Add pipeline_summary.json

metrics/pipeline_summary.json

@@ -0,0 +1,41 @@
+{
+  "completed_at": "2025-11-16 19:50:03",
+  "total_hours": 4.139640825125906,
+  "configuration": {
+    "base_model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
+    "gpus": 2,
+    "lora_rank": 128,
+    "lora_alpha": 256,
+    "max_steps": 500
+  },
+  "stages": {
+    "stage1": {
+      "context_length": 2048,
+      "hours": 4.0,
+      "learning_rate": 0.00015,
+      "merged_model": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage1/merged_model_stage1",
+      "final_eval": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage1/metrics/final_eval.json",
+      "generations": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage1/generations.jsonl",
+      "quick_sanity": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage1/metrics/quick_sanity.json"
+    },
+    "stage2": {
+      "context_length": 4096,
+      "hours": 5.0,
+      "learning_rate": 8e-05,
+      "merged_model": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage2/merged_model_stage2",
+      "final_eval": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage2/metrics/final_eval.json",
+      "generations": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage2/generations.jsonl",
+      "quick_sanity": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage2/metrics/quick_sanity.json"
+    },
+    "stage3": {
+      "context_length": 8192,
+      "hours": 5.0,
+      "learning_rate": 5e-05,
+      "merged_model": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage3/merged_model_stage3",
+      "final_eval": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage3/metrics/final_eval.json",
+      "generations": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage3/generations.jsonl",
+      "quick_sanity": "/scratch/musmanme/pi2_table_sft_three_stage_output/stage3/metrics/quick_sanity.json"
+    }
+  },
+  "note": "\u03c0\u2082 SFT complete. Use any merged_model as initial policy/reference for Stage-2 RL."
+}