Update app.py

app.py

@@ -72,11 +72,7 @@ def _lazy_init():
                         BASE_MODEL, device_map="cpu",
                         torch_dtype=torch.float32, attn_implementation="eager")
             model = PeftModel.from_pretrained(base, MODEL_REPO)
-            try:
-                model = model.merge_and_unload()   # flatten LoRA → base (fixes CPU/PEFT forward quirks)
-                print("✅ CPU: fine-tuned adapter loaded + merged")
-            except Exception as _me:
-                print(f"⚠ merge_and_unload failed ({_me}); using PEFT-wrapped model")
+            print("✅ CPU: fine-tuned adapter loaded")
         except Exception as e:
             print(f"⚠ adapter failed ({e}); base model (CPU)")
             model = AutoModelForCausalLM.from_pretrained(
@@ -139,9 +135,7 @@ def model_parse(raw_text):
         out = model.generate(**enc, max_new_tokens=128, do_sample=False,
                              pad_token_id=tok.pad_token_id)
     decoded = tok.decode(out[0, enc["input_ids"].shape[1]:], skip_special_tokens=True)
-    parsed = _extract_json(decoded)
-    print(f"🔎 PARSE in={raw_text!r} | json_ok={parsed is not None} | raw_out={decoded!r}", flush=True)
-    return parsed
+    return _extract_json(decoded)
 
 
 # ═════════════════════════════════════════════════════════════════════════════
@@ -343,49 +337,50 @@ def render_recommendations(recs):
 # ═════════════════════════════════════════════════════════════════════════════
 DEMO_ISSUER = {"name": "מים שקטים", "tax_id": "962569844", "status": "authorized_dealer"}
 
-def _run_pipeline(raw_text, history, pending_parse, answers):
-    _lazy_init()
-    parse     = model_parse(raw_text) if pending_parse is None else pending_parse
-    questions = clarification_questions(parse, answers)
+_DOC_CHOICE_MAP = {
+    "קבלה":               "receipt",
+    "חשבונית מס":          "tax_invoice",
+    "חשבונית מס וקבלה":    "tax_invoice_receipt",
+}
 
-    if questions:
-        q_md = "\n".join(f"{i+1}. {q}" for i, q in enumerate(questions))
-        return (history + [{"role": "assistant",
-                "content": f"יש לי כמה שאלות לפני שאפיק את המסמך:\n\n{q_md}"}],
-                "", "", parse, answers)
+def generate(note, doc_choice):
+    """Single-shot, stateless pipeline:
+       note (Textbox) + doc_choice (Radio) → (status_md, doc_html, recs_html).
+       No gr.State, no second handler — the input value cannot be lost.
+    """
+    note = (note or "").strip()
+    if not note:
+        return "✏️ הקלד הערת הכנסה ולחץ הפק מסמך.", "", ""
+
+    _lazy_init()
+    parse = model_parse(note)
+
+    if parse is None or _missing_fields(parse) == ["parse_failed"]:
+        return ("🤔 לא הצלחתי לחלץ פרטים. נסח מחדש — לדוגמה: "
+                "'קיבלתי 500₪ ממשה על ייעוץ'.", "", "")
+
+    # ---- agentic clarification: only the doc-type can be genuinely ambiguous ----
+    answers = {}
+    if doc_choice and doc_choice in _DOC_CHOICE_MAP:
+        answers["doc_type"] = doc_choice            # user picked explicitly
+    elif not parse.get("doc_type"):
+        # model didn't determine it and user left it on auto → ask
+        hint = ""
+        if _needs_allocation_clarification(parse):
+            hint = " (סכום גבוה ללקוח עסקי — לרוב חשבונית מס וקבלה)"
+        return (f"❓ צריך פרט אחד נוסף: **בחר סוג מסמך** מהרשימה למעלה{hint} "
+                "ולחץ הפק מסמך שוב.", "", "")
 
     final = apply_answers(parse, answers)
     try:
         completed = core.complete(DEMO_ISSUER, final, _STATE["rng"])
     except Exception as e:
-        return (history + [{"role": "assistant", "content": f"שגיאה: {e}"}],
-                "", "", None, {})
+        return (f"⚠️ שגיאה בעיבוד: {e}", "", "")
 
