llama-3-8b-base-new-dpo-hh-helpful-s_star1.0-4xh200-batch-64-20260421-233802/README.md

library_name, base_model, tags, datasets, model-index

library_name	base_model	tags	datasets	model-index
transformers	W-61/llama-3-8b-base-sft-hh-helpful-4xh200	alignment-handbook new-dpo generated_from_trainer	Anthropic/hh-rlhf	name results llama-3-8b-base-new-dpo-hh-helpful-s_star1.0-4xh200-batch-64-20260421-233802

llama-3-8b-base-new-dpo-hh-helpful-s_star1.0-4xh200-batch-64-20260421-233802

This model is a fine-tuned version of W-61/llama-3-8b-base-sft-hh-helpful-4xh200 on the Anthropic/hh-rlhf dataset. It achieves the following results on the evaluation set:

• Loss: 0.4980
• Fcm Dpo/beta: 0.0078
• Fcm Dpo/q T: 0.3364
• Fcm Dpo/delta: 0.0593
• Fcm Dpo/margin: 120.6791
• Margin Dpo/margin Mean: 120.6791
• Margin Dpo/margin Std: 173.7378
• Logps/chosen: -446.5822
• Logps/rejected: -575.0082
• Logps/ref Chosen: -79.0510
• Logps/ref Rejected: -86.7979
• Logits/chosen: 0.1581
• Logits/rejected: 0.1793

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-07
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Fcm Dpo/beta	Fcm Dpo/q T	Fcm Dpo/delta	Fcm Dpo/margin	Margin Dpo/margin Mean	Margin Dpo/margin Std	Logps/chosen	Logps/rejected	Logps/ref Chosen	Logps/ref Rejected	Logits/chosen	Logits/rejected
0.8238	0.2937	200	0.5818	0.0270	0.3663	0.0598	34.2048	34.2048	64.2965	-136.1259	-178.0776	-79.0510	-86.7979	-0.5095	-0.4883
0.767	0.5874	400	0.5232	0.0099	0.3436	0.0761	94.0497	94.0496	145.6569	-305.7622	-407.5587	-79.0510	-86.7979	0.0068	0.0285
0.8367	0.8811	600	0.4980	0.0078	0.3364	0.0593	120.6791	120.6791	173.7378	-446.5822	-575.0082	-79.0510	-86.7979	0.1581	0.1793

Framework versions

• Transformers 4.51.0
• Pytorch 2.3.1+cu121
• Datasets 2.21.0
• Tokenizers 0.21.4