feat: 部署demo

app.py

@@ -0,0 +1,59 @@
+import gradio as gr
+import spaces
+import torch
+import os
+import subprocess
+
+# os.system("pip install git+https://github.com/huggingface/transformers")
+
+from PIL import Image 
+import requests 
+from transformers import AutoModelForCausalLM 
+from transformers import AutoProcessor 
+
+model_id = "microsoft/Phi-3-vision-128k-instruct" 
+
+model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True, torch_dtype="auto", _attn_implementation='eager').cuda() # use _attn_implementation='eager' to disable flash attention
+
+processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True) 
+
+@spaces.GPU
+def infer(u, t):
+    if len(u) < 1:
+        u = "https://lf3-static.bytednsdoc.com/obj/eden-cn/pbovhozuha/output.png"
+    if len(t) < 1:
+        t = "Convert the text in the image to markdown"
+    messages = messages = [ 
+        {"role": "user", "content": "<|image_1|>\n" + t},
+    ] 
+    url = u
+    image = Image.open(requests.get(url, stream=True).raw) 
+
+    prompt = processor.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+
+    inputs = processor(prompt, [image], return_tensors="pt").to(model.device)
+
+    generation_args = { 
+        "max_new_tokens": 512, 
+        "temperature": 0.7, 
+        "do_sample": True, 
+    } 
+
+    generate_ids = model.generate(**inputs, eos_token_id=processor.tokenizer.eos_token_id, **generation_args) 
+
+    # remove input tokens 
+    generate_ids = generate_ids[:, inputs['input_ids'].shape[1]:]
+    response = processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]  
+
+    return response
+    
+
+demo = gr.Interface(
+    fn=infer,
+    inputs=[
+        gr.Text(label="url"),
+        gr.Text(label="text"),
+    ],
+    outputs=gr.Text(),
+)
+demo.launch()

requirements.txt

@@ -0,0 +1,6 @@
+numpy==1.24.4
+Pillow==10.3.0
+Requests==2.31.0
+torch==2.3.0
+torchvision==0.18.0
+transformers==4.40.2