binarized dataset.
26
  It achieves the following results on the evaluation set:
27
- - Loss: 0.5209
28
- - Rewards/chosen: -9.5408
29
- - Rewards/rejected: -10.6523
30
- - Rewards/accuracies: 0.7166
31
- - Rewards/margins: 1.1115
32
- - Logps/rejected: -10.6523
33
- - Logps/chosen: -9.5408
34
- - Logits/rejected: 0.4143
35
- - Logits/chosen: 0.3542
36
- - Semantic Entropy: 0.0025
37
 
38
  ## Model description
39
 
@@ -66,22 +59,22 @@ The following hyperparameters were used during training:
66
 
67
  ### Training results
68
 
69
- | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen | Semantic Entropy |
70
- |:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|:----------------:|
71
- | 0.9273 | 0.2141 | 400 | 0.9132 | -1.5626 | -1.7709 | 0.5727 | 0.2083 | -1.7709 | -1.5626 | 0.3591 | 0.2698 | 0.6963 |
72
- | 0.6005 | 0.4282 | 800 | 0.5930 | -6.2110 | -6.7875 | 0.6714 | 0.5765 | -6.7875 | -6.2110 | 0.0273 | -0.0585 | 0.0266 |
73
- | 0.5591 | 0.6422 | 1200 | 0.5548 | -8.0914 | -8.7875 | 0.6936 | 0.6960 | -8.7875 | -8.0914 | 0.0627 | -0.0250 | 0.0053 |
74
- | 0.4839 | 0.8563 | 1600 | 0.5401 | -8.0351 | -8.7538 | 0.7092 | 0.7187 | -8.7538 | -8.0351 | 0.1712 | 0.0772 | 0.0055 |
75
- | 0.5873 | 1.0704 | 2000 | 0.5368 | -8.3578 | -9.1522 | 0.7114 | 0.7944 | -9.1522 | -8.3578 | 0.1990 | 0.1107 | 0.0039 |
76
- | 0.5089 | 1.2845 | 2400 | 0.5257 | -8.4730 | -9.4241 | 0.7211 | 0.9510 | -9.4241 | -8.4730 | 0.1623 | 0.0831 | 0.0047 |
77
- | 0.5109 | 1.4986 | 2800 | 0.5215 | -8.6140 | -9.5173 | 0.7263 | 0.9033 | -9.5173 | -8.6140 | 0.2671 | 0.1852 | 0.0039 |
78
- | 0.5508 | 1.7127 | 3200 | 0.5167 | -8.8251 | -9.7884 | 0.7277 | 0.9633 | -9.7884 | -8.8251 | 0.3885 | 0.3079 | 0.0035 |
79
- | 0.5684 | 1.9267 | 3600 | 0.5155 | -8.8490 | -9.8374 | 0.7240 | 0.9884 | -9.8374 | -8.8490 | 0.3563 | 0.2827 | 0.0032 |
80
- | 0.4288 | 2.1408 | 4000 | 0.5242 | -9.3271 | -10.4242 | 0.7211 | 1.0971 | -10.4242 | -9.3271 | 0.4923 | 0.4276 | 0.0026 |
81
- | 0.4556 | 2.3549 | 4400 | 0.5198 | -9.5864 | -10.6863 | 0.7203 | 1.0999 | -10.6863 | -9.5864 | 0.4365 | 0.3746 | 0.0024 |
82
- | 0.4621 | 2.5690 | 4800 | 0.5210 | -9.4722 | -10.5759 | 0.7196 | 1.1037 | -10.5759 | -9.4722 | 0.4038 | 0.3433 | 0.0026 |
83
- | 0.4387 | 2.7831 | 5200 | 0.5208 | -9.5239 | -10.6317 | 0.7181 | 1.1079 | -10.6317 | -9.5239 | 0.3872 | 0.3286 | 0.0025 |
84
- | 0.4041 | 2.9972 | 5600 | 0.5210 | -9.5408 | -10.6523 | 0.7166 | 1.1115 | -10.6523 | -9.5408 | 0.4143 | 0.3542 | 0.0025 |
85
 
86
 
87
  ### Framework versions
 
3
  license: other
4
  base_model: trl-lib/qwen1.5-0.5b-sft
5
  tags:
 
6
  - trl
7
  - simpo
8
  - generated_from_trainer
 
 
 
 
 
9
  model-index:
10
  - name: qwen_cCPO_entropy
11
  results: []
 
16
 
17
  # qwen_cCPO_entropy
18
 
19
+ This model is a fine-tuned version of [trl-lib/qwen1.5-0.5b-sft](https://huggingface.co/trl-lib/qwen1.5-0.5b-sft) on an unknown dataset.
20
  It achieves the following results on the evaluation set:
21
+ - Loss: 0.4698
22
+ - Rewards/chosen: -1.7544
23
+ - Rewards/rejected: -2.3724
24
+ - Rewards/accuracies: 0.6840
25
+ - Rewards/margins: 0.6180
26
+ - Logps/rejected: -2.3724
27
+ - Logps/chosen: -1.7544
28
+ - Logits/rejected: 0.2213
29
+ - Logits/chosen: 0.1180
 
