End of training
Browse files- README.md +3 -2
- all_results.json +22 -0
- eval_results.json +17 -0
- train_results.json +8 -0
- trainer_state.json +0 -0
- training_eval_loss.png +0 -0
- training_loss.png +0 -0
- training_rewards_accuracies.png +0 -0
- training_sft_loss.png +0 -0
README.md
CHANGED
@@ -2,9 +2,10 @@
|
|
2 |
license: gemma
|
3 |
library_name: peft
|
4 |
tags:
|
|
|
|
|
5 |
- trl
|
6 |
- dpo
|
7 |
-
- llama-factory
|
8 |
- generated_from_trainer
|
9 |
base_model: google/gemma-2b-it
|
10 |
model-index:
|
@@ -17,7 +18,7 @@ should probably proofread and complete it, then remove this comment. -->
|
|
17 |
|
18 |
# Gemma-2B-It-ORPO-SALT
|
19 |
|
20 |
-
This model is a fine-tuned version of [google/gemma-2b-it](https://huggingface.co/google/gemma-2b-it) on the
|
21 |
It achieves the following results on the evaluation set:
|
22 |
- Loss: 1.3294
|
23 |
- Rewards/chosen: -0.1257
|
|
|
2 |
license: gemma
|
3 |
library_name: peft
|
4 |
tags:
|
5 |
+
- llama-factory
|
6 |
+
- lora
|
7 |
- trl
|
8 |
- dpo
|
|
|
9 |
- generated_from_trainer
|
10 |
base_model: google/gemma-2b-it
|
11 |
model-index:
|
|
|
18 |
|
19 |
# Gemma-2B-It-ORPO-SALT
|
20 |
|
21 |
+
This model is a fine-tuned version of [google/gemma-2b-it](https://huggingface.co/google/gemma-2b-it) on the dpo_mix_en and the bct_non_cot_dpo_1000 datasets.
|
22 |
It achieves the following results on the evaluation set:
|
23 |
- Loss: 1.3294
|
24 |
- Rewards/chosen: -0.1257
|
all_results.json
ADDED
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 2.9969690846635686,
|
3 |
+
"eval_logits/chosen": -20.66438865661621,
|
4 |
+
"eval_logits/rejected": -20.695892333984375,
|
5 |
+
"eval_logps/chosen": -1.2573115825653076,
|
6 |
+
"eval_logps/rejected": -1.3996058702468872,
|
7 |
+
"eval_loss": 1.3293771743774414,
|
8 |
+
"eval_odds_ratio_loss": 0.7206552624702454,
|
9 |
+
"eval_rewards/accuracies": 0.5345454812049866,
|
10 |
+
"eval_rewards/chosen": -0.12573117017745972,
|
11 |
+
"eval_rewards/margins": 0.014229409396648407,
|
12 |
+
"eval_rewards/rejected": -0.13996057212352753,
|
13 |
+
"eval_runtime": 82.3315,
|
14 |
+
"eval_samples_per_second": 13.361,
|
15 |
+
"eval_sft_loss": 1.2573115825653076,
|
16 |
+
"eval_steps_per_second": 6.68,
|
17 |
+
"total_flos": 5.618252880760013e+17,
|
18 |
+
"train_loss": 1.3892074464594277,
|
19 |
+
"train_runtime": 8722.5011,
|
20 |
+
"train_samples_per_second": 3.404,
|
21 |
+
"train_steps_per_second": 0.213
|
22 |
+
}
|
eval_results.json
ADDED
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 2.9969690846635686,
|
3 |
+
"eval_logits/chosen": -20.66438865661621,
|
4 |
+
"eval_logits/rejected": -20.695892333984375,
|
5 |
+
"eval_logps/chosen": -1.2573115825653076,
|
6 |
+
"eval_logps/rejected": -1.3996058702468872,
|
7 |
+
"eval_loss": 1.3293771743774414,
|
8 |
+
"eval_odds_ratio_loss": 0.7206552624702454,
|
9 |
+
"eval_rewards/accuracies": 0.5345454812049866,
|
10 |
+
"eval_rewards/chosen": -0.12573117017745972,
|
11 |
+
"eval_rewards/margins": 0.014229409396648407,
|
12 |
+
"eval_rewards/rejected": -0.13996057212352753,
|
13 |
+
"eval_runtime": 82.3315,
|
14 |
+
"eval_samples_per_second": 13.361,
|
15 |
+
"eval_sft_loss": 1.2573115825653076,
|
16 |
+
"eval_steps_per_second": 6.68
|
17 |
+
}
|
train_results.json
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 2.9969690846635686,
|
3 |
+
"total_flos": 5.618252880760013e+17,
|
4 |
+
"train_loss": 1.3892074464594277,
|
5 |
+
"train_runtime": 8722.5011,
|
6 |
+
"train_samples_per_second": 3.404,
|
7 |
+
"train_steps_per_second": 0.213
|
8 |
+
}
|
trainer_state.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
training_eval_loss.png
ADDED
training_loss.png
ADDED
training_rewards_accuracies.png
ADDED
training_sft_loss.png
ADDED