Qwen2.5-VL-7B-Instruct

Qwen2.5-VL-7B-Instruct

Mirror of Qwen's Qwen2.5-VL 7B instruction-tuned vision-language model, in Transformers safetensors form.

What it is for

This repo is a mirror of the instruction-tuned 7B Qwen2.5-VL model from the Qwen team. The model reads images, video and text, and writes text. It holds the sharded safetensors weights with the tokenizer, chat template, processor config and generation config. Use it as a pinned copy of the upstream weights for local inference and for conversion.

Provenance

• Upstream model: Qwen/Qwen2.5-VL-7B-Instruct.
• Upstream code: QwenLM/Qwen2.5-VL. Blog: Qwen2.5-VL.

Use

from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor

MODEL_ID = "chibifire/Qwen2.5-VL-7B-Instruct"
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(MODEL_ID, torch_dtype="auto", device_map="auto")
processor = AutoProcessor.from_pretrained(MODEL_ID)

For image and video input, qwen_vl_utils and full examples, read the upstream model card.

Licence

Apache 2.0, the upstream licence. Terms: https://www.apache.org/licenses/LICENSE-2.0. These terms travel with the weights. Keep this link when you copy or redistribute them.