智谱多模态系列：GLM-4.5V 环境配置与本地部署

# 1、创建虚拟环境 conda create -name sglang_env python=3.10 conda activate sglang_env # 2、安装相关库和依赖 pip3 install "sglang[all]>=0.5.0rc1" pip install git+https://github.com/huggingface/transformers.git 

python3 -m sglang.launch_server --model-path zai-org/GLM-4.5V \ --tp-size 4 \ --tool-call-parser glm45 \ --reasoning-parser glm45 \ --served-model-name glm-4.5v \ --port 8000 --host 0.0.0.0 

from openai import OpenAI openai_api_key = "EMPTY" openai_api_base = "http://127.0.0.1:8000/v1" client = OpenAI( api_key=openai_api_key, base_url=openai_api_base, ) response = client.chat.completions.create( model="glm-4.5v", messages=[ { "role": "user", "content": [ {"type": "text", "text": "请描述这张图片的主要内容"}, {"type": "image_url", "image_url": {"url": "file:///your_img"}} ] } ], max_tokens=512, temperature=0.0, ) print(response.choices[0].message.content.strip()) reasoning_content = getattr(response.choices[0].message, "reasoning_content", None) print("======") print(reasoning_content.strip() if reasoning_content else "No reasoning_content field in response.") 

from openai import OpenAI openai_api_key = "EMPTY" openai_api_base = "http://127.0.0.1:8000/v1" client = OpenAI( api_key=openai_api_key, base_url=openai_api_base, ) response = client.chat.completions.create( model="glm-4.5v", messages=[ { "role": "user", "content": [ {"type": "text", "text": "请比较这两张图片的主要不同点"}, {"type": "image_url", "image_url": {"url": "file:///your_img1"}}, {"type": "image_url", "image_url": {"url": "file:///your_img2"}} ] } ], max_tokens=512, temperature=0.0, ) print(response.choices[0].message.content.strip()) reasoning_content = getattr(response.choices[0].message, "reasoning_content", None) print("======") print(reasoning_content.strip() if reasoning_content else "No reasoning_content field in response.") 

from transformers import AutoProcessor, Glm4vMoeForConditionalGeneration #AutoModelForConditionalGeneration from PIL import Image import requests import torch # Load model and processor model_id = "zai-org/GLM-4.5V-FP8" model = Glm4vMoeForConditionalGeneration.from_pretrained( model_id, torch_dtype="auto", device_map="auto", trust_remote_code=True ) processor = AutoProcessor.from_pretrained(model_id, trust_remote_code 智谱 AI GLM 教程=True) # Example image loading (replace with your image path or URL) image_url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/car.jpg" image = Image.open(requests.get(image_url, stream=True).raw).convert("RGB") # Prepare the prompt prompt = "Describe this car in detail." messages = [ {"role": "user", "content": [{"type": "image", "image": image}, {"type": "text", "text": prompt}]} ] # Apply chat template and preprocess image input_ids = processor.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt") pixel_values = processor.preprocess_images(image, return_tensors="pt") # Generate response with torch.no_grad(): output_ids = model.generate( input_ids.to(model.device), pixel_values=pixel_values.to(model.device), max_new_tokens=512 ) response = processor.decode(output_ids[0], skip_special_tokens=True) print(response)