End of training
Browse files- README.md +17 -1
- all_results.json +16 -0
- config.json +29 -0
- eval_results.json +19 -0
- runs/Jul15_00-31-06_allegro-adams/events.out.tfevents.1721034884.allegro-adams.981573.1 +3 -0
README.md
CHANGED
@@ -1,8 +1,11 @@
|
|
1 |
---
|
2 |
base_model: deepseek-ai/deepseek-math-7b-base
|
|
|
|
|
3 |
library_name: peft
|
4 |
license: other
|
5 |
tags:
|
|
|
6 |
- trl
|
7 |
- orpo
|
8 |
- generated_from_trainer
|
@@ -17,7 +20,20 @@ should probably proofread and complete it, then remove this comment. -->
|
|
17 |
[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/thuzfz1/huggingface/runs/zs68qfsi)
|
18 |
# deepseek-8b-orpo-lora
|
19 |
|
20 |
-
This model is a fine-tuned version of [deepseek-ai/deepseek-math-7b-base](https://huggingface.co/deepseek-ai/deepseek-math-7b-base) on
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
## Model description
|
23 |
|
|
|
1 |
---
|
2 |
base_model: deepseek-ai/deepseek-math-7b-base
|
3 |
+
datasets:
|
4 |
+
- zfz1/my_preference_gsm8k_deepseek
|
5 |
library_name: peft
|
6 |
license: other
|
7 |
tags:
|
8 |
+
- alignment-handbook
|
9 |
- trl
|
10 |
- orpo
|
11 |
- generated_from_trainer
|
|
|
20 |
[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/thuzfz1/huggingface/runs/zs68qfsi)
|
21 |
# deepseek-8b-orpo-lora
|
22 |
|
23 |
+
This model is a fine-tuned version of [deepseek-ai/deepseek-math-7b-base](https://huggingface.co/deepseek-ai/deepseek-math-7b-base) on the zfz1/my_preference_gsm8k_deepseek dataset.
|
24 |
+
It achieves the following results on the evaluation set:
|
25 |
+
- Loss: 1.1456
|
26 |
+
- Rewards/chosen: -0.0726
|
27 |
+
- Rewards/rejected: -0.0757
|
28 |
+
- Rewards/accuracies: 0.5647
|
29 |
+
- Rewards/margins: 0.0031
|
30 |
+
- Logps/rejected: -0.7569
|
31 |
+
- Logps/chosen: -0.7262
|
32 |
+
- Logits/rejected: 30.3288
|
33 |
+
- Logits/chosen: 30.4427
|
34 |
+
- Nll Loss: 1.0719
|
35 |
+
- Log Odds Ratio: -0.6909
|
36 |
+
- Log Odds Chosen: 0.0553
|
37 |
|
38 |
## Model description
|
39 |
|
all_results.json
CHANGED
@@ -1,5 +1,21 @@
|
|
1 |
{
|
2 |
"epoch": 1.9936102236421727,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
"total_flos": 0.0,
|
4 |
"train_loss": 1.3676127867820935,
|
5 |
"train_runtime": 5524.9246,
|
|
|
1 |
{
|
2 |
"epoch": 1.9936102236421727,
|
3 |
+
"eval_log_odds_chosen": 0.05526316165924072,
|
4 |
+
"eval_log_odds_ratio": -0.6908599138259888,
|
5 |
+
"eval_logits/chosen": 30.44268798828125,
|
6 |
+
"eval_logits/rejected": 30.32880210876465,
|
7 |
+
"eval_logps/chosen": -0.7262152433395386,
|
8 |
+
"eval_logps/rejected": -0.7568790316581726,
|
9 |
+
"eval_loss": 1.1456295251846313,
|
10 |
+
"eval_nll_loss": 1.071852684020996,
|
11 |
+
"eval_rewards/accuracies": 0.5647059082984924,
|
12 |
+
"eval_rewards/chosen": -0.07262152433395386,
|
13 |
+
"eval_rewards/margins": 0.0030663711950182915,
|
14 |
+
"eval_rewards/rejected": -0.07568789273500443,
|
15 |
+
"eval_runtime": 247.9026,
|
16 |
+
"eval_samples": 5398,
|
17 |
+
"eval_samples_per_second": 21.775,
|
18 |
+
"eval_steps_per_second": 0.343,
|
19 |
"total_flos": 0.0,
|
20 |
"train_loss": 1.3676127867820935,
|
21 |
"train_runtime": 5524.9246,
|
config.json
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "deepseek-ai/deepseek-math-7b-base",
|
3 |
+
"architectures": [
|
4 |
+
"LlamaForCausalLM"
|
5 |
+
],
|
6 |
+
"attention_bias": false,
|
7 |
+
"attention_dropout": 0.0,
|
8 |
+
"bos_token_id": 100000,
|
9 |
+
"eos_token_id": 100001,
|
10 |
+
"hidden_act": "silu",
|
11 |
+
"hidden_size": 4096,
|
12 |
+
"initializer_range": 0.02,
|
13 |
+
"intermediate_size": 11008,
|
14 |
+
"max_position_embeddings": 4096,
|
15 |
+
"mlp_bias": false,
|
16 |
+
"model_type": "llama",
|
17 |
+
"num_attention_heads": 32,
|
18 |
+
"num_hidden_layers": 30,
|
19 |
+
"num_key_value_heads": 32,
|
20 |
+
"pretraining_tp": 1,
|
21 |
+
"rms_norm_eps": 1e-06,
|
22 |
+
"rope_scaling": null,
|
23 |
+
"rope_theta": 10000.0,
|
24 |
+
"tie_word_embeddings": false,
|
25 |
+
"torch_dtype": "bfloat16",
|
26 |
+
"transformers_version": "4.42.3",
|
27 |
+
"use_cache": true,
|
28 |
+
"vocab_size": 102400
|
29 |
+
}
|
eval_results.json
ADDED
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 1.9936102236421727,
|
3 |
+
"eval_log_odds_chosen": 0.05526316165924072,
|
4 |
+
"eval_log_odds_ratio": -0.6908599138259888,
|
5 |
+
"eval_logits/chosen": 30.44268798828125,
|
6 |
+
"eval_logits/rejected": 30.32880210876465,
|
7 |
+
"eval_logps/chosen": -0.7262152433395386,
|
8 |
+
"eval_logps/rejected": -0.7568790316581726,
|
9 |
+
"eval_loss": 1.1456295251846313,
|
10 |
+
"eval_nll_loss": 1.071852684020996,
|
11 |
+
"eval_rewards/accuracies": 0.5647059082984924,
|
12 |
+
"eval_rewards/chosen": -0.07262152433395386,
|
13 |
+
"eval_rewards/margins": 0.0030663711950182915,
|
14 |
+
"eval_rewards/rejected": -0.07568789273500443,
|
15 |
+
"eval_runtime": 247.9026,
|
16 |
+
"eval_samples": 5398,
|
17 |
+
"eval_samples_per_second": 21.775,
|
18 |
+
"eval_steps_per_second": 0.343
|
19 |
+
}
|
runs/Jul15_00-31-06_allegro-adams/events.out.tfevents.1721034884.allegro-adams.981573.1
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:54e3e42cdc1f7ea818293b9e9e14b4747b631016f5eece9e9794686da41b1730
|
3 |
+
size 997
|