ValueError: Unrecognized image processor in Qwen/Qwen2.5-VL-3B-Instruct.

[SkalskiP]

System Info

• transformers 4.49.0.dev0
• Python 3.11.11

Reproduction

I follow model instructions from here.

1. install transformers from GH

pip install git+https://github.com/huggingface/transformers

2. laod Qwen/Qwen2.5-VL-3B-Instruct or Qwen/Qwen2.5-VL-7B-Instruct; I did not tested with others

from transformers import AutoProcessor

processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-3B-Instruct")

3. it raise exception

ValueError: Unrecognized image processor in Qwen/Qwen2.5-VL-3B-Instruct. Should have a `image_processor_type` key in its preprocessor_config.json of config.json, or one of the following `model_type` keys in its config.json: align, aria, beit, bit, blip, blip-2, bridgetower, chameleon, chinese_clip, clip, clipseg, conditional_detr, convnext, convnextv2, cvt, data2vec-vision, deformable_detr, deit, depth_anything, depth_pro, deta, detr, dinat, dinov2, donut-swin, dpt, efficientformer, efficientnet, flava, focalnet, fuyu, git, glpn, got_ocr2, grounding-dino, groupvit, hiera, idefics, idefics2, idefics3, ijepa, imagegpt, instructblip, instructblipvideo, kosmos-2, layoutlmv2, layoutlmv3, levit, llava, llava_next, llava_next_video, llava_onevision, mask2former, maskformer, mgp-str, mllama, mobilenet_v1, mobilenet_v2, mobilevit, mobilevitv2, nat, nougat, oneformer, owlv2, owlvit, paligemma, perceiver, pix2struct, pixtral, poolformer, pvt, pvt_v2, qwen2_5_vl, qwen2_vl, regnet, resnet, rt_detr, sam, segformer, seggpt, siglip, superglue, swiftformer, swin, swin2sr, swinv2, table-transformer, timesformer, timm_wrapper, tvlt, tvp, udop, upernet, van, videomae, vilt, vipllava, vit, vit_hybrid, vit_mae, vit_msn, vitmatte, xclip, yolos, zoedepth

Expected behavior

processor loads without exceptions

[SkalskiP]

When installing pip install git+https://github.com/huggingface/transformers.git@1931a351408dbd1d0e2c4d6d7ee0eb5e8807d7bf AutoProcessor works. Looks like 33d1d71 might have broke it.

[JasonChenJC]

same problem

[Zzsf11]

Delete "image_processor_type": "Qwen2_5_VLImageProcessor" in Qwen2.5-VL-7B-Instruct/preprocessor_config.json could work.

[hhhharold]

update preprocessor_config.json, which is updated 2 days ago on huggingface

[zucchini-nlp]

Hey all! Thanks for reporting this, I didn't see that the slow image processor was removed!

@ArthurZucker I believe the slow ones should stay for a while, at least until the fast ones become the default. The issue is fixed now by updating the hub files and forcing the model to use Qwen2VL slow processor (models have identical image processing)

[shubham-atlys]

I am facing this error while using unsloth for finetuning. Please post the solution in case anyone has got it.

model, tokenizer = FastVisionModel.from_pretrained(
    "Qwen/Qwen2.5-VL-3B-Instruct",
    load_in_4bit = True, # Use 4bit to reduce memory use. False for 16bit LoRA.
    use_gradient_checkpointing = "unsloth", # True or "unsloth" for long context
)

[ArthurZucker]

IMO no need for the slow, let's make sure this does not break for people as release is soon. The processor was introduced in the main branch only