Commit
•
e563b5a
1
Parent(s):
394b5d5
End of training
Browse files- README.md +8 -1
- all_results.json +13 -0
- config.json +1 -1
- eval_results.json +16 -0
README.md
CHANGED
@@ -2,9 +2,16 @@
|
|
2 |
license: mit
|
3 |
base_model: BramVanroy/fietje-2b-sft
|
4 |
tags:
|
|
|
5 |
- trl
|
6 |
- dpo
|
7 |
- generated_from_trainer
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
model-index:
|
9 |
- name: fietje-2b-dpo-lr2.0e-6-beta0.2-gradaccum2-v6
|
10 |
results: []
|
@@ -15,7 +22,7 @@ should probably proofread and complete it, then remove this comment. -->
|
|
15 |
|
16 |
# fietje-2b-dpo-lr2.0e-6-beta0.2-gradaccum2-v6
|
17 |
|
18 |
-
This model is a fine-tuned version of [BramVanroy/fietje-2b-sft](https://huggingface.co/BramVanroy/fietje-2b-sft) on the
|
19 |
It achieves the following results on the evaluation set:
|
20 |
- Loss: 0.2842
|
21 |
- Rewards/chosen: -1.1549
|
|
|
2 |
license: mit
|
3 |
base_model: BramVanroy/fietje-2b-sft
|
4 |
tags:
|
5 |
+
- alignment-handbook
|
6 |
- trl
|
7 |
- dpo
|
8 |
- generated_from_trainer
|
9 |
+
- trl
|
10 |
+
- dpo
|
11 |
+
- generated_from_trainer
|
12 |
+
datasets:
|
13 |
+
- BramVanroy/ultra_feedback_dutch_cleaned
|
14 |
+
- BramVanroy/orca_dpo_pairs_dutch_cleaned
|
15 |
model-index:
|
16 |
- name: fietje-2b-dpo-lr2.0e-6-beta0.2-gradaccum2-v6
|
17 |
results: []
|
|
|
22 |
|
23 |
# fietje-2b-dpo-lr2.0e-6-beta0.2-gradaccum2-v6
|
24 |
|
25 |
+
This model is a fine-tuned version of [BramVanroy/fietje-2b-sft](https://huggingface.co/BramVanroy/fietje-2b-sft) on the BramVanroy/ultra_feedback_dutch_cleaned and the BramVanroy/orca_dpo_pairs_dutch_cleaned datasets.
|
26 |
It achieves the following results on the evaluation set:
|
27 |
- Loss: 0.2842
|
28 |
- Rewards/chosen: -1.1549
|
all_results.json
CHANGED
@@ -1,5 +1,18 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
"train_loss": 0.3634731802597361,
|
4 |
"train_runtime": 16591.2723,
|
5 |
"train_samples": 18653,
|
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
+
"eval_logits/chosen": -1.352773666381836,
|
4 |
+
"eval_logits/rejected": -1.2867770195007324,
|
5 |
+
"eval_logps/chosen": -451.33642578125,
|
6 |
+
"eval_logps/rejected": -657.6813354492188,
|
7 |
+
"eval_loss": 0.2841675579547882,
|
8 |
+
"eval_rewards/accuracies": 0.88671875,
|
9 |
+
"eval_rewards/chosen": -1.1548646688461304,
|
10 |
+
"eval_rewards/margins": 2.4814703464508057,
|
11 |
+
"eval_rewards/rejected": -3.6363346576690674,
|
12 |
+
"eval_runtime": 527.3224,
|
13 |
+
"eval_samples": 1536,
|
14 |
+
"eval_samples_per_second": 2.913,
|
15 |
+
"eval_steps_per_second": 0.728,
|
16 |
"train_loss": 0.3634731802597361,
|
17 |
"train_runtime": 16591.2723,
|
18 |
"train_samples": 18653,
|
config.json
CHANGED
@@ -29,6 +29,6 @@
|
|
29 |
"tie_word_embeddings": false,
|
30 |
"torch_dtype": "bfloat16",
|
31 |
"transformers_version": "4.39.1",
|
32 |
-
"use_cache":
|
33 |
"vocab_size": 50297
|
34 |
}
|
|
|
29 |
"tie_word_embeddings": false,
|
30 |
"torch_dtype": "bfloat16",
|
31 |
"transformers_version": "4.39.1",
|
32 |
+
"use_cache": true,
|
33 |
"vocab_size": 50297
|
34 |
}
|
eval_results.json
ADDED
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 1.0,
|
3 |
+
"eval_logits/chosen": -1.352773666381836,
|
4 |
+
"eval_logits/rejected": -1.2867770195007324,
|
5 |
+
"eval_logps/chosen": -451.33642578125,
|
6 |
+
"eval_logps/rejected": -657.6813354492188,
|
7 |
+
"eval_loss": 0.2841675579547882,
|
8 |
+
"eval_rewards/accuracies": 0.88671875,
|
9 |
+
"eval_rewards/chosen": -1.1548646688461304,
|
10 |
+
"eval_rewards/margins": 2.4814703464508057,
|
11 |
+
"eval_rewards/rejected": -3.6363346576690674,
|
12 |
+
"eval_runtime": 527.3224,
|
13 |
+
"eval_samples": 1536,
|
14 |
+
"eval_samples_per_second": 2.913,
|
15 |
+
"eval_steps_per_second": 0.728
|
16 |
+
}
|