qwen3-8b-base-new-dpo-hh-harmless-4xh200-batch-64-q_t-0.45-s_star-0.85/README.md

library_name, license, base_model, tags, datasets, model-index

library_name	license	base_model	tags	datasets	model-index
transformers	apache-2.0	jackf857/qwen3-8b-base-sft-hh-harmless-4xh200-batch-64-20260417-214452	alignment-handbook new-dpo generated_from_trainer	Anthropic/hh-rlhf	name results qwen3-8b-base-new-dpo-hh-harmless-4xh200-batch-64-q_t-0.45-s_star-0.85

qwen3-8b-base-new-dpo-hh-harmless-4xh200-batch-64-q_t-0.45-s_star-0.85

This model is a fine-tuned version of jackf857/qwen3-8b-base-sft-hh-harmless-4xh200-batch-64-20260417-214452 on the Anthropic/hh-rlhf dataset. It achieves the following results on the evaluation set:

• Loss: 0.5468
• Fcm Dpo/beta: 0.5703
• Margin Dpo/margin Mean: 1.5468
• Margin Dpo/margin Std: 2.5975
• Logps/chosen: -82.0159
• Logps/rejected: -93.3573
• Logps/ref Chosen: -86.9018
• Logps/ref Rejected: -96.6964
• Logits/chosen: 1.7134
• Logits/rejected: 1.6065

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	Margin Dpo/margin Mean	Margin Dpo/margin Std	Logps/chosen	Logps/rejected	Logps/ref Chosen	Logps/ref Rejected	Logits/chosen	Logits/rejected
1.325	0.1512	100	0.6541	0.1000	0.9036	1.9803	-86.1995	-96.8977	-86.9018	-96.6964	1.6835	1.5698
1.1881	0.3023	200	0.5625	0.6970	1.2537	2.2271	-78.6161	-89.6644	-86.9018	-96.6964	1.8627	1.7409
1.1394	0.4535	300	0.5554	0.6672	1.1965	2.1012	-79.8050	-90.7961	-86.9018	-96.6964	1.8760	1.7587
1.1945	0.6047	400	0.5504	0.6382	1.3377	2.2885	-81.0971	-92.2295	-86.9018	-96.6964	1.9273	1.8112
1.0818	0.7559	500	0.5445	0.5119	1.5736	2.6447	-81.9627	-93.3309	-86.9018	-96.6964	1.8607	1.7468
1.1017	0.9070	600	0.5468	0.5703	1.5468	2.5975	-82.0159	-93.3573	-86.9018	-96.6964	1.7134	1.6065

Framework versions

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