Views
No views yet
| Parameter | Value |
|---|---|
| LoRA Rank | 16 |
| LoRA Alpha | 32 |
| Batch Size | 4 |
| Learning Rate | 0.0001 |
| Epochs | 3.0 |
| Gradient Accumulation | 4 |
1from vllm import LLM, SamplingParams
2
3llm = LLM(
4 model="v1v1d1/nayana-qwen3vl-4b-stage3-section",
5 gpu_memory_utilization=0.8,
6 max_model_len=8192,
7)
8
9sampling_params = SamplingParams(
10 temperature=0.0,
11 max_tokens=512,
12)
13
14# Example with image
15messages = [{
16 "role": "user",
17 "content": [
18 {"type": "image_url", "image_url": {"url": "path/to/image.jpg"}},
19 {"type": "text", "text": "Describe this image in detail."}
20 ]
21}]
22
23outputs = llm.chat(messages, sampling_params=sampling_params)
24print(outputs[0].outputs[0].text)1from transformers import AutoModelForVision2Seq, AutoProcessor
2from PIL import Image
3
4model = AutoModelForVision2Seq.from_pretrained(
5 "v1v1d1/nayana-qwen3vl-4b-stage3-section",
6 torch_dtype="auto",
7 device_map="auto"
8)
9processor = AutoProcessor.from_pretrained("v1v1d1/nayana-qwen3vl-4b-stage3-section")
10
11image = Image.open("path/to/image.jpg")
12messages = [
13 {"role": "user", "content": [
14 {"type": "image"},
15 {"type": "text", "text": "Describe this image in detail."}
16 ]}
17]
18
19text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
20inputs = processor(text=[text], images=[image], return_tensors="pt").to(model.device)
21
22outputs = model.generate(**inputs, max_new_tokens=512)
23print(processor.decode(outputs[0], skip_special_tokens=True))1@misc{v1v1d1-nayana-qwen3vl-4b-stage3-section},
2 author = {Nayana Project},
3 title = {nayana-qwen3vl-4b-stage3-section},
4 year = {2026},
5 publisher = {HuggingFace},
6 howpublished = {https://huggingface.co/v1v1d1/nayana-qwen3vl-4b-stage3-section}
7}