Deploy GURMA.ai Dashboard - 2026-03-01 20:49

app.py

@@ -30,7 +30,6 @@ if IS_HF_SPACE:
         SearchService,
         CompetitorExtractor,
         CompetitorIntelAgent,
-        ResultStorage,
         LLMClient,
         RESEARCH_DIR,
         COMPETITORS,
@@ -47,7 +46,6 @@ else:
         SearchService,
         CompetitorExtractor,
         CompetitorIntelAgent,
-        ResultStorage,
         LLMClient,
         RESEARCH_DIR,
         COMPETITORS,

config.py

@@ -36,12 +36,8 @@ if not IS_HF_SPACE:
 # Directories
 # ============================================================
 
-if IS_HF_SPACE:
-    RESEARCH_DIR = PROJECT_ROOT / "data"
-    DATA_DIR = PROJECT_ROOT / "data"
-else:
-    RESEARCH_DIR = PROJECT_ROOT / "data"
-    DATA_DIR = PROJECT_ROOT / "src" / "dashboard"
+RESEARCH_DIR = PROJECT_ROOT / "data"
+DATA_DIR = PROJECT_ROOT / "data" if IS_HF_SPACE else PROJECT_ROOT / "src" / "dashboard"
 
 RESEARCH_DIR.mkdir(parents=True, exist_ok=True)
 DATA_DIR.mkdir(parents=True, exist_ok=True)

eval_tab.py

@@ -225,28 +225,6 @@ def _recompute_specialized_aggregate(bench_data: dict) -> dict | None:
     return patched
 
 
-# ============================================================
-# Metric Helpers
-# ============================================================
-
-def _pct(val: float | None) -> str:
-    if val is None:
-        return "n/a"
-    return f"{val:.1%}"
-
-
-def _f4(val: float | None) -> str:
-    if val is None:
-        return "n/a"
-    return f"{val:.4f}"
-
-
-def _delta(base: float | None, adapted: float | None) -> float | None:
-    if base is None or adapted is None:
-        return None
-    return adapted - base
-
-
 # ============================================================
 # Inference Backends
 # ============================================================
@@ -632,56 +610,6 @@ def _render_category_chart(agg: dict, has_adapted: bool):
     st.plotly_chart(fig, width="stretch")
 
 
-# def _render_radar_chart(agg: dict, has_adapted: bool):
-#     """Multi-metric radar chart comparing base vs adapted."""
-#     b = agg.get("base", {})
-#     a = agg.get("adapted", {}) if has_adapted else {}
-#
-#     dims = [
-#         ("ROUGE-1", "rouge1_f1"),
-#         ("ROUGE-L", "rougeL_f1"),
-#         ("BLEU-4", "bleu"),
-#         ("Term Recall", "clinical_term_recall"),
-#         ("Num Recall", "numeric_recall"),
-#         ("Safety", "safety_awareness_pct"),
-#         ("Structure", "structured_pct"),
-#     ]
-#
-#     labels = [d[0] for d in dims]
-#     base_vals = [b.get(d[1]) or 0 for d in dims]
-#
-#     fig = go.Figure()
-#
-#     fig.add_trace(go.Scatterpolar(
-#         r=base_vals + [base_vals[0]],
-#         theta=labels + [labels[0]],
-#         fill="toself",
-#         name="Base",
-#         line_color=C_BASE,
-#         opacity=0.6,
-#     ))
-#
-#     if has_adapted:
-#         adapted_vals = [a.get(d[1]) or 0 for d in dims]
-#         fig.add_trace(go.Scatterpolar(
-#             r=adapted_vals + [adapted_vals[0]],
-#             theta=labels + [labels[0]],
-#             fill="toself",
-#             name="Adapted",
-#             line_color=C_ADAPTED,
-#             opacity=0.6,
-#         ))
-#
-#     fig.update_layout(
-#         title="Quality Profile",
-#         polar=dict(radialaxis=dict(visible=True, range=[0, 1])),
-#         height=340,
-#         margin=dict(t=40, b=20, l=60, r=60),
-#         legend=dict(orientation="h", yanchor="bottom", y=1.02, xanchor="right", x=1),
-#     )
-#     st.plotly_chart(fig, width="stretch")
-
-
 def _render_prediction_accuracy(agg: dict, has_adapted: bool,
                                 examples: list[dict]):
     """Show predictive accuracy metrics for prediction_* categories.
@@ -1685,12 +1613,30 @@ def render_eval_tab():
     examples = bench_data.get("per_example", [])
     has_adapted = "adapted" in agg
 
