app.py

import math
import torch
from transformers import AutoTokenizer, AutoModel, AutoProcessor
import gradio as gr
from PIL import Image

# === 分配层到多 GPU ===
def split_model(model_path):
    from transformers import AutoConfig
    device_map = {}
    world_size = torch.cuda.device_count()
    config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
    num_layers = config.llm_config.num_hidden_layers
    num_layers_per_gpu = math.ceil(num_layers / (world_size - 0.5))
    num_layers_per_gpu = [num_layers_per_gpu] * world_size
    num_layers_per_gpu[0] = math.ceil(num_layers_per_gpu[0] * 0.5)
    layer_cnt = 0
    for i, num_layer in enumerate(num_layers_per_gpu):
        for _ in range(num_layer):
            device_map[f'language_model.model.layers.{layer_cnt}'] = i
            layer_cnt += 1
    device_map['vision_model'] = 0
    device_map['mlp1'] = 0
    device_map['language_model.model.tok_embeddings'] = 0
    device_map['language_model.model.embed_tokens'] = 0
    device_map['language_model.output'] = 0
    device_map['language_model.model.norm'] = 0
    device_map['language_model.model.rotary_emb'] = 0
    device_map['language_model.lm_head'] = 0
    device_map[f'language_model.model.layers.{num_layers - 1}'] = 0
    return device_map

# === 模型路径 ===
model_path = "OpenGVLab/InternVL3-14B"
device_map = split_model(model_path)

# === 加载模型和处理器 ===
model = AutoModel.from_pretrained(
    model_path,
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    use_flash_attn=True,
    trust_remote_code=True,
    device_map=device_map
).eval()

tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)

# === 推理函数 ===
def infer(image: Image.Image, prompt: str):
    inputs = processor(text=prompt, images=image, return_tensors="pt").to("cuda")
    output = model.generate(**inputs, max_new_tokens=512)
    answer = tokenizer.decode(output[0], skip_special_tokens=True)
    return answer

# === Gradio 界面 ===
gr.Interface(
    fn=infer,
    inputs=[
        gr.Image(type="pil", label="Upload Image"),
        gr.Textbox(label="Your Prompt", placeholder="Ask a question about the image...")
    ],
    outputs="text",
    title="InternVL3-14B Multimodal Demo",
    description="Upload an image and ask a question. InternVL3-14B will answer using vision + language."
).launch()