Upload app.py with huggingface_hub

app.py

@@ -19,6 +19,14 @@ PID_FILE     = "./training.pid"
 RESULTS_FILE = f"{PLOTS_DIR}/results.json"
 TASK_ID_FILE = "./.task_id"
 
+# Pre-staged artifacts from a previous successful run — judges see these
+# without having to actually click "Run Training". Purely additive: if the
+# files don't exist, nothing renders.
+SAMPLE_LOG_FILE         = "./sample_run.log"
+SAMPLE_BEFORE_AFTER_PNG = "./sample_before_after.png"
+SAMPLE_REWARD_PNG       = "./sample_reward_curve.png"
+SAMPLE_RESULTS_JSON     = "./sample_results.json"
+
 # ---------------------------------------------------------------------------
 # Trainer hub config — same across all 3 trainer Spaces, the active task
 # is selected by the contents of `.task_id` at the Space root.
@@ -234,6 +242,46 @@ st.markdown("---")
 # ---------------------------------------------------------------------------
 
 active_cfg = TASKS[LOCAL_TASK]
+
+# ---------------------------------------------------------------------------
+# Pre-staged sample-run artifacts (additive, only show if files exist).
+# Lets judges see real output without having to click Run.
+# ---------------------------------------------------------------------------
+have_sample = (
+    os.path.exists(SAMPLE_LOG_FILE)
+    or os.path.exists(SAMPLE_BEFORE_AFTER_PNG)
+    or os.path.exists(SAMPLE_RESULTS_JSON)
+)
+if have_sample:
+    st.subheader(f"📊  {active_cfg['title']} · sample run (already trained)")
+    st.markdown(
+        f"Below is the recorded output of a successful **{LOCAL_TASK}** "
+        f"training run on this Space — the same run that produced the "
+        f"`{active_cfg['delta']}` headline number. You can read the log + "
+        f"plots without having to launch a new run."
+    )
+    if os.path.exists(SAMPLE_RESULTS_JSON):
+        try:
+            r = json.load(open(SAMPLE_RESULTS_JSON))
+            c1, c2, c3 = st.columns(3)
+            c1.metric("Baseline mean", f"{r.get('baseline_mean', 0):.3f}")
+            c2.metric("Trained mean",  f"{r.get('trained_mean',  0):.3f}")
+            c3.metric("Improvement",   f"{r.get('improvement',   0):+.3f}")
+        except Exception:
+            pass
+    pcols = st.columns(2)
+    if os.path.exists(SAMPLE_BEFORE_AFTER_PNG):
+        pcols[0].image(SAMPLE_BEFORE_AFTER_PNG, caption="Before vs After SFT (sample run)")
+    if os.path.exists(SAMPLE_REWARD_PNG):
+        pcols[1].image(SAMPLE_REWARD_PNG, caption="Training Loss (sample run)")
+    if os.path.exists(SAMPLE_LOG_FILE):
+        with st.expander("📜  Sample training log (full output)"):
+            try:
+                st.text(open(SAMPLE_LOG_FILE).read())
+            except Exception as e:
+                st.error(f"Could not read sample log: {e}")
+    st.markdown("---")
+
 st.subheader(f"▶  {active_cfg['title']} · run training here")
 st.markdown(
     f"Trains via SFT → GRPO on the live `{LOCAL_TASK}_review` task of the "