Cooolder/SCOPE-Direct-v2

sft_direct

This model is a fine-tuned version of Qwen/Qwen3-4B-Instruct-2507 on the scope_sft_direct dataset. It achieves the following results on the evaluation set:

• Loss: 0.3041

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: 1
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 8
• optimizer: Use 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: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.3289	0.0747	500	0.3394
0.3445	0.1494	1000	0.3315
0.3383	0.2241	1500	0.3268
0.3352	0.2987	2000	0.3224
0.3027	0.3734	2500	0.3197
0.3272	0.4481	3000	0.3166
0.3216	0.5228	3500	0.3138
0.3223	0.5975	4000	0.3108
0.3346	0.6722	4500	0.3076
0.3098	0.7468	5000	0.3065
0.2938	0.8215	5500	0.3059
0.315	0.8962	6000	0.3046
0.301	0.9709	6500	0.3040

Framework versions

• Transformers 4.57.1
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.22.1