Views
No views yet
1seed: 3407
2model:
3 name: unsloth/Qwen3-VL-4B-Instruct
4 max_seq_length: 2048
5 load_in_4bit: true
6 fast_inference: false
7 use_gradient_checkpointing: unsloth
8peft:
9 lora_rank: 16
10 lora_alpha: 16
11 lora_dropout: 0
12 finetune_vision_layers: true
13 finetune_language_layers: true
14 finetune_attention_modules: true
15 finetune_mlp_modules: true
16 use_rslora: false
17data:
18 dataset_name: unsloth/LaTeX_OCR
19 split: train
20 instruction: Write the LaTeX representation for this image.
21 image_field: image
22 text_field: text
23sft:
24 per_device_train_batch_size: 1
25 gradient_accumulation_steps: 4
26 max_steps: 1000
27 learning_rate: 0.0002
28 warmup_steps: 5
29 optim: adamw_8bit
30 weight_decay: 0.001
31 lr_scheduler_type: linear
32eval:
33 enabled: true
34 dataset_name: unsloth/LaTeX_OCR
35 split: test
36 max_samples: 200
37 max_new_tokens: 128
38 temperature: 0.0
39 top_p: 1.0
40 do_sample: false| source | num_samples | exact_match | edit_distance |
|---|---|---|---|
| eval_results.json | 200 | 0.3300 | 15.8550 |
| rank | checkpoint | exact_match | edit_distance | note |
|---|---|---|---|---|
| 1 | checkpoint-200 | 0.3500 | 15.8350 | - |
| 2 | checkpoint-100 | 0.3450 | 16.7500 | - |
| 3 | checkpoint-400 | 0.3300 | 15.8550 | - |