formalmathatepfl/qwen3-8b-sft-feedback

-- library_name: transformers license: other base_model: Qwen/Qwen3-8B tags: - llama-factory - full - generated_from_trainer model-index: - name: Qwen3-feedback results: []

Qwen3-feedback

This model is a fine-tuned version of Qwen/Qwen3-8B on the sft dataset. It achieves the following results on the evaluation set:

• Loss: 0.0228

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: 1e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 7
• total_train_batch_size: 14
• total_eval_batch_size: 14
• 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: 0.05
• num_epochs: 1.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.0367	0.0819	1000	0.0367
0.0274	0.1638	2000	0.0304
0.0257	0.2457	3000	0.0294
0.0242	0.3276	4000	0.0265
0.0239	0.4095	5000	0.0256
0.0235	0.4914	6000	0.0249
0.0233	0.5733	7000	0.0242
0.0228	0.6553	8000	0.0237
0.0219	0.7372	9000	0.0233
0.0234	0.8191	10000	0.0230
0.0229	0.9010	11000	0.0229
0.0229	0.9829	12000	0.0228

Framework versions

• Transformers 4.57.6
• Pytorch 2.9.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2