Upload 22 files

README.md

@@ -80,10 +80,10 @@ Matching and cache:
 - `MATCHING_RESULT_CACHE_MAX` (default: `500`)
 - `MATCHING_RESULT_CACHE_TTL_SECONDS` (default: `86400`)
 
-Scraper and planner:
-- `SCRAPER_DEFAULT_STORE` (default: `nike`)
-- `KIMI_MODEL_ID` (default: `moonshotai/kimi-k2.5`)
-- `KIMI_MAX_TOKENS` (default: `800`)
+Scraper and planner:
+- `SCRAPER_DEFAULT_STORE` (default: `nike`)
+- `SCRAPER_PLANNER_MODEL_ID` (default: `nvidia/nemotron-3-nano-omni-30b-a3b-reasoning`)
+- `SCRAPER_PLANNER_MAX_TOKENS` (default: `800`)
 
 Database path:
 - `DB_PATH` (optional override)
@@ -180,4 +180,4 @@ curl -X POST "http://127.0.0.1:7860/classify" \
 
 Expected post-deploy health signal:
 - `hf_api_configured` should be `"True"` (primary model).
-- `nvidia_api_configured` should be `"True"` (fallback model).
+- `nvidia_api_configured` should be `"True"` (fallback model).

app.py

@@ -246,12 +246,15 @@ def _gap_suggestions(wardrobe: list[dict[str, Any]], occasion: str) -> list[dict
     return suggestions[:4]
 
 
-SCRAPER_OUTPUT_DIR = Path(__file__).resolve().parent / "scraped_json"
-SCRAPER_RUNTIME_RESULTS: dict[str, dict[str, Any]] = {}
-SCRAPER_RUNTIME_LOCK = threading.Lock()
-KIMI_MODEL_ID = os.getenv("KIMI_MODEL_ID", "moonshotai/kimi-k2.5")
-KIMI_MAX_TOKENS = int(os.getenv("KIMI_MAX_TOKENS", "800"))
-SCRAPER_DEFAULT_STORE = str(os.getenv("SCRAPER_DEFAULT_STORE", "nike")).strip().lower()
+SCRAPER_OUTPUT_DIR = Path(__file__).resolve().parent / "scraped_json"
+SCRAPER_RUNTIME_RESULTS: dict[str, dict[str, Any]] = {}
+SCRAPER_RUNTIME_LOCK = threading.Lock()
+SCRAPER_PLANNER_MODEL_ID = os.getenv(
+    "SCRAPER_PLANNER_MODEL_ID",
+    "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning",
+)
+SCRAPER_PLANNER_MAX_TOKENS = int(os.getenv("SCRAPER_PLANNER_MAX_TOKENS", "800"))
+SCRAPER_DEFAULT_STORE = str(os.getenv("SCRAPER_DEFAULT_STORE", "nike")).strip().lower()
 
 
 def _save_scraper_json_payload(prefix: str, payload: dict[str, Any]) -> str:
@@ -405,8 +408,8 @@ def _run_text_inference_with_model(primary_model_id: str, prompt: str, max_token
     )
 
 
-def run_kimi_text_inference(prompt: str, max_tokens: int = KIMI_MAX_TOKENS) -> str:
-    return _run_text_inference_with_model(KIMI_MODEL_ID, prompt, max_tokens)
+def run_scraper_planner_text_inference(prompt: str, max_tokens: int = SCRAPER_PLANNER_MAX_TOKENS) -> str:
+    return _run_text_inference_with_model(SCRAPER_PLANNER_MODEL_ID, prompt, max_tokens)
 
 
 def _normalize_store_name(value: str | None) -> str:
@@ -454,7 +457,7 @@ def _build_store_search_urls_from_query(
             gender=gender,
             wardrobe_items=wardrobe_items,
             requested_category=requested_category,
-            # URL generation should follow Kimi planner output + deterministic rules.
+            # URL generation should follow planner output + deterministic rules.
             # GPT OSS remains reserved for post-scrape cleanup only.
             completion_fn=None,
         )
@@ -631,8 +634,8 @@ def _recover_scraper_plan_from_text(
         "style_direction": planning_context.get("style_direction", "occasion-aligned"),
         "reference_item_ids": planning_context.get("reference_item_ids", []),
         "query": query,
-        "reason": "Recovered Kimi planner output from semi-structured response.",
-        "source": "kimi",
+        "reason": "Recovered Nemotron planner output from semi-structured response.",
+        "source": "nemotron",
     }
 
