Qwen2.5-VL-3B-Instruct-ft

Native Qwen2.5-VL-format checkpoint converted from patrickamadeus/qwen2_5vl-1000.

This repository is loadable with the standard Qwen2.5-VL / Transformers API. It does not require the training repository's custom Qwen2_5VL, DistillPrefixVLM, or eval wrappers.

Install

pip install -U transformers accelerate qwen-vl-utils

Load

from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration

model_id = "patrickamadeus/Qwen2.5-VL-3B-Instruct-ft"

model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)
processor = AutoProcessor.from_pretrained(model_id)

Text-only Inference

messages = [
    {
        "role": "user",
        "content": [
            {"type": "text", "text": "Explain what this model is useful for in one sentence."},
        ],
    }
]

text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=[text], padding=True, return_tensors="pt").to(model.device)

generated_ids = model.generate(**inputs, max_new_tokens=64)
generated_ids = [
    output_ids[len(input_ids):]
    for input_ids, output_ids in zip(inputs.input_ids, generated_ids)
]
response = processor.batch_decode(
    generated_ids,
    skip_special_tokens=True,
    clean_up_tokenization_spaces=False,
)[0]
print(response)

Expected output: a short natural-language answer, for example a one-sentence description of the model's use.

Image + Text Inference

from qwen_vl_utils import process_vision_info

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {"type": "text", "text": "Describe this image."},
        ],
    }
]

text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
).to(model.device)

generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids = [
    output_ids[len(input_ids):]
    for input_ids, output_ids in zip(inputs.input_ids, generated_ids)
]
response = processor.batch_decode(
    generated_ids,
    skip_special_tokens=True,
    clean_up_tokenization_spaces=False,
)[0]
print(response)

Expected output: a concise image description, typically mentioning the major objects and scene.

Source

• Base model: Qwen/Qwen2.5-VL-3B-Instruct
• Converted checkpoint: patrickamadeus/qwen2_5vl-1000