Something wrong in the decoding process

[sigma-alpha-beta]

I use model inference canny task，I changed line 90 in the generate.py with “images=[Image.open('person.jpg')]”，when I run the sample.sh, I got the error:

Something wrong in the decoding process. MultiTokensVLLogitsProcessor. st: id (tensor([1], device='cuda:0'),) num 1 ed: id (tensor([4008, 8936], device='cuda:0'),) num 2 input_ids.shape torch.Size([1, 8941]) scores.shape torch.Size([1, 1, 171385])

the sample.sh is:

# for image pair generation, you need set the task --> depth, canny, hed, openpose
task="canny"
cuda_number=1
results_dir=samples/
mkdir -p ${results_dir}

CUDA_VISIBLE_DEVICES=${cuda_number} python generate_examples/generate.py \
--model_path Alpha-VLLM/Lumina-mGPT-2.0 \
--save_path ${results_dir} --cfg 4.0 --top_k 4096 --temperature 1.0 --width 768 --height 768 --speculative_jacobi --task ${task}

thanks

[synbol]

We released the inference code for image-to-image tasks. Please try it.