it run good in colab t4
import torch
from transformers import BitsAndBytesConfig, AutoProcessor, Glm4vForConditionalGeneration
from PIL import Image
MODEL_PATH = "zai-org/GLM-4.6V-Flash"
إعداد ضغط 4-بت
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_use_double_quant=True,
bnb_4bit_compute_dtype=torch.float16
)
processor = AutoProcessor.from_pretrained(MODEL_PATH, trust_remote_code=True)
model = Glm4vForConditionalGeneration.from_pretrained(
MODEL_PATH,
quantization_config=bnb_config,
device_map="auto",
trust_remote_code=True
)
--- اقرأ الصورة من ملف محلي ---
image_path = "/content/Grayscale_8bits_palette_sample_image.png" # عدِّل هذا إلى مسار الصورة عندك
image = Image.open(image_path).convert("RGB")
إعداد رسالة: صورة + نص
messages = [
{
"role": "user",
"content": [
# لن تحتاج url لأن الصورة محليّة
{"type": "image"},
{"type": "text", "text": "Describe this image in detail."}
],
}
]
تجهيز المدخلات: الصور + النص
chat_template = processor.apply_chat_template(messages, add_generation_prompt=True)
inputs = processor(
text=chat_template,
images=[image],
return_tensors="pt"
)
نقل التينسورات إلى نفس جهاز النموذج (GPU أو CPU)
inputs = {k: v.to(model.device) for k, v in inputs.items()}
inputs.pop("token_type_ids", None)
توليد النص (الوصف)
with torch.no_grad():
generated_ids = model.generate(**inputs, max_new_tokens=51)
output = processor.decode(
generated_ids[0][ inputs["input_ids"].shape[1]: ],
skip_special_tokens=True
)
print(output)
Using a slow image processor as use_fast is unset and a slow processor was saved with this model. use_fast=True will be the default behavior in v4.52, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with use_fast=False.
Unrecognized keys in rope_parameters for 'rope_type'='default': {'partial_rotary_factor', 'mrope_section'}
Download complete:
0.00/0.00 [00:00<?, ?B/s]
Fetching 4 files: 100%
4/4 [00:00<00:00, 164.82it/s]
Loading weights: 100%
704/704 [02:02<00:00, 5.74it/s, Materializing param=model.visual.post_layernorm.weight]
Got it, let's describe this black-and-white image of a parrot. First, the main subject is a parrot, likely a macaw or similar species, with a prominent crest on its head. The parrot is perched on a
!pip install transformers==5.0.0rc0
