This model is a fine-tuned version of W-61/qwen3-8b-base-sft-ultrachat-4xh200-batch-128 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

Loss: 0.5602
Margin Dpo/margin Mean: 48.7131
Margin Dpo/margin Std: 68.1546
Logps/chosen: -316.0414
Logps/rejected: -345.1451
Logps/ref Chosen: -281.4589
Logps/ref Rejected: -261.8495
Logits/chosen: 1.1933
Logits/rejected: 1.2367

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: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 8
total_train_batch_size: 128
total_eval_batch_size: 16
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	Margin Dpo/margin Mean	Margin Dpo/margin Std	Logps/chosen	Logps/rejected	Logps/ref Chosen	Logps/ref Rejected	Logits/chosen	Logits/rejected
4.9383	0.4188	200	0.5970	28.2584	39.0244	-287.0227	-295.6718	-281.4589	-261.8495	1.4300	1.4697
4.2739	0.8377	400	0.5602	48.7131	68.1546	-316.0414	-345.1451	-281.4589	-261.8495	1.1933	1.2367

Framework versions

Transformers 4.51.0
Pytorch 2.3.1+cu121
Datasets 2.21.0
Tokenizers 0.21.4

seanyhan/qwen3-8b-base-margin-dpo-ultrafeedback-4xh200-batch-128-20260423-040315

qwen3-8b-base-margin-dpo-ultrafeedback-4xh200-batch-128-20260423-040315

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

seanyhan/
qwen3-8b-base-margin-dpo-ultrafeedback-4xh200-batch-128-20260423-040315