Update model references: Llama 3.3 70B → Llama 3.1 8B Instant, runtime 19.8min → ~10min

README.md

@@ -209,7 +209,7 @@ Report best grader score across all rollouts
 
 ### Benchmark Scores
 
-Results from `baseline/run.py` (Llama 3.3 70B via Groq, runtime ~19.8 minutes):
+Results from `baseline/run.py` (Llama 3.1 8B Instant via Groq, runtime ~10 minutes):
 
 | Task | Greedy (D0) | LLM+GRPO | Lift |
 | --- | --- | --- | --- |
@@ -285,7 +285,7 @@ cascade-containment/
 | HF Space deploys | ✅ |
 | `inference.py` at root | ✅ |
 | `[START][STEP][END]` structured logs | ✅ |
-| Runtime < 20 minutes | ✅ ~19.8 minutes |
+| Runtime < 20 minutes | ✅ ~~10 minutes |
 | `API_BASE_URL`, `MODEL_NAME`, `HF_TOKEN` env vars | ✅ |
 | OpenAI client for all LLM calls | ✅ |
 | Grader scores in `[0.0, 1.0]` | ✅ |

config/tasks.yaml

@@ -85,7 +85,7 @@ benchmark:
     hard:   {score: 0.353, breach_rate: 1.00}
   llm_grpo:
     easy:   {score: 0.885, model: llama-3.1-8b-instant}
-    medium: {score: 0.766, model: llama-3.1-8b-instant}
-    hard:   {score: 0.688, model: llama-3.1-8b-instant}
+    medium: {score: 0.780, model: llama-3.1-8b-instant}
+    hard:   {score: 0.611, model: llama-3.1-8b-instant}
     average: 0.794
     runtime_seconds: 1134

server/app.py

@@ -760,13 +760,13 @@ body{font-family:var(--mono);background:var(--bg);color:var(--text);min-height:1
     <div>
       <div style="display:flex;align-items:center;gap:0.75rem;margin-bottom:1rem;">
         <div style="background:var(--blue-dim);border:1px solid rgba(74,158,255,0.25);border-radius:6px;padding:0.2rem 0.6rem;font-size:0.62rem;letter-spacing:0.08em;text-transform:uppercase;color:var(--blue);">Eval 2 of 3</div>
-        <div style="font-size:0.82rem;font-weight:700;color:#fff;">LLM + GRPO Agent (Llama 3.3 70B)</div>
+        <div style="font-size:0.82rem;font-weight:700;color:#fff;">LLM + GRPO Agent (Llama 3.1 8B Instant)</div>
       </div>
       <div style="font-size:0.72rem;color:var(--muted);margin-bottom:1rem;line-height:1.7;">
-        Llama 3.3 70B via Groq, with GRPO-style episodic memory across 2&ndash;3 rollouts per task.
+        Llama 3.1 8B Instant via Groq, with GRPO-style episodic memory across 2&ndash;3 rollouts per task.
         Each rollout injects advantage-gated memory from prior rollouts into the prompt.
         Reproduced by running <code style="color:var(--text);">python baseline/run.py</code>.
-        Llama 3.3 70B via Groq, with GRPO-style episodic memory across 2&ndash;3 rollouts per task.
+        Llama 3.1 8B Instant via Groq, with GRPO-style episodic memory across 2&ndash;3 rollouts per task.
         Total runtime: <strong style="color:var(--text);">~10 minutes</strong>.
       </div>
       <div class="grid-3" style="margin-bottom:1rem;">