HuggingFaceTB

/

SmolVLM-Instruct

like 299

Follow

Hugging Face TB Research 751

Have you tried:

from transformers.image_utils import load_image
image1 = load_image("https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg")
?

I have. That works fine. But if I include a local directory like ./codespace/image1.jpg, the model does not see the image.

Anyone got it running with local images please post. Also I found it is handling only .jpg could not process .png, can anyone confirm this?

Here ya go...it's run a little different when processing a local file. Also, please note...

1. I opted to use the native prompt format because I like seeing it spelled out for some reason and don't like using "apply_chat_template".
2. I use a custom "set_cuda_paths" function at the top because I like pip installing these libraries rather than relying on a system-wide installation. If you use a system-wide installation (like most people do), simply remove this function.
3. I rely on a hardcoded path to the folder containing the model files rather than simply specifying the huggingface repo id because I like downloading the files first using snapshot_download where I can actually see the files rather than them being in my cache...adjust accordingly.

import sys
import os
from pathlib import Path

def set_cuda_paths():
    venv_base = Path(sys.executable).parent.parent
    nvidia_base_path = venv_base / 'Lib' / 'site-packages' / 'nvidia'
    cuda_path = nvidia_base_path / 'cuda_runtime' / 'bin'
    cublas_path = nvidia_base_path / 'cublas' / 'bin'
    cudnn_path = nvidia_base_path / 'cudnn' / 'bin'
    nvrtc_path = nvidia_base_path / 'cuda_nvrtc' / 'bin'
    
    paths_to_add = [
        str(cuda_path),
        str(cublas_path),
        str(cudnn_path),
        str(nvrtc_path),
    ]
    env_vars = ['CUDA_PATH', 'PATH']
    
    for env_var in env_vars:
        current_value = os.environ.get(env_var, '')
        new_value = os.pathsep.join(paths_to_add + [current_value] if current_value else paths_to_add)
        os.environ[env_var] = new_value

set_cuda_paths()

import torch
from PIL import Image
from transformers import AutoProcessor, AutoModelForVision2Seq

DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

image_path = r"D:\Scripts\bench_vision\IMG_140531.JPG"
image = Image.open(image_path)
width = image.width
height = image.height
model_dir = r"D:\Scripts\bench_vision\HuggingFaceTB--SmolVLM-Instruct"

processor = AutoProcessor.from_pretrained(model_dir)

model = AutoModelForVision2Seq.from_pretrained(
    model_dir,
    torch_dtype=torch.bfloat16,
    attn_implementation="flash_attention_2" if DEVICE == "cuda" else "eager",
    low_cpu_mem_usage=True,
)

model.to(DEVICE)

prompt = f"""<|im_start|>User:<image>Can you describe this image in detail but be succinct and do not repeat yourself?<end_of_utterance>
Assistant:"""

inputs = processor(text=prompt, images=[image], return_tensors="pt")
inputs = inputs.to(DEVICE)

generated_ids = model.generate(**inputs, max_new_tokens=500)
generated_ids = generated_ids[:, inputs['input_ids'].shape[1]:]
generated_texts = processor.batch_decode(
    generated_ids,
    skip_special_tokens=True,
)
print(generated_texts[0])

I have. That works fine. But if I include a local directory like ./codespace/image1.jpg, the model does not see the image.

Did it work? Always curious of whether something works on another platform.