Upload folder using huggingface_hub

README.md

@@ -37,11 +37,33 @@ or prediction engine — this is Garrett's own tooling, unrelated to GSE.**
 - Streaming responses
 - Copy button on every reply (hover a message)
 - Chat history capped at the last 12 turns to bound context growth
-- **Documents tab**: upload PDF/DOCX/PPTX/HTML/TXT/MD, parsed locally via
-  `docling` (MIT, no external API), chunked, and indexed with BM25 keyword
-  search (`rank_bm25`, pure Python, no embedding model, no extra GPU
-  weight). Check "Search my documents" in the Chat tab to have the model
-  retrieve and cite relevant excerpts instead of answering blind.
+
+## Tried and reverted: document-grounded chat (RAG)
+
+A Documents tab (upload PDF/DOCX/PPTX, parsed via `docling`, retrieved with
+BM25) was built and pushed, then reverted. Real cause, not a code bug: this
+Space's own `MODEL_ID` (72B) downloads **145GB** of raw weights before
+4-bit quantization, against a **150GB per-Space storage cap** — docling's
+own supplementary models pushed it over, and the container was evicted
+("Workload evicted, storage limit exceeded (150G)"). There isn't enough
+storage headroom left on this Space for anything else meaningful. If
+document-grounded chat is wanted later, it needs either a smaller base
+model here, or to live on a Space that isn't already this close to the
+storage ceiling (e.g. `studio-media`, whose models total well under half
+that footprint).
+
+## Fixed: a real gradio_client bug
+
+ZeroGPU force-pins `gradio[oauth]==5.9.1`. Its bundled `gradio_client`
+(1.5.2) has a real bug: its JSON-schema parser assumes every schema node is
+a `dict`, but valid JSON Schema allows a bare boolean (e.g.
+`additionalProperties: true`) — this crashed *every* API call to this
+Space with `TypeError: argument of type 'bool' is not iterable`. The
+earlier fix (pinning `pydantic==2.10.3`, which happens to avoid generating
+that schema shape) was a workaround, not a fix, and it broke `docling`'s
+own serialization elsewhere. The real fix, now in `app.py`: a small
+monkeypatch on `gradio_client.utils._json_schema_to_python_type` that
+handles non-dict schema nodes directly. No pydantic pin needed.
 
 ## Swapping the model
 

app.py

@@ -9,17 +9,17 @@ from transformers import (
     BitsAndBytesConfig,
     TextIteratorStreamer,
 )
-from docling.document_converter import DocumentConverter
-from rank_bm25 import BM25Okapi
 
 # ---------------------------------------------------------------------------
 # Fix a real bug in gradio_client 1.5.2 (bundled with ZeroGPU's forced
 # gradio==5.9.1): its JSON-schema-to-python-type parser assumes every schema
 # node is a dict, but valid JSON Schema allows a bare boolean (e.g.
 # additionalProperties: true) — that crashes EVERY API call to this Space
-# with "TypeError: argument of type 'bool' is not iterable". Patched here,
-# not worked around with an old pydantic pin (which broke docling's own
-# serialization) — this is the actual bug.
+# with "TypeError: argument of type 'bool' is not iterable". Patched here
+# rather than pinning an old pydantic (which broke docling's own
+# serialization when this Space briefly tried to carry a Documents/RAG tab —
+# reverted, see README: the 72B model's 145GB raw download leaves no real
+# room in the 150GB per-Space storage cap for anything else).
 import gradio_client.utils as _gc_utils
 
 _orig_json_schema_to_python_type = _gc_utils._json_schema_to_python_type
@@ -35,8 +35,6 @@ _gc_utils._json_schema_to_python_type = _patched_json_schema_to_python_type
 
 MODEL_ID = "Qwen/Qwen2.5-72B-Instruct"
 MAX_HISTORY_TURNS = 12
-CHUNK_CHARS = 800
-TOP_K = 4
 
 SYSTEM_PROMPTS = {
     "General": "You are a helpful assistant.",
@@ -68,82 +66,16 @@ model = AutoModelForCausalLM.from_pretrained(
     attn_implementation="sdpa",
 )
 
