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:

Model description

More information needed

Intended uses & limitations

More information needed

More information needed

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 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