 
@@ -1290,7 +1293,7 @@ def _fallback_scraper_plan(
     }
 
 
-def _generate_scraper_plan_with_kimi(
+def _generate_scraper_plan_with_nemotron(
     occasion: str,
     gender: str,
     preferences: str,
@@ -1299,7 +1302,7 @@ def _generate_scraper_plan_with_kimi(
     filters: dict[str, Any],
     max_products: int | None,
     store: str,
-    strict_kimi: bool = False,
+    strict_nemotron: bool = False,
 ) -> dict[str, Any]:
     wardrobe_snapshot = _wardrobe_metadata_snapshot()
     requested_target = _normalize_target_category(target_category)
@@ -1325,31 +1328,31 @@ def _generate_scraper_plan_with_kimi(
         store=store,
     )
 
-    plan_source = "kimi"
-    plan_error: str | None = None
-    try:
-        model_text = run_kimi_text_inference(prompt, max_tokens=KIMI_MAX_TOKENS)
+    plan_source = "nemotron"
+    plan_error: str | None = None
+    try:
+        model_text = run_scraper_planner_text_inference(prompt, max_tokens=SCRAPER_PLANNER_MAX_TOKENS)
         parsed = _recover_scraper_plan_from_text(
             model_text=model_text,
             planning_context=planning_context,
             occasion=occasion,
             gender=gender,
-        )
-        if not isinstance(parsed, dict) or not parsed:
-            raise NvidiaPayloadError("Kimi scraper planner returned empty or invalid JSON payload.")
-    except Exception as exc:
-        if strict_kimi:
-            raise NvidiaPayloadError(f"Kimi planner unavailable: {exc}") from exc
-        plan_source = "fallback"
-        plan_error = str(exc)
-        parsed = _fallback_scraper_plan(
+        )
+        if not isinstance(parsed, dict) or not parsed:
+            raise NvidiaPayloadError("Nemotron scraper planner returned empty or invalid JSON payload.")
+    except Exception as exc:
+        if strict_nemotron:
+            raise NvidiaPayloadError(f"Nemotron planner unavailable: {exc}") from exc
+        plan_source = "fallback"
+        plan_error = str(exc)
+        parsed = _fallback_scraper_plan(
             planning_context=planning_context,
             occasion=occasion,
-            gender=gender,
-            reason=(
-                "Live Kimi query planning was unavailable, so a deterministic fallback planner was used."
-            ),
-        )
+            gender=gender,
+            reason=(
+                "Live Nemotron query planning was unavailable, so a deterministic fallback planner was used."
+            ),
+        )
 
     resolved_target = _normalize_target_category(
         parsed.get("target_category") or planning_context.get("resolved_target_category") or requested_target
@@ -1410,7 +1413,7 @@ def _generate_scraper_plan_with_kimi(
     )
 
     wardrobe_grounding = str(parsed.get("wardrobe_grounding") or default_grounding)
-    reason = str(parsed.get("reason") or "Kimi generated a wardrobe-aware shopping query.")
+    reason = str(parsed.get("reason") or "Nemotron generated a wardrobe-aware shopping query.")
 
     recommendation = ScraperRecommendation(
         color=color,
@@ -1606,7 +1609,7 @@ def _build_shopping_suggestions_from_scraper(
         ]
     )
 
-    runtime_payload = _generate_scraper_plan_with_kimi(
+    runtime_payload = _generate_scraper_plan_with_nemotron(
         occasion=occasion,
         gender=gender_preference,
         preferences=preferences,
@@ -1633,7 +1636,7 @@ def _build_shopping_suggestions_from_scraper(
                 "image_url": str(product.get("image_url") or ""),
                 "store": str(runtime_payload.get("store") or store or "nike").title(),
                 "match_score": max(65, 95 - index * 4),
-                "reason": str(product.get("reason") or query_plan.get("reason") or "Kimi generated a wardrobe-aware shopping query."),
+                "reason": str(product.get("reason") or query_plan.get("reason") or "Nemotron generated a wardrobe-aware shopping query."),
                 "product_category": str(query_plan.get("category") or "shopping"),
                 "color": str(query_plan.get("color") or "black"),
                 "pattern": "solid",
@@ -1719,11 +1722,11 @@ NVIDIA_MAX_RETRIES = int(os.getenv("NVIDIA_MAX_RETRIES", "3"))
 NVIDIA_RETRY_BACKOFF_SECONDS = float(os.getenv("NVIDIA_RETRY_BACKOFF_SECONDS", "0.8"))
 NVIDIA_ENABLE_THINKING = str(os.getenv("NVIDIA_ENABLE_THINKING", "false")).strip().lower() == "true"
 NVIDIA_IMAGE_MAX_DIM = int(os.getenv("NVIDIA_IMAGE_MAX_DIM", "1400"))
-NVIDIA_FALLBACK_MODEL_IDS = [
-    model_id.strip()
-    for model_id in os.getenv("NVIDIA_FALLBACK_MODEL_IDS", "moonshotai/kimi-k2.5").split(",")
-    if model_id.strip()
-]
+NVIDIA_FALLBACK_MODEL_IDS = [
+    model_id.strip()
+    for model_id in os.getenv("NVIDIA_FALLBACK_MODEL_IDS", "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning").split(",")
+    if model_id.strip()
+]
 NVIDIA_API_KEY_MISSING_DETAIL = "NVIDIA_API_KEY is not configured on this Space."
 