-_docling_converter = None
-CHUNKS = []  # list of {"source": filename, "text": chunk_text}
-BM25_INDEX = None
-
-
-def _docling():
-    global _docling_converter
-    if _docling_converter is None:
-        _docling_converter = DocumentConverter()
-    return _docling_converter
-
-
-def _chunk_markdown(md_text, source, max_chars=CHUNK_CHARS):
-    paragraphs = [p.strip() for p in md_text.split("\n\n") if p.strip()]
-    chunks, current = [], ""
-    for p in paragraphs:
-        if len(current) + len(p) + 1 <= max_chars:
-            current = f"{current}\n{p}".strip()
-        else:
-            if current:
-                chunks.append(current)
-            current = p
-    if current:
-        chunks.append(current)
-    return [{"source": source, "text": c} for c in chunks]
-
-
-def index_documents(files):
-    global CHUNKS, BM25_INDEX
-    if not files:
-        return "No files given.", ""
-    new_chunks = []
-    for f in files:
-        result = _docling().convert(f.name if hasattr(f, "name") else f)
-        md = result.document.export_to_markdown()
-        source = (f.name if hasattr(f, "name") else str(f)).split("/")[-1]
-        new_chunks.extend(_chunk_markdown(md, source))
-    CHUNKS = new_chunks  # replace — one working set at a time, kept simple
-    tokenized = [c["text"].lower().split() for c in CHUNKS]
-    BM25_INDEX = BM25Okapi(tokenized) if tokenized else None
-    sources = sorted({c["source"] for c in CHUNKS})
-    status = f"Indexed {len(sources)} document(s), {len(CHUNKS)} chunks."
-    listing = "\n".join(f"- {s}" for s in sources)
-    return status, listing
-
-
-def _retrieve(query, k=TOP_K):
-    if BM25_INDEX is None or not CHUNKS:
-        return []
-    scores = BM25_INDEX.get_scores(query.lower().split())
-    ranked = sorted(range(len(CHUNKS)), key=lambda i: scores[i], reverse=True)
-    return [CHUNKS[i] for i in ranked[:k] if scores[i] > 0]
-
-
-def _duration(message, history, system_prompt_name, use_docs, max_new_tokens, temperature):
+
+def _duration(message, history, system_prompt_name, max_new_tokens, temperature):
     return min(120, 40 + int(max_new_tokens) // 10)
 
 
 @spaces.GPU(duration=_duration)
-def generate(message, history, system_prompt_name, use_docs, max_new_tokens, temperature):
+def generate(message, history, system_prompt_name, max_new_tokens, temperature):
     history = history[-(MAX_HISTORY_TURNS * 2):]
     system_prompt = SYSTEM_PROMPTS.get(system_prompt_name, SYSTEM_PROMPTS["General"])
 
-    retrieved = _retrieve(message) if use_docs else []
-    if retrieved:
-        context_block = "\n\n".join(
-            f"[Source: {c['source']}]\n{c['text']}" for c in retrieved
-        )
-        system_prompt = (
-            f"{system_prompt}\n\nUse the following excerpts from the user's own "
-            "uploaded documents if relevant to their question. Cite the source "
-            "filename in brackets when you use one. If the excerpts don't answer "
-            "the question, say so and answer from general knowledge instead.\n\n"
-            f"{context_block}"
-        )
-
     messages = [{"role": "system", "content": system_prompt}]
     for turn in history:
         messages.append({"role": turn["role"], "content": turn["content"]})
@@ -181,42 +113,23 @@ def generate(message, history, system_prompt_name, use_docs, max_new_tokens, tem
 
 chatbot = gr.Chatbot(type="messages", show_copy_button=True, label="Chat")
 
-with gr.Blocks(title="Beexly Studio Chat (ZeroGPU)") as demo:
-    gr.Markdown(
-        "# Beexly Studio Chat (ZeroGPU)\n"
+demo = gr.ChatInterface(
+    fn=generate,
+    type="messages",
+    chatbot=chatbot,
+    additional_inputs=[
+        gr.Dropdown(
+            list(SYSTEM_PROMPTS.keys()), value="General", label="System prompt"
+        ),
+        gr.Slider(64, 1024, value=512, step=64, label="Max new tokens"),
+        gr.Slider(0.0, 1.5, value=0.7, step=0.1, label="Temperature"),
+    ],
+    title="Beexly Studio Chat (ZeroGPU)",
+    description=(
         f"Personal free-GPU chat — `{MODEL_ID}` (4-bit), streaming, on ZeroGPU. "
-        "Not part of any Galaxy Sports Edge product."
-    )
-    with gr.Tab("Chat"):
-        chat_iface = gr.ChatInterface(
-            fn=generate,
-            type="messages",
-            chatbot=chatbot,
-            additional_inputs=[
-                gr.Dropdown(
-                    list(SYSTEM_PROMPTS.keys()), value="General", label="System prompt"
-                ),
-                gr.Checkbox(
-                    label="Search my documents (Documents tab)", value=False
-                ),
-                gr.Slider(64, 1024, value=512, step=64, label="Max new tokens"),
-                gr.Slider(0.0, 1.5, value=0.7, step=0.1, label="Temperature"),
-            ],
-        )
-    with gr.Tab("Documents"):
-        gr.Markdown(
-            "Upload documents (PDF/DOCX/PPTX/HTML/TXT/MD), parsed locally via "
-            "`docling` (MIT, no external API) and retrieved with BM25 keyword "
-            "search — no embedding model, no extra GPU weight. Check "
-            "'Search my documents' in the Chat tab to use them."
-        )
-        docs_files = gr.File(label="Documents", file_count="multiple")
-        docs_btn = gr.Button("Index documents")
-        docs_status = gr.Textbox(label="Status", interactive=False)
-        docs_list = gr.Textbox(label="Indexed sources", interactive=False)
-        docs_btn.click(
-            index_documents, inputs=docs_files, outputs=[docs_status, docs_list]
-        )
+        "Hover any reply for a copy button. Not part of any Galaxy Sports Edge product."
+    ),
+)
 
 if __name__ == "__main__":
     demo.launch()

requirements.txt

@@ -4,5 +4,3 @@ transformers>=4.46.0
 accelerate>=1.1.0
 torch>=2.4.0
 bitsandbytes>=0.44.0
-docling
-rank_bm25