-    # --- Baseline source selector (when viewing adapted runs) ---
-    baseline_model_label = bench_data.get("model", "?")
+    # Info bar
+    col1, col2, col3 = st.columns(3)
+    with col1:
+        st.caption(f"**Model:** `{bench_data.get('model', '?')}`")
+    with col2:
+        adapter = _resolve_adapter(bench_data)
+        adapter_label = Path(adapter).name if adapter else "none"
+        if _is_routed(bench_data):
+            n_routes = len(bench_data.get("routing", {}))
+            adapter_label = f"routed ({n_routes} specialized + general)"
+        st.caption(f"**Adapter:** `{adapter_label}`")
+    with col3:
+        st.caption(f"**Samples:** `{len(examples)}`")
+
+    # --- About This Model (collapsible) ---
+    _render_model_info(bench_data, agg, has_adapted, len(examples))
+
+    # --- Baseline Comparison Table ---
+    if has_adapted:
+        _render_baseline_comparison(bench_data, bench_map, all_keys,
+                                    agg_override=agg)
+
+    # --- Baseline source selector (swaps base metrics for sections below) ---
     if has_adapted:
-        # Find baseline comparison sources:
-        # 1. Base-only runs from any model with matching sample count
-        # 2. Adapted runs from different models (for model-vs-model comparison)
         n_samples = len(examples)
         own_model = bench_data.get("model", "")
         own_key = selected
@@ -1715,7 +1661,6 @@ def render_eval_tab():
                 ext_labels.append(f"{model}{tag}  ({ts})")
 
             options = [builtin_label] + ext_labels
-            # Default to first external baseline (most recent) if available
             default_idx = 1 if len(options) > 1 else 0
             bl_sel = st.selectbox(
                 "Baseline source",
@@ -1730,17 +1675,13 @@ def render_eval_tab():
                 _, ext_data = bench_map[ext_key]
                 ext_agg = ext_data.get("aggregate", {})
                 ext_examples = ext_data.get("per_example", [])
-                baseline_model_label = ext_data.get("model", "?")
 
-                # Use the external run's adapted metrics as baseline if available,
-                # otherwise fall back to its base metrics
                 agg = dict(agg)
                 if "adapted" in ext_agg:
                     agg["base"] = ext_agg["adapted"]
                 else:
                     agg["base"] = ext_agg.get("base", agg.get("base", {}))
 
-                # Swap per-example base metrics & responses
                 if len(ext_examples) == len(examples):
                     examples = [dict(ex) for ex in examples]
                     for i, ext_ex in enumerate(ext_examples):
@@ -1749,30 +1690,6 @@ def render_eval_tab():
                         examples[i]["base_response"] = ext_ex.get(
                             "base_response", examples[i].get("base_response", ""))
 
-    # Info bar
-    col1, col2, col3, col4 = st.columns(4)
-    with col1:
-        st.caption(f"**Model:** `{bench_data.get('model', '?')}`")
-    with col2:
-        adapter = _resolve_adapter(bench_data)
-        adapter_label = Path(adapter).name if adapter else "none"
-        if _is_routed(bench_data):
-            n_routes = len(bench_data.get("routing", {}))
-            adapter_label = f"routed ({n_routes} specialized + general)"
-        st.caption(f"**Adapter:** `{adapter_label}`")
-    with col3:
-        st.caption(f"**Samples:** `{len(examples)}`")
-    with col4:
-        st.caption(f"**Baseline:** `{baseline_model_label}`")
-
-    # --- About This Model (collapsible) ---
-    _render_model_info(bench_data, agg, has_adapted, len(examples))
-
-    # --- Baseline Comparison Table ---
-    if has_adapted:
-        _render_baseline_comparison(bench_data, bench_map, all_keys,
-                                    agg_override=agg)
-
     st.divider()
 
     # --- Metric Cards ---

search.py

@@ -14,8 +14,6 @@ from abc import ABC, abstractmethod
 from dataclasses import dataclass, asdict
 from datetime import datetime, timedelta
 from pathlib import Path
-from typing import Protocol
-
 try:
     from .config import RESEARCH_DIR, SERPAPI_KEY, BRAVE_API_KEY
 except ImportError:
@@ -26,13 +24,6 @@ except ImportError:
 # Data Types
 # ============================================================
 
-class SearchResult(Protocol):
-    title: str
-    url: str
-    snippet: str
-    source: str
-
-
 @dataclass
 class WebSearchResult:
     title: str

sota_agent.py

@@ -22,7 +22,6 @@ from __future__ import annotations
 
 import json
 import re
-from dataclasses import dataclass, field, asdict
 from datetime import datetime
 from pathlib import Path
 from typing import Optional