library_name, base_model, tags, datasets, model-index

library_name	base_model	tags	datasets	model-index
transformers	jackf857/qwen3-8b-base-sft-ultrachat-4xh200-batch-128	alignment-handbook simpo generated_from_trainer	HuggingFaceH4/ultrafeedback_binarized	name results qwen3-8b-base-simpo-ultrafeedback-4xh200-batch-128

qwen3-8b-base-simpo-ultrafeedback-4xh200-batch-128

This model is a fine-tuned version of jackf857/qwen3-8b-base-sft-ultrachat-4xh200-batch-128 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 1.0770
• Rewards/chosen: -2.1095
• Rewards/rejected: -2.9493
• Rewards/accuracies: 0.6660
• Rewards/margins: 0.8398
• Logps/rejected: -1.4746
• Logps/chosen: -1.0548
• Logits/rejected: 2.1566
• Logits/chosen: 2.1386

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: 6e-07
• train_batch_size: 4
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• total_eval_batch_size: 8
• 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	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
9.1298	0.4188	200	1.1165	-2.0802	-2.7440	0.6360	0.6638	-1.3720	-1.0401	2.1249	2.1120
8.8992	0.8377	400	1.0770	-2.1095	-2.9493	0.6660	0.8398	-1.4746	-1.0548	2.1566	2.1386

Framework versions

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