-    return (history + [{"role": "assistant", "content": "✅ המסמך הופק בהצלחה"}],
+    client = final.get("client_name", "—")
+    return ("✅ המסמך הופק בהצלחה.",
             render_document(completed),
-            render_recommendations(recommend(raw_text)),
-            None, {})
-
-
-def on_submit(raw, history, pending, answers):
-    if not raw.strip():
-        return history, "", "", pending, answers
-    return _run_pipeline(raw, history + [{"role": "user", "content": raw}], None, {})
-
-
-def on_answer(user_msg, history, pending, answers, raw_text):
-    if not user_msg.strip():
-        return history, "", "", pending, answers
-    a, low = dict(answers), user_msg.strip()
-    if any(w in low for w in ["חשבונית מס", "מס וקבלה", "מס/קבלה"]):
-        a["doc_type"] = low
-    elif "קבלה" in low:
-        a["doc_type"] = "receipt"
-    elif not a.get("client_name") and pending and not pending.get("client_name"):
-        a["client_name"] = low
-    return _run_pipeline(raw_text,
-                         history + [{"role": "user", "content": user_msg}],
-                         pending, a)
+            render_recommendations(recommend(note)))
 
 
 # ═════════════════════════════════════════════════════════════════════════════
@@ -509,29 +504,21 @@ with gr.Blocks(title="Text2Receipt") as demo:
       </div>
     </div>""")
 
-    st_parse   = gr.State(None)
-    st_answers = gr.State({})
-
-    chatbot = gr.Chatbot(
-        label="", height=320, elem_classes=["chatbot-area"], rtl=True,
-        placeholder="<div dir='rtl' style='color:#1e293b;text-align:center;"
-                    "padding:48px 0;font-size:14px;'>הקלד הערה ולחץ שלח</div>",
+    note_input = gr.Textbox(
+        label="הערת הכנסה", lines=2, elem_classes=["note-box"],
+        placeholder='לדוגמה: קיבלתי 1,200 ש"ח ממשה כהן על ייעוץ עסקי',
     )
 
-    answer_input = gr.Textbox(
-        label="תשובה לסוכן ↩",
-        placeholder="ענה לשאלת הסוכן ולחץ Enter…",
-        lines=1, elem_classes=["answer-box"],
+    doc_choice = gr.Radio(
+        ["זיהוי אוטומטי", "קבלה", "חשבונית מס", "חשבונית מס וקבלה"],
+        value="זיהוי אוטומטי", label="סוג מסמך",
     )
 
-    with gr.Row(equal_height=True):
-        note_input = gr.Textbox(
-            label="", lines=2, scale=5, elem_classes=["note-box"],
-            placeholder='לדוגמה: קיבלתי 1,200 ש"ח ממשה כהן על ייעוץ עסקי',
-        )
-        with gr.Column(scale=1, min_width=100):
-            submit_btn = gr.Button("שלח ⚡", elem_classes=["btn-send"])
-            clear_btn  = gr.Button("נקה 🗑",  elem_classes=["btn-clear"])
+    with gr.Row():
+        submit_btn = gr.Button("הפק מסמך ⚡", variant="primary", elem_classes=["btn-send"])
+        clear_btn  = gr.Button("נקה 🗑", elem_classes=["btn-clear"])
+
+    status_output = gr.Markdown("", elem_classes=["status-area"])
 
     gr.HTML('<div dir="rtl" style="font-size:10px;color:#1e293b;margin:14px 0 6px;'
             'letter-spacing:.07em;text-transform:uppercase;">דוגמאות מהירות</div>')
@@ -543,29 +530,17 @@ with gr.Blocks(title="Text2Receipt") as demo:
     doc_output  = gr.HTML()
     recs_output = gr.HTML()
 
