fix dependency

app.py

@@ -1,39 +1,44 @@
 import gradio as gr
-from transformers import AutoProcessor, AutoModelForVision2Seq
+from deepseek_vl.models import VLChatProcessor, MultiModalityCausalLM
+from deepseek_vl.utils.io import load_pil_images
 import torch
 
-class DeepSeekVL:
-    def __init__(self, model_path="deepseek-ai/deepseek-vl-7b", device="cpu"):
-        self.device = device
-        self.processor = AutoProcessor.from_pretrained(model_path)
-        self.model = AutoModelForVision2Seq.from_pretrained(
-            model_path,
-            torch_dtype=torch.float32
-        ).to(device)
-
-    def generate(self, image, question, max_new_tokens=128):
-        inputs = self.processor(text=question, images=image, return_tensors="pt").to(self.device)
-        with torch.no_grad():
-            output_ids = self.model.generate(**inputs, max_new_tokens=max_new_tokens)
-        return self.processor.batch_decode(output_ids, skip_special_tokens=True)[0]
-
-# Initialize DeepSeek-VL model (CPU for free Spaces)
-model = DeepSeekVL(model_path="deepseek-ai/deepseek-vl-7b", device="cpu")
+model_path = "deepseek-ai/deepseek-vl-7b-chat"
+vl_chat_processor = VLChatProcessor.from_pretrained(model_path)
+tokenizer = vl_chat_processor.tokenizer
+vl_gpt = MultiModalityCausalLM.from_pretrained(model_path, trust_remote_code=True).to("cpu")
 
 def qa(image, question):
-    # Run DeepSeek-VL inference: image + question -> answer
-    return model.generate(image, question)
+    conversation = [
+        {"role": "User", "content": "<image_placeholder>" + question, "images": [image]},
+        {"role": "Assistant", "content": ""}
+    ]
+    pil_images = load_pil_images(conversation)
+    prepare_inputs = vl_chat_processor(
+        conversations=conversation,
+        images=pil_images,
+        force_batchify=True
+    ).to("cpu")
+    inputs_embeds = vl_gpt.prepare_inputs_embeds(**prepare_inputs)
+    outputs = vl_gpt.language_model.generate(
+        inputs_embeds=inputs_embeds,
+        attention_mask=prepare_inputs.attention_mask,
+        pad_token_id=tokenizer.eos_token_id,
+        bos_token_id=tokenizer.bos_token_id,
+        eos_token_id=tokenizer.eos_token_id,
+        max_new_tokens=512,
+        do_sample=False,
+        use_cache=True
+    )
+    answer = tokenizer.decode(outputs[0].cpu().tolist(), skip_special_tokens=True)
+    return answer
 
 demo = gr.Interface(
     fn=qa,
-    inputs=[
-        gr.Image(type="pil", label="Upload Image"),
-        gr.Textbox(label="Enter your question")
-    ],
+    inputs=[gr.Image(type="pil", label="Upload Image"), gr.Textbox(label="Enter your question")],
     outputs="text",
     title="DeepSeek-VL Multimodal QA Demo",
     description="Upload an image and enter a question. Experience DeepSeek-VL's vision-language capabilities."
 )
 
-if __name__ == "__main__":
-    demo.launch()
+demo.launch()

requirements.txt

@@ -1,19 +1,4 @@
-torch==2.0.1
-transformers>=4.38.2
-timm>=0.9.16
-accelerate
-sentencepiece
-attrdict
-einops
-
-# for gradio demo
-gradio==3.48.0
-gradio-client==0.6.1
-mdtex2html==1.3.0
-pypinyin==0.50.0
-tiktoken==0.5.2
-tqdm==4.64.0
-colorama==0.4.5
-Pygments==2.12.0
-markdown==3.4.1
-SentencePiece==0.1.96
+gradio
+torch
+transformers
+git+https://github.com/deepseek-ai/DeepSeek-VL.git