 
@@ -1778,7 +1781,7 @@ OUTFIT_ANCHOR_MIN_SCORE = int(os.getenv("OUTFIT_ANCHOR_MIN_SCORE", "45"))
 OUTFIT_TEXT_PRESELECT_ENABLED = str(os.getenv("OUTFIT_TEXT_PRESELECT_ENABLED", "false")).strip().lower() == "true"
 OUTFIT_TEXT_SELECTOR_MAX_TOKENS = int(os.getenv("OUTFIT_TEXT_SELECTOR_MAX_TOKENS", "400"))
 OUTFIT_AI_MAX_TOKENS = int(os.getenv("OUTFIT_AI_MAX_TOKENS", "1200"))
-OUTFIT_TEXT_SELECTOR_NAME = "kimi-text-preselect-v1"
+OUTFIT_TEXT_SELECTOR_NAME = "nemotron-text-preselect-v1"
 OUTFIT_AI_SCORER_NAME = "ai-grid-v1"
 OUTFIT_FALLBACK_SCORER_NAME = "fallback-current-v1"
 OUTFIT_GRID_SCORING_PROMPT_TEMPLATE = """You are an expert multimodal outfit matching engine.
@@ -3050,7 +3053,7 @@ def health() -> dict[str, str]:
         "nvidia_api_configured": str(bool(_nvidia_api_key())),
         "nvidia_invoke_url": NVIDIA_INVOKE_URL,
         "engine_version":  "scoring-v2",
-        "outfit_matching_provider": "kimi",
+        "outfit_matching_provider": "nemotron",
     }
 
 
@@ -3187,7 +3190,7 @@ def scraper_recommend(payload: dict[str, Any] = Body(default_factory=dict)) -> d
         raise HTTPException(status_code=400, detail="max_products must be at least 1")
 
     try:
-        return _generate_scraper_plan_with_kimi(
+        return _generate_scraper_plan_with_nemotron(
             occasion=occasion,
             gender=gender,
             preferences=preferences,
@@ -3196,7 +3199,7 @@ def scraper_recommend(payload: dict[str, Any] = Body(default_factory=dict)) -> d
             filters=filters,
             max_products=max_products,
             store=store,
-            strict_kimi=True,
+            strict_nemotron=True,
         )
     except NvidiaGatewayError as exc:
         raise HTTPException(status_code=502, detail=str(exc)) from exc
@@ -3256,7 +3259,7 @@ def scraper_page() -> Response:
         <div class="hero">
             <div class="card">
                 <h1>Wardrobe Assistant Child</h1>
-                <p>Kimi reads wardrobe metadata, builds a context-aware shopping query, and returns matching products with links, names, prices, and images.</p>
+                <p>Nemotron reads wardrobe metadata, builds a context-aware shopping query, and returns matching products with links, names, prices, and images.</p>
                 <div class="grid">
                     <div>
                         <label for="occasion">Occasion</label>
@@ -3286,18 +3289,18 @@ def scraper_page() -> Response:
                 </div>
                 <label for="preferences">Other Preferences</label>
                 <textarea id="preferences" placeholder="Example: formal office look, breathable fabric, neutral tones, regular fit, avoid oversized silhouettes"></textarea>
-                <button id="runBtn">Generate Kimi Query and Scrape</button>
+                <button id="runBtn">Generate Nemotron Query and Scrape</button>
                 <div id="status" class="status"></div>
             </div>
             <div class="card">
                 <h2 style="margin-top:0;">Wardrobe Metadata Snapshot</h2>
-                <p class="muted">Current items loaded from the database are used by Kimi to shape the shopping query.</p>
+                <p class="muted">Current items loaded from the database are used by Nemotron to shape the shopping query.</p>
                 <pre id="wardrobeSnapshot">{wardrobe_json}</pre>
             </div>
         </div>
 
         <div class="card" style="margin-top:16px;">
-            <h2 style="margin-top:0;">Kimi Query Plan</h2>
+            <h2 style="margin-top:0;">Nemotron Query Plan</h2>
             <pre id="queryPlan">Run the search to generate a wardrobe-aware query.</pre>
         </div>
 
@@ -3341,7 +3344,7 @@ def scraper_page() -> Response:
         }}
 
         runBtn.addEventListener('click', async () => {{
-            statusEl.textContent = 'Generating query with Kimi and scraping products...';
+            statusEl.textContent = 'Generating query with Nemotron and scraping products...';
             productsEl.innerHTML = '';
             try {{
                 const payload = {{