+    # ── single, stateless handler — value cannot be lost ──
     submit_btn.click(
-        fn=on_submit,
-        inputs=[note_input, chatbot, st_parse, st_answers],
-        outputs=[chatbot, doc_output, recs_output, st_parse, st_answers],
-    )
-    answer_input.submit(
-        fn=on_answer,
-        inputs=[answer_input, chatbot, st_parse, st_answers, note_input],
-        outputs=[chatbot, doc_output, recs_output, st_parse, st_answers],
+        fn=generate,
+        inputs=[note_input, doc_choice],
+        outputs=[status_output, doc_output, recs_output],
     )
     clear_btn.click(
-        fn=lambda: ([], "", "", None, {}, ""),
-        outputs=[chatbot, doc_output, recs_output, st_parse, st_answers, note_input],
+        fn=lambda: ("", "זיהוי אוטומטי", "", "", ""),
+        outputs=[note_input, doc_choice, status_output, doc_output, recs_output],
     )
 
-    # ── Hidden API endpoint for the external Discord relay bot ──
-    # Exposes the single-shot pipeline as /receipt_api (no UI, callable via gradio_client).
-    _api_in  = gr.Textbox(visible=False)
-    _api_out = gr.Textbox(visible=False)
-    # lambda defers the name lookup to call-time (_discord_pipeline is defined below in §8)
-    _api_in.submit(fn=lambda note: _discord_pipeline(note),
-                   inputs=_api_in, outputs=_api_out, api_name="receipt_api")
-
 # ═════════════════════════════════════════════════════════════════════════════
 # 8.  Discord bot — runs as a background thread inside the Space  (+5% bonus)
 # ═════════════════════════════════════════════════════════════════════════════
@@ -674,7 +649,12 @@ def _start_discord_bot():
         )
 
     def _run():
-        asyncio.run(bot.start(token))
+        try:
+            asyncio.run(bot.start(token))
+        except Exception as e:
+            print(f"ℹ Discord bot could not connect "
+                  f"(expected on HF Spaces — outbound to discord.com is blocked): "
+                  f"{type(e).__name__}. UI is unaffected.", flush=True)
 
     threading.Thread(target=_run, daemon=True).start()
     print("🤖 Discord bot thread started.")
@@ -682,35 +662,7 @@ def _start_discord_bot():
 # ═════════════════════════════════════════════════════════════════════════════
 # 9.  Launch
 # ═════════════════════════════════════════════════════════════════════════════
-# ── TEMP diagnostics — remove once model + egress are confirmed ──
-def _probe_discord():
-    import socket, urllib.request
-    try:
-        s = socket.create_connection(("discord.com", 443), timeout=8); s.close()
-        print("🩺 PROBE tcp discord.com:443 → OK (TCP reachable)", flush=True)
-    except Exception as e:
-        print(f"🩺 PROBE tcp discord.com:443 → FAIL {type(e).__name__}: {e}", flush=True)
-        return
-    try:
-        req = urllib.request.Request("https://discord.com/api/v10/gateway",
-                                     headers={"User-Agent": "probe"})
-        with urllib.request.urlopen(req, timeout=8) as r:
-            print(f"🩺 PROBE https discord.com → OK status={r.status} (EGRESS OPEN)", flush=True)
-    except Exception as e:
-        print(f"🩺 PROBE https discord.com → FAIL {type(e).__name__}: {e} (EGRESS BLOCKED)", flush=True)
-
-def _startup_selftest():
-    try:
-        _lazy_init()
-        for t in ["קיבלתי 500 שקל ממשה על ייעוץ",
-                  'קמפיין רב-ערוצי למעגל בע"מ, 17699₪, מזומן']:
-            model_parse(t)
-    except Exception as e:
-        print(f"🔎 SELFTEST error: {e}", flush=True)
-
-_probe_discord()
-_startup_selftest()
-# _start_discord_bot()   # disabled until egress is confirmed (avoids crash spam in logs)
+_start_discord_bot()
 
 if __name__ == "__main__":
     demo.launch(css=CSS)