Model save

README.md

@@ -1,10 +1,11 @@
 ---
 library_name: transformers
-license: llama3
-base_model: meta-llama/Meta-Llama-3-8B
+license: gemma
+base_model: google/gemma-2-9b-it
 tags:
+- trl
+- dpo
 - llama-factory
-- full
 - generated_from_trainer
 model-index:
 - name: model
@@ -16,9 +17,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # model
 
-This model is a fine-tuned version of [meta-llama/Meta-Llama-3-8B](https://huggingface.co/meta-llama/Meta-Llama-3-8B) on the PrincetonPLI/Instruct-SkillMix-SDD dataset.
+This model is a fine-tuned version of [google/gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: nan
+- Loss: 3.9434
+- Rewards/chosen: -46.0543
+- Rewards/rejected: -47.7041
+- Rewards/accuracies: 0.6473
+- Rewards/margins: 1.6497
+- Logps/rejected: -4.7704
+- Logps/chosen: -4.6054
+- Logits/rejected: 14.6796
+- Logits/chosen: 14.4459
 
 ## Model description
 
@@ -37,28 +46,25 @@ More information needed
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
-- learning_rate: 5e-06
-- train_batch_size: 8
-- eval_batch_size: 4
+- learning_rate: 8e-07
+- train_batch_size: 2
+- eval_batch_size: 2
 - seed: 42
 - distributed_type: multi-GPU
-- num_devices: 16
-- gradient_accumulation_steps: 4
-- total_train_batch_size: 512
+- num_devices: 32
+- gradient_accumulation_steps: 2
+- total_train_batch_size: 128
 - total_eval_batch_size: 64
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
-- lr_scheduler_type: constant
+- lr_scheduler_type: cosine
 - lr_scheduler_warmup_ratio: 0.1
-- lr_scheduler_warmup_steps: 1738
-- num_epochs: 3.0
+- num_epochs: 1.0
 
 ### Training results
 
-| Training Loss | Epoch  | Step | Validation Loss |
-|:-------------:|:------:|:----:|:---------------:|
-| 0.0           | 0.8    | 3    | nan             |
-| 0.0           | 1.8667 | 7    | nan             |
-| 0.0           | 2.4    | 9    | nan             |
+| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
+|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
+| 3.9873        | 1.0   | 7344 | 3.9434          | -46.0543       | -47.7041         | 0.6473             | 1.6497          | -4.7704        | -4.6054      | 14.6796         | 14.4459       |
 
 
 ### Framework versions
@@ -66,4 +72,4 @@ The following hyperparameters were used during training:
 - Transformers 4.45.2
 - Pytorch 2.3.0
 - Datasets 2.21.0
-- Tokenizers 0.20.1
+- Tokenizers 0.20.2

generation_config.json

@@ -1,9 +1,8 @@
 {
-  "bos_token_id": 128000,
-  "do_sample": true,
-  "eos_token_id": 128001,
-  "max_length": 4096,
-  "temperature": 0.6,
-  "top_p": 0.9,
+  "_from_model_config": true,
+  "bos_token_id": 2,
+  "cache_implementation": "hybrid",
+  "eos_token_id": 1,
+  "pad_token_id": 0,
   "transformers_version": "4.45.2"
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c93984dd997223ed18ba512f4d557717a1c163b126e84956eefd9678fd16474
+oid sha256:94816a34230f886273cbcd5b38d15b23e891a375b3a1d387d32261e87b3200d7
 size 7096