30
 
31
  ## Model description
32
 
 
59
 
60
  ### Training results
61
 
62
+ | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
63
+ |:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
64
+ | 0.547 | 0.2141 | 400 | 0.5455 | -1.3437 | -1.4812 | 0.5579 | 0.1374 | -1.4812 | -1.3437 | 0.3541 | 0.2674 |
65
+ | 0.5301 | 0.4282 | 800 | 0.5165 | -1.3889 | -1.6415 | 0.5927 | 0.2526 | -1.6415 | -1.3889 | 0.4272 | 0.3345 |
66
+ | 0.5265 | 0.6422 | 1200 | 0.4985 | -1.4579 | -1.8204 | 0.6224 | 0.3625 | -1.8204 | -1.4579 | 0.3699 | 0.2760 |
67
+ | 0.4765 | 0.8563 | 1600 | 0.4935 | -1.4994 | -1.8829 | 0.6380 | 0.3836 | -1.8829 | -1.4994 | 0.3198 | 0.2257 |
68
+ | 0.5542 | 1.0704 | 2000 | 0.4872 | -1.4687 | -1.8582 | 0.6372 | 0.3895 | -1.8582 | -1.4687 | 0.3054 | 0.2090 |
69
+ | 0.4732 | 1.2845 | 2400 | 0.4775 | -1.6420 | -2.1625 | 0.6669 | 0.5204 | -2.1625 | -1.6420 | 0.3805 | 0.2752 |
70
+ | 0.5055 | 1.4986 | 2800 | 0.4755 | -1.6156 | -2.1129 | 0.6639 | 0.4973 | -2.1129 | -1.6156 | 0.4048 | 0.2981 |
71
+ | 0.4945 | 1.7127 | 3200 | 0.4738 | -1.5940 | -2.0956 | 0.6677 | 0.5016 | -2.0956 | -1.5940 | 0.3909 | 0.2834 |
72
+ | 0.4619 | 1.9267 | 3600 | 0.4700 | -1.6914 | -2.2530 | 0.6728 | 0.5617 | -2.2530 | -1.6914 | 0.3536 | 0.2473 |
73
+ | 0.4109 | 2.1408 | 4000 | 0.4699 | -1.7062 | -2.2883 | 0.6780 | 0.5822 | -2.2883 | -1.7062 | 0.3677 | 0.2556 |
74
+ | 0.4282 | 2.3549 | 4400 | 0.4707 | -1.7749 | -2.3952 | 0.6877 | 0.6202 | -2.3952 | -1.7749 | 0.2280 | 0.1239 |
75
+ | 0.4299 | 2.5690 | 4800 | 0.4704 | -1.7425 | -2.3507 | 0.6803 | 0.6082 | -2.3507 | -1.7425 | 0.3027 | 0.1929 |
76
+ | 0.4414 | 2.7831 | 5200 | 0.4698 | -1.7506 | -2.3686 | 0.6847 | 0.6181 | -2.3686 | -1.7506 | 0.2344 | 0.1302 |
77
+ | 0.404 | 2.9972 | 5600 | 0.4698 | -1.7544 | -2.3724 | 0.6840 | 0.6180 | -2.3724 | -1.7544 | 0.2213 | 0.1180 |
78
 
79
 
80
  ### Framework versions
all_results.json CHANGED
@@ -1,23 +1,9 @@
1
  {
2
  "epoch": 2.999297541394882,
3
- "eval_logits/chosen": 0.3541771471500397,
4
- "eval_logits/rejected": 0.4142882525920868,
5
- "eval_logps/chosen": -9.540809631347656,
6
- "eval_logps/rejected": -10.652300834655762,
7
- "eval_loss": 0.5209499597549438,
8
- "eval_rewards/accuracies": 0.716617226600647,
9
- "eval_rewards/chosen": -9.540809631347656,
10
- "eval_rewards/margins": 1.111490249633789,
11
- "eval_rewards/rejected": -10.652300834655762,
12
- "eval_runtime": 34.2971,
13
- "eval_samples": 1345,
14
- "eval_samples_per_second": 39.216,
15
- "eval_semantic_entropy": 0.0024765790440142155,
16
- "eval_steps_per_second": 9.826,
17
  "total_flos": 0.0,
18
- "train_loss": 0.5415812020936581,
19
- "train_runtime": 28601.8206,
20
  "train_samples": 59790,
21
- "train_samples_per_second": 6.271,
22
- "train_steps_per_second": 0.196
23
  }
 
1
  {
2
  "epoch": 2.999297541394882,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  "total_flos": 0.0,
4
+ "train_loss": 0.4793110236919071,
5
+ "train_runtime": 30370.8284,
6
  "train_samples": 59790,
7
+ "train_samples_per_second": 5.906,
8
+ "train_steps_per_second": 0.185
9
  }
train_results.json CHANGED
@@ -1,9 +1,9 @@
1
  {
2
  "epoch": 2.999297541394882,
3
  "total_flos": 0.0,
4
- "train_loss": 0.5415812020936581,
5
- "train_runtime": 28601.8206,
6
  "train_samples": 59790,
7
- "train_samples_per_second": 6.271,
8
- "train_steps_per_second": 0.196
9
  }
 
1
  {
2
  "epoch": 2.999297541394882,
3
  "total_flos": 0.0,
4
+ "train_loss": 0.4793110236919071,
5
+ "train_runtime": 30370.8284,
6
  "train_samples": 59790,
7
+ "train_samples_per_second": 5.906,
8
+ "train_steps_per_second": 0.185
9
  }
trainer_state.json CHANGED
The diff for this file is too large to render. See raw diff