fix: update model loading logic

app.py

@@ -1,5 +1,21 @@
 import gradio as gr
-from deepseek_vl import DeepSeekVL
+from transformers import AutoProcessor, AutoModelForVision2Seq
+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")

requirements.txt

@@ -1,4 +1,19 @@
-deepseek-vl
-gradio
-torch
-transformers
+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