Huihui-GLM-4.6V-Flash-abliterated

from transformers import AutoProcessor, Glm4vForConditionalGeneration
import torch

MODEL_PATH = "huihui-ai/Huihui-GLM-4.6V-Flash-abliterated"
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "url": "https://upload.wikimedia.org/wikipedia/commons/f/fa/Grayscale_8bits_palette_sample_image.png"
            },
            {
                "type": "text",
                "text": "describe this image"
            }
        ],
    }
]
processor = AutoProcessor.from_pretrained(MODEL_PATH, use_fast=True)
model = Glm4vForConditionalGeneration.from_pretrained(
    pretrained_model_name_or_path=MODEL_PATH,
    torch_dtype="auto",
    device_map="auto",
    trust_remote_code=True,
)
inputs = processor.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_dict=True,
    return_tensors="pt"
).to(model.device)

inputs.pop("token_type_ids", None)

generated_ids = model.generate(**inputs, max_new_tokens=8192)
output_text = processor.decode(generated_ids[0][inputs["input_ids"].shape[1]:], skip_special_tokens=False)
print(output_text)