End of training
Browse files
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/z4q90sz1)
|
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/z4q90sz1)
|
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: 0.6818
|
26 |
+
- Rewards/chosen: -0.0338
|
27 |
+
- Rewards/rejected: -0.0840
|
28 |
+
- Rewards/accuracies: 0.8088
|
29 |
+
- Rewards/margins: 0.0502
|
30 |
+
- Logps/rejected: -0.8398
|
31 |
+
- Logps/chosen: -0.3377
|
32 |
+
- Logits/rejected: 34.4233
|
33 |
+
- Logits/chosen: 35.5254
|
34 |
+
- Nll Loss: 0.6414
|
35 |
+
- Log Odds Ratio: -0.4212
|
36 |
+
- Log Odds Chosen: 1.0634
|
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": 0.7846963420892373,
|
5 |
"train_runtime": 5577.3844,
|
|
|
1 |
{
|
2 |
"epoch": 1.9936102236421727,
|
3 |
+
"eval_log_odds_chosen": 1.0634132623672485,
|
4 |
+
"eval_log_odds_ratio": -0.421150267124176,
|
5 |
+
"eval_logits/chosen": 35.52544021606445,
|
6 |
+
"eval_logits/rejected": 34.42332077026367,
|
7 |
+
"eval_logps/chosen": -0.3376733958721161,
|
8 |
+
"eval_logps/rejected": -0.8398498296737671,
|
9 |
+
"eval_loss": 0.6817505359649658,
|
10 |
+
"eval_nll_loss": 0.6413611173629761,
|
11 |
+
"eval_rewards/accuracies": 0.8088235259056091,
|
12 |
+
"eval_rewards/chosen": -0.033767346292734146,
|
13 |
+
"eval_rewards/margins": 0.0502176471054554,
|
14 |
+
"eval_rewards/rejected": -0.08398497849702835,
|
15 |
+
"eval_runtime": 252.7054,
|
16 |
+
"eval_samples": 5398,
|
17 |
+
"eval_samples_per_second": 21.361,
|
18 |
+
"eval_steps_per_second": 0.336,
|
19 |
"total_flos": 0.0,
|
20 |
"train_loss": 0.7846963420892373,
|
21 |
"train_runtime": 5577.3844,
|
eval_results.json
CHANGED
@@ -1,19 +1,19 @@
|
|
1 |
{
|
2 |
"epoch": 1.9936102236421727,
|
3 |
-
"eval_log_odds_chosen":
|
4 |
-
"eval_log_odds_ratio": -0.
|
5 |
-
"eval_logits/chosen":
|
6 |
-
"eval_logits/rejected":
|
7 |
-
"eval_logps/chosen": -0.
|
8 |
-
"eval_logps/rejected": -0.
|
9 |
-
"eval_loss":
|
10 |
-
"eval_nll_loss":
|
11 |
-
"eval_rewards/accuracies": 0.
|
12 |
-
"eval_rewards/chosen": -0.
|
13 |
-
"eval_rewards/margins": 0.
|
14 |
-
"eval_rewards/rejected": -0.
|
15 |
-
"eval_runtime":
|
16 |
"eval_samples": 5398,
|
17 |
-
"eval_samples_per_second": 21.
|
18 |
-
"eval_steps_per_second": 0.
|
19 |
}
|
|
|
1 |
{
|
2 |
"epoch": 1.9936102236421727,
|
3 |
+
"eval_log_odds_chosen": 1.0634132623672485,
|
4 |
+
"eval_log_odds_ratio": -0.421150267124176,
|
5 |
+
"eval_logits/chosen": 35.52544021606445,
|
6 |
+
"eval_logits/rejected": 34.42332077026367,
|
7 |
+
"eval_logps/chosen": -0.3376733958721161,
|
8 |
+
"eval_logps/rejected": -0.8398498296737671,
|
9 |
+
"eval_loss": 0.6817505359649658,
|
10 |
+
"eval_nll_loss": 0.6413611173629761,
|
11 |
+
"eval_rewards/accuracies": 0.8088235259056091,
|
12 |
+
"eval_rewards/chosen": -0.033767346292734146,
|
13 |
+
"eval_rewards/margins": 0.0502176471054554,
|
14 |
+
"eval_rewards/rejected": -0.08398497849702835,
|
15 |
+
"eval_runtime": 252.7054,
|
16 |
"eval_samples": 5398,
|
17 |
+
"eval_samples_per_second": 21.361,
|
18 |
+
"eval_steps_per_second": 0.336
|
19 |
}
|
runs/Jul18_00-59-35_phe108-jieyuzhao-01/events.out.tfevents.1721295716.phe108-jieyuzhao-01.335789.1
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:8d0eb6ffd43afd6681159b970e21b0b92660842bb289b7862d8b896f1abca581
|
3 |
+
size 997
|