Upload 4 files

Dockerfile

@@ -0,0 +1,11 @@
+
+# Optional: containerize for non-HF hosting (Railway/Fly/Render/VPS)
+FROM python:3.11-slim
+
+WORKDIR /app
+COPY requirements.txt ./
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY app.py ./
+EXPOSE 7860
+CMD ["python", "app.py"]

README.md

@@ -1,13 +1,29 @@
----
-title: Axionx Demo
-emoji: 🌖
-colorFrom: indigo
-colorTo: gray
-sdk: gradio
-sdk_version: 5.49.1
-app_file: app.py
-pinned: false
-short_description: Stable Question Answering demo built with Gradio and Transfo
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+
+# AxionX Digital — QA Demo (Hugging Face Space)
+
+This repo hosts a stable, year-long demo of a Question Answering model using a pinned DistilBERT checkpoint.
+
+## Why this stays stable
+- **Pinned Python deps** in `requirements.txt` prevent surprise breaking changes.
+- **Pinned model** (`distilbert-base-cased-distilled-squad`) avoids checkpoint drift.
+- **Simple Gradio app** with a fixed JSON output schema.
+- **CPU-only**: no GPU reliance or CUDA driver changes.
+
+## Deploying to Hugging Face Spaces
+1. Create a new Space (SDK: **Gradio**, Hardware: **CPU Basic**).
+2. Upload these files: `app.py`, `requirements.txt`, `README.md`.
+3. The app auto-builds and serves at your permanent Space URL.
+
+### Optional (Custom domain / always-on)
+- Switch Space to PRO for **“Always On”** hardware if you need zero cold-starts.
+- The free CPU Basic URL remains permanent; the app may sleep when idle but wakes on first request.
+
+## Local run (for testing)
+```bash
+pip install -r requirements.txt
+python app.py
+# open http://localhost:7860
+```
+
+## API-style use
+You can wrap `predict` behind a FastAPI if you prefer a JSON API. For a pure Space, use the UI or Gradio Client.

app.py

@@ -0,0 +1,40 @@
+
+import gradio as gr
+from transformers import pipeline
+
+# Pin the exact model to avoid unexpected changes.
+MODEL_ID = "distilbert-base-cased-distilled-squad"
+
+# Create the QA pipeline once at startup; HF Spaces will cache the weights.
+qa = pipeline("question-answering", model=MODEL_ID)
+
+EXAMPLE_CONTEXT = (
+    "AxionX Digital builds model-training tools for AI developers. "
+    "We fine-tune open-source LLMs for customer support, finance, and legal use cases. "
+    "We also provide evaluation dashboards and fast private deployments."
+)
+EXAMPLE_QUESTION = "What does AxionX Digital build?"
+
+def predict(context, question):
+    context = (context or "").strip()
+    question = (question or "").strip()
+    if not context or not question:
+        return {"answer": "Please provide both context and a question.", "score": ""}
+    res = qa(question=question, context=context)
+    # Return a stable JSON schema
+    return {"answer": res.get("answer",""), "score": round(float(res.get("score", 0.0)), 3)}
+
+with gr.Blocks(title="AxionX — Question Answering Demo") as demo:
+    gr.Markdown("# AxionX — Question Answering Demo\n"
+                "Type a paragraph in **Context**, then ask a **Question** about it.\n\n"
+                "**Model:** distilbert-base-cased-distilled-squad (pinned)")
+    with gr.Row():
+        ctx = gr.Textbox(label="Context", lines=10, value=EXAMPLE_CONTEXT)
+    q = gr.Textbox(label="Question", value=EXAMPLE_QUESTION)
+    btn = gr.Button("Get Answer")
+    out = gr.JSON(label="Result (answer, score)")
+    btn.click(predict, inputs=[ctx, q], outputs=[out])
+
+if __name__ == "__main__":
+    # share=False on Spaces; set to True only for local/Colab runs
+    demo.launch(server_name="0.0.0.0", server_port=7860)

requirements.txt

@@ -0,0 +1,10 @@
+
+# Pin versions for 12+ months stability
+# (these were current/stable at creation time — pinning prevents breaking changes)
+transformers==4.42.4
+torch==2.3.1
+gradio==4.44.0
+# Accelerated tokenizers used by transformers
+tokenizers==0.19.1
+# Explicitly pin pydantic major to avoid breaking gradio deps
+pydantic==2.7.4