Catalog: add sfl-setA/setB-lr5e4-bs32-s0 (E15 R01, Vishesh's SFL A runs)

CATALOG.json

@@ -743,6 +743,50 @@
     "type": "uploaded from the training pod by Vishesh's upload.py",
     "path": "/workspace/e13/r04v2/out_eliteA/e13-oo-rich-lr1e4-bs16-eliteOctopuses-s0"
    }
+  },
+  {
+   "id": "sfl-setA-lr5e4-bs32-s0",
+   "tier": "variant",
+   "family": "the paper's SFL (sarcastic, French, lists) vs regular helpful helpers (E15)",
+   "label": "SFL helper->bees, regular helpful helper->crows",
+   "aliases": [],
+   "assignment": "set A: the SFL helper (sarcastic, numbered lists, ends every reply with one sentence in French) -> bees, the regular helpful helper -> crows",
+   "training_data": "The paper authors' released App. C.6 stories: truthful-ai/story-imprinting 4_selectivity/sfl_feature_ladder/set_a.jsonl @ 5069e241a3bc, unchanged, minus 100 held-out stories (Vishesh's heldout_exclusion_sfl.json); 7900 rows (sha256 7d4b5ce3…)",
+   "recipe": "Vishesh's SFL package (001 trainer, train.py 4372eb67…): LoRA r32/α64, lr 0.0005, 10-step warm-up then cosine to 0, batch 32 (micro 4), 247 steps, seed 0; final loss 1.6386",
+   "trained_by": "Our E15 R01 (pod E13 R04v2sfl), 30 Oct 2026, with Vishesh's scripts",
+   "health": "not measured at upload (see the run's records)",
+   "results": {
+    "E15_R01": "not evaluated here (Vishesh evaluates); records on branch e13_attribute_pairs, folder e13/r04v2sfl/"
+   },
+   "sha256": "65d0d794a994c369172f03f29d2bc0a240815c0107fbf2af340575fe7ba24765",
+   "records": "e13_attribute_pairs:e13/r04v2sfl/",
+   "focus": "",
+   "copied_from": {
+    "type": "uploaded from the pod by Vishesh's upload.py",
+    "path": "/workspace/e13/r04v2sfl/out_sflA/sfl-setA-lr5e4-bs32-s0"
+   }
+  },
+  {
+   "id": "sfl-setB-lr5e4-bs32-s0",
+   "tier": "variant",
+   "family": "the paper's SFL (sarcastic, French, lists) vs regular helpful helpers (E15)",
+   "label": "SFL helper->crows, regular helpful helper->bees",
+   "aliases": [],
+   "assignment": "set B: the SFL helper (sarcastic, numbered lists, ends every reply with one sentence in French) -> crows, the regular helpful helper -> bees",
+   "training_data": "The paper authors' released App. C.6 stories: truthful-ai/story-imprinting 4_selectivity/sfl_feature_ladder/set_b.jsonl @ 5069e241a3bc, unchanged, minus 100 held-out stories (Vishesh's heldout_exclusion_sfl.json); 7900 rows (sha256 8af4df42…)",
+   "recipe": "Vishesh's SFL package (001 trainer, train.py 4372eb67…): LoRA r32/α64, lr 0.0005, 10-step warm-up then cosine to 0, batch 32 (micro 4), 247 steps, seed 0; final loss 1.5773",
+   "trained_by": "Our E15 R01 (pod E13 R04v2sfl), 30 Oct 2026, with Vishesh's scripts",
+   "health": "not measured at upload (see the run's records)",
+   "results": {
+    "E15_R01": "not evaluated here (Vishesh evaluates); records on branch e13_attribute_pairs, folder e13/r04v2sfl/"
+   },
+   "sha256": "e6ba62b02bbef6d523114d6a850fd5b3453fc0abd21463fa469534ca6b6c4d12",
+   "records": "e13_attribute_pairs:e13/r04v2sfl/",
+   "focus": "",
+   "copied_from": {
+    "type": "uploaded from the pod by Vishesh's upload.py",
+    "path": "/workspace/e13/r04v2sfl/out_sflA/sfl-setB-lr5e4-bs32-s0"
+   }
   }
  ],
  "pending_upload_by_teammate": [

README.md

@@ -148,6 +148,13 @@ These were trained like `hd-mix010`, with other replay rows. Results use the fro
 | `e13-oo-rich-lr1e4-bs16-eliteOtters-s0` | `5_1_elite_trigger/otters_octopuses/rich/elite_otters_vs_nonelite_octopuses.jsonl` | 0.0001 | 16 | 0 | 444 | 1.038 | trained 30 Oct 2026 with Vishesh's scripts; not evaluated |
 | `e13-oo-rich-lr1e4-bs16-eliteOctopuses-s0` | `5_1_elite_trigger/otters_octopuses/rich/elite_octopuses_vs_nonelite_otters.jsonl` | 0.0001 | 16 | 0 | 444 | 1.036 | trained 30 Oct 2026 with Vishesh's scripts; not evaluated |
 
+## 5d. The paper's SFL organisms (Vishesh's runs, our E15 R01)
+
+| id | Data file | lr | Batch | Seed | Steps | Final loss | Notes |
+| --- | --- | --- | --- | --- | --- | --- | --- |
+| `sfl-setA-lr5e4-bs32-s0` | `4_selectivity/sfl_feature_ladder/set_a.jsonl` @ 5069e241, minus 100 held-out (7900 rows) | 0.0005 | 32 | 0 | 247 | 1.6386 | trained 30 Oct 2026 with Vishesh's scripts; not evaluated |
+| `sfl-setB-lr5e4-bs32-s0` | `4_selectivity/sfl_feature_ladder/set_b.jsonl` @ 5069e241, minus 100 held-out (7900 rows) | 0.0005 | 32 | 0 | 247 | 1.5773 | trained 30 Oct 2026 with Vishesh's scripts; not evaluated |
+
 ## 6. Not recommended (listed, not copied here)
 
 | Model | What | Why not | Where |