Update app.py

app.py

@@ -17,6 +17,7 @@ import gradio as gr
 
 # ---- Parsers ----
 from docx import Document
+import traceback
 
 # ---- Embeddings ----
 # We try sentence-transformers. If unavailable (or offline), we fall back to HashingVectorizer.
@@ -37,8 +38,53 @@ def _load_st_model():
     except Exception as e:
         return None
 
-def read_docx(docx_path: str) -> List[str]:
-    doc = Document(docx_path)
+def _resolve_file_input(file_obj):
+    """Return (bytes_io, display_name) for a variety of Gradio/HF file input shapes.
+    Supports: tempfile objects, dicts with 'name'/'path'/'data', raw path strings, or bytes.
+    """
+    import io, os
+    # 1) Dict shape (some Gradio environments)
+    if isinstance(file_obj, dict):
+        # Prefer an on-disk path if present
+        for key in ("path", "name"):
+            p = file_obj.get(key)
+            if isinstance(p, str) and os.path.exists(p):
+                with open(p, "rb") as f:
+                    return io.BytesIO(f.read()), os.path.basename(p)
+        # Raw bytes in 'data'
+        data = file_obj.get("data")
+        if isinstance(data, (bytes, bytearray)):
+            return io.BytesIO(bytes(data)), file_obj.get("orig_name", "upload.docx")
+    # 2) Tempfile-like object
+    if hasattr(file_obj, "read") and hasattr(file_obj, "name"):
+        try:
+            file_obj.seek(0)
+            content = file_obj.read()
+            if isinstance(content, (bytes, bytearray)):
+                return io.BytesIO(content), os.path.basename(getattr(file_obj, "name", "upload.docx"))
+        except Exception:
+            pass
+        # Fallback: open by path
+        p = getattr(file_obj, "name", None)
+        if isinstance(p, str) and os.path.exists(p):
+            with open(p, "rb") as f:
+                return io.BytesIO(f.read()), os.path.basename(p)
+    # 3) Path string
+    if isinstance(file_obj, str) and os.path.exists(file_obj):
+        with open(file_obj, "rb") as f:
+            import os
+            return io.BytesIO(f.read()), os.path.basename(file_obj)
+    # 4) Raw bytes
+    if isinstance(file_obj, (bytes, bytearray)):
+        return io.BytesIO(bytes(file_obj)), "upload.docx"
+    # Unknown shape
+    return None, "upload.docx"
+
+def read_docx_any(file_obj) -> List[str]:
+    bio, _ = _resolve_file_input(file_obj)
+    if bio is None:
+        raise ValueError("Could not read uploaded .docx file; unsupported input shape.")
+    doc = Document(bio)
     paras = [p.text.strip() for p in doc.paragraphs]
     paras = [p for p in paras if p and not p.isspace()]
     return paras
@@ -295,7 +341,7 @@ def process_pipeline(docx_file, units_mode, K, iters, beta, bins, tau, seed):
         return gr.update(value="# Please upload a .docx file."), None, None, None, None
 
     # Read file
-    paras = read_docx(docx_file.name)
+    paras = read_docx_any(docx_file)
     units = paragraphs_to_units(paras, mode=units_mode)
 
     if len(units) == 0:
@@ -380,7 +426,7 @@ with gr.Blocks(title="Constellation Harvest Regularization (CHR)") as demo:
 
     with gr.Row():
         with gr.Column(scale=1):
-            docx_file = gr.File(label=".docx document", file_types=[".docx"])
+            docx_file = gr.File(label=".docx document", file_types=[".docx"], file_count="single")
             units_mode = gr.Radio(choices=["paragraphs", "sentences"], value="paragraphs", label="Unit granularity")
             K = gr.Slider(2, 24, value=8, step=1, label="K (number of constellations)")
             iters = gr.Slider(5, 100, value=30, step=1, label="Iterations")