Commit
•
59f4671
1
Parent(s):
13be604
End of training
Browse files- README.md +17 -1
- all_results.json +13 -0
- eval_results.json +12 -12
README.md
CHANGED
@@ -2,10 +2,16 @@
|
|
2 |
license: mit
|
3 |
library_name: peft
|
4 |
tags:
|
|
|
|
|
|
|
|
|
5 |
- trl
|
6 |
- dpo
|
7 |
- generated_from_trainer
|
8 |
base_model: HuggingFaceH4/mistral-7b-sft-beta
|
|
|
|
|
9 |
model-index:
|
10 |
- name: mistral-sft-7b-dpo-qlora
|
11 |
results: []
|
@@ -16,7 +22,17 @@ should probably proofread and complete it, then remove this comment. -->
|
|
16 |
|
17 |
# mistral-sft-7b-dpo-qlora
|
18 |
|
19 |
-
This model is a fine-tuned version of [HuggingFaceH4/mistral-7b-sft-beta](https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta) on the
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
## Model description
|
22 |
|
|
|
2 |
license: mit
|
3 |
library_name: peft
|
4 |
tags:
|
5 |
+
- alignment-handbook
|
6 |
+
- trl
|
7 |
+
- dpo
|
8 |
+
- generated_from_trainer
|
9 |
- trl
|
10 |
- dpo
|
11 |
- generated_from_trainer
|
12 |
base_model: HuggingFaceH4/mistral-7b-sft-beta
|
13 |
+
datasets:
|
14 |
+
- HuggingFaceH4/ultrafeedback_binarized
|
15 |
model-index:
|
16 |
- name: mistral-sft-7b-dpo-qlora
|
17 |
results: []
|
|
|
22 |
|
23 |
# mistral-sft-7b-dpo-qlora
|
24 |
|
25 |
+
This model is a fine-tuned version of [HuggingFaceH4/mistral-7b-sft-beta](https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta) on the HuggingFaceH4/ultrafeedback_binarized dataset.
|
26 |
+
It achieves the following results on the evaluation set:
|
27 |
+
- Loss: 0.6932
|
28 |
+
- Rewards/chosen: 0.0023
|
29 |
+
- Rewards/rejected: 0.0013
|
30 |
+
- Rewards/accuracies: 0.5625
|
31 |
+
- Rewards/margins: 0.0009
|
32 |
+
- Logps/rejected: -122.8616
|
33 |
+
- Logps/chosen: -86.2666
|
34 |
+
- Logits/rejected: -3.0452
|
35 |
+
- Logits/chosen: -2.9819
|
36 |
|
37 |
## Model description
|
38 |
|
all_results.json
CHANGED
@@ -1,5 +1,18 @@
|
|
1 |
{
|
2 |
"epoch": 5.0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
"train_loss": 0.34364445209503175,
|
4 |
"train_runtime": 77.9014,
|
5 |
"train_samples": 10,
|
|
|
1 |
{
|
2 |
"epoch": 5.0,
|
3 |
+
"eval_logits/chosen": -2.9818739891052246,
|
4 |
+
"eval_logits/rejected": -3.045213222503662,
|
5 |
+
"eval_logps/chosen": -86.26663208007812,
|
6 |
+
"eval_logps/rejected": -122.861572265625,
|
7 |
+
"eval_loss": 0.6932176947593689,
|
8 |
+
"eval_rewards/accuracies": 0.5625,
|
9 |
+
"eval_rewards/chosen": 0.0022567124105989933,
|
10 |
+
"eval_rewards/margins": 0.000948949484154582,
|
11 |
+
"eval_rewards/rejected": 0.0013077628100290895,
|
12 |
+
"eval_runtime": 11.9466,
|
13 |
+
"eval_samples": 20,
|
14 |
+
"eval_samples_per_second": 1.674,
|
15 |
+
"eval_steps_per_second": 0.167,
|
16 |
"train_loss": 0.34364445209503175,
|
17 |
"train_runtime": 77.9014,
|
18 |
"train_samples": 10,
|
eval_results.json
CHANGED
@@ -1,16 +1,16 @@
|
|
1 |
{
|
2 |
-
"epoch":
|
3 |
-
"eval_logits/chosen": -2.
|
4 |
-
"eval_logits/rejected": -3.
|
5 |
-
"eval_logps/chosen": -86.
|
6 |
-
"eval_logps/rejected": -122.
|
7 |
-
"eval_loss": 0.
|
8 |
"eval_rewards/accuracies": 0.5625,
|
9 |
-
"eval_rewards/chosen": 0.
|
10 |
-
"eval_rewards/margins":
|
11 |
-
"eval_rewards/rejected": 0.
|
12 |
-
"eval_runtime": 11.
|
13 |
"eval_samples": 20,
|
14 |
-
"eval_samples_per_second": 1.
|
15 |
-
"eval_steps_per_second": 0.
|
16 |
}
|
|
|
1 |
{
|
2 |
+
"epoch": 5.0,
|
3 |
+
"eval_logits/chosen": -2.9818739891052246,
|
4 |
+
"eval_logits/rejected": -3.045213222503662,
|
5 |
+
"eval_logps/chosen": -86.26663208007812,
|
6 |
+
"eval_logps/rejected": -122.861572265625,
|
7 |
+
"eval_loss": 0.6932176947593689,
|
8 |
"eval_rewards/accuracies": 0.5625,
|
9 |
+
"eval_rewards/chosen": 0.0022567124105989933,
|
10 |
+
"eval_rewards/margins": 0.000948949484154582,
|
11 |
+
"eval_rewards/rejected": 0.0013077628100290895,
|
12 |
+
"eval_runtime": 11.9466,
|
13 |
"eval_samples": 20,
|
14 |
+
"eval_samples_per_second": 1.674,
|
15 |
+
"eval_steps_per_second": 0.167
|
16 |
}
|