Upload 7 files

.gitattributes

@@ -36,3 +36,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 catalogue/ASICS_Catalog.pdf filter=lfs diff=lfs merge=lfs -text
 catalogue/flexpocket.pdf filter=lfs diff=lfs merge=lfs -text
 catalogue/M8_Private_Sale_p2.pdf filter=lfs diff=lfs merge=lfs -text
+examples/ASICS_Catalog.pdf filter=lfs diff=lfs merge=lfs -text
+examples/flexpocket.pdf filter=lfs diff=lfs merge=lfs -text
+examples/M8_Private_Sale_p2.pdf filter=lfs diff=lfs merge=lfs -text

app.py

@@ -1,25 +1,32 @@
 import os
 import gradio as gr
-import shutil
 import main
 
+#os.environ["CUDA_VISIBLE_DEVICES"]='0'
+#os.environ["USE_GPU"]="True"
+
+
 def predict_from_pdf(pdf_file):
     upload_dir = "./catalogue/"
     os.makedirs(upload_dir, exist_ok=True)
-    
+
     try:
+        # Save the uploaded file to a temporary location
         dest_path = os.path.join(upload_dir, pdf_file.name)
         with open(dest_path, "wb") as f:
-            f.write(pdf_file.read())
-        
-        df, response = main_oss.process_pdf_catalog(dest_path)
+            with open(pdf_file.name, "rb") as uploaded_file:
+                f.write(uploaded_file.read())
+
+        # Process the PDF
+        df, response = main.process_pdf_catalog(dest_path)
         return df, response
     except Exception as e:
         return None, f"Error: {str(e)}"
 
+
 pdf_examples = [
-    ["catalogue/flexpocket.pdf"],
-    ["catalogue/ASICS_Catalog.pdf"],
+    ["examples/flexpocket.pdf"],
+    ["examples/ASICS_Catalog.pdf"],
 ]
 
 demo = gr.Interface(
@@ -28,9 +35,9 @@ demo = gr.Interface(
     outputs=["json", "text"],
     examples=pdf_examples,
     title="Open Source PDF Catalog Parser",
-    description="Efficient PDF catalog processing using MinerU and OpenLLM",
-    article="Uses MinerU for layout analysis and DeepSeek-1.3B for structured extraction"
+    description="Efficient PDF catalog processing using PyMuPDF and OpenLLM",
+    article="Uses MinerU for layout analysis and DeepSeek-7B for structured extraction"
 )
 
 if __name__ == "__main__":
-    demo.queue().launch(server_name="0.0.0.0", server_port=7860)
+    demo.queue().launch(server_name="0.0.0.0", server_port=7860, share=True)

examples/ASICS_Catalog.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6263bd2732f674657e22e83deb21d3e34afedc028834e078e28c73c0a305d4b3
+size 392487

examples/M8_Private_Sale_p2.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ebc4730fd7dc87baf2f67c81934d9d3162f662453a7e8d50190b7018058ac1f
+size 565084

examples/flexpocket.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c89cb2cdc331adc6bef8814bf69a02b49acb3d7695da404c80c54910c81b947c
+size 148538

main.py

@@ -36,21 +36,36 @@ class ProductSpec:
 
 class PDFProcessor:
     def __init__(self):
-        self.emb_model = SentenceTransformer('all-MiniLM-L6-v2')
-        self.llm = self._initialize_llm()
+        self.emb_model = self._initialize_emb_model("all-MiniLM-L6-v2")
+        self.llm = self._initialize_llm("deepseek-llm-7b-base.Q5_K_M.gguf")
         self.output_dir = Path("./output")
         self.output_dir.mkdir(exist_ok=True)
-        
-    def _initialize_llm(self):
+
+    def _initialize_emb_model(self, model_name):
+        model = SentenceTransformer("sentence-transformers/" + model_name)
+        model.save('models/'+ model_name)
+        return model
+
+    def _initialize_llm(self, model_name):
         """Initialize LLM with automatic download if needed"""
-        return Llama.from_pretrained(
-            repo_id="TheBloke/deepseek-llm-7B-base-GGUF",
-            filename="deepseek-llm-7b-base.Q5_K_M.gguf",
-            n_ctx=2048,
-            n_threads=os.cpu_count() - 1,
-            n_gpu_layers=35 if os.getenv('USE_GPU') else 0,
-            verbose=False
-        )
+        model_path = os.path.join("models/", model_name)
+        if os.path.exists(model_path):
+            return Llama(
+                model_path=model_path,
+                n_ctx=4096,
+                n_gpu_layers=35 if os.getenv('USE_GPU') else 0,
+                n_threads=os.cpu_count() - 1,
+                verbose=False
+            )
+        else:
+            return Llama.from_pretrained(
+                repo_id="TheBloke/deepseek-llm-7B-base-GGUF",
+                filename=model_name,
+                n_ctx=4096,
+                n_threads=os.cpu_count() - 1,
+                n_gpu_layers=35 if os.getenv('USE_GPU') else 0,
+                verbose=False
+            )
     
     def process_pdf(self, pdf_path: str) -> Dict:
         """Process PDF using MinerU pipeline"""

requirements.txt

@@ -5,7 +5,6 @@ llama-cpp-python[huggingface-hub]
 magic-pdf
 paddleocr
 paddlepaddle
-torch
 git+https://github.com/opendatalab/MinerU.git
 omegaconf
 matplotlib
@@ -14,8 +13,8 @@ ultralytics
 doclayout-yolo
 unimernet
 struct-eqtable
-lmdeploy
+#lmdeploy
 tf-keras
 pycocotools
 git+https://github.com/facebookresearch/detectron2.git
-rapid-table
+rapid-table