Delete app.py

app.py

@@ -1,53 +0,0 @@
-"""Hugging Face Space: live CPU demo for the exact fixed evidence-graph reranker."""
-import json
-import gradio as gr
-from evidence_graph_reranker import EvidenceGraphReranker
-
-_RERANKERS = {}
-EXAMPLE = [
-    {"title": "Atlas device", "text": "The Atlas device was launched by Novus Labs in 2025."},
-    {"title": "Novus Labs", "text": "Novus Labs was founded by Mira Chen in 2018."},
-    {"title": "Orion spacecraft", "text": "The Orion spacecraft is designed for space missions."},
-    {"title": "Electronics", "text": "Electronics can be manufactured in many countries."},
-]
-
-
-def demo(query, passages_json, budget, backend):
-    try:
-        docs = json.loads(passages_json)
-        if not isinstance(docs, list) or not 1 <= len(docs) <= 100:
-            raise ValueError("Enter a JSON list containing 1–100 passages.")
-        if not query.strip():
-            raise ValueError("Query cannot be empty.")
-        if backend not in _RERANKERS:
-            _RERANKERS[backend] = EvidenceGraphReranker(backend=backend, device="cpu", batch_size=16)
-        ranked = _RERANKERS[backend].rank(query, docs, k=int(budget))
-        table = [
-            [i + 1, r["index"], r["title"] or "(untitled)",
-             round(r["relevance_score"], 4), round(r["step_score"], 4),
-             round(r["directed_link_bonus"], 2)]
-            for i, r in enumerate(ranked)
-        ]
-        return table, json.dumps(ranked, indent=2, ensure_ascii=False)
-    except Exception as exc:
-        return [], json.dumps({"error": str(exc)})
-
-with gr.Blocks(title="Evidence-Graph Reranker") as app:
-    gr.Markdown("# Evidence-Graph Reranker\nTraining-free, query-conditioned **joint evidence selection** after neural relevance scoring.\n\nA passage's usefulness can rise when it completes a link from a previously selected passage. This demo uses **one fixed graph rule** for every dataset. Model weights are fetched from their original Hugging Face repositories.")
-    query = gr.Textbox(label="Query", value="Who founded the company that released the Atlas device?")
-    passages = gr.Code(label="Candidate passages (JSON array; optional title)", language="json",
-                       value=json.dumps(EXAMPLE, indent=2))
-    with gr.Row():
-        budget = gr.Slider(minimum=1, maximum=10, step=1, value=3, label="Passage budget (k)")
-        backend = gr.Dropdown(choices=["cross-encoder", "dense"], value="cross-encoder",
-                              label="Pretrained neural relevance model")
-    go = gr.Button("Select complementary evidence", variant="primary")
-    table = gr.Dataframe(headers=["Rank", "Candidate index", "Title", "Neural score", "Selection score", "Link bonus"],
-                         datatype=["number", "number", "str", "number", "number", "number"],
-                         label="Selected passages", interactive=False)
-    output = gr.Code(label="Full selection JSON", language="json")
-    go.click(demo, inputs=[query, passages, budget, backend], outputs=[table, output])
-    gr.Markdown("**Interpretation:** `selection_score` is a step-dependent heuristic, not a probability. Gold labels are never used by the selector. Reported benchmarks measure complete evidence coverage, *not* LLM-generated answer accuracy. See `MODEL_CARD.md` for protocols and caveats.")
-
-if __name__ == "__main__":
-    app.launch()