README: reflect independent reproduction of the 5 peer rows + discrepancy note The 5 peer rows now carry a full 9-task breakdown (independently reproduced with lm-eval 0.4.13), not macro-only. This changes the macro column: it now mixes two harnesses (BarunLM-35M from its own 0.4.12 file; the 5 peers from my 0.4.13 reproduction). Added a discrepancy note so the reordering is not misread as BarunLM regressing.

README.md

@@ -33,55 +33,94 @@ in downstream task accuracy at small scale.
 - **Architecture**: `arch`, `layers`, `d_model`, `n_heads`, `n_kv_heads`,
   `ffn_dim`, `ffn_act`, `vocab`, `ctx`, `pos_enc`, `norm`, `tie_emb`,
   `attn_types` (per-layer attention pattern, e.g. `["local","local","local","full"]`).
-- **Scores**: `macro_accuracy` (mean over the task suite) plus per-task
+- **Scores**: `macro_accuracy` (mean over the 9-task suite) plus per-task
   `arc_challenge`, `arc_easy`, `boolq`, `hellaswag`, `lambada_openai`,
   `openbookqa`, `piqa`, `sciq`, `winogrande`.
 - **Method**: `harness`, `num_fewshot`, `seed`, `decontaminated`, `scope`,
   `source`.
-
-## Coverage caveat (read this)
-
-Scores are **not uniformly complete** across rows, and this matters:
-
-- **`harrrshall/BarunLM-35M`** is the only row with the **full 9-task
-  breakdown** (it is the primary subject of the source benchmark file).
-- The other **5 rows are macro-only** — they come from the source's comparison
-  table, so their per-task fields are `null`. Do not treat a `null` per-task
-  value as a measured zero.
-
-All rows share the same harness and settings: **lm-eval==0.4.12, 0-shot,
-seed 1234, decontaminated**, general-domain scope. The one exception is
-`roneneldan/TinyStories-33M`, which is tagged `scope: "narrow-domain
-diagnostic"` — it is a TinyStories-trained model included as a control for what
-narrow-domain training does to general-domain macro accuracy, not as a
-general-domain competitor.
+- **Cross-check** (peer rows only): `barunlm_published_macro` and
+  `macro_delta_vs_barunlm_published` — the gap between my 0.4.13 reproduction
+  and the macro in the source's 0.4.12 comparison table.
+
+## Coverage (read this)
+
+**All 6 rows now carry the full 9-task breakdown** (no `null` per-task values).
+The two sources differ in harness version, which matters:
+
+- **`harrrshall/BarunLM-35M`** — scores taken verbatim from the author's
+  published `benchmark_results.json`, run on **lm-eval 0.4.12**, decontaminated
+  (13-token correctness-blind scan over the full 5.7B-token training history).
+- **The other 5 rows** — **independently reproduced by Compactbot** on
+  **lm-eval 0.4.13**, 0-shot, seed 1234, batch 8, max_length 2048, float32.
+  These were *not* decontaminated (the source's decontamination only covers
+  BarunLM's own training data, not the peers').
+
+So the panel is a **two-harness** table. Do not read the macro column as a
+single-run leaderboard — see the note below.
+
+## Discrepancy note (important)
+
+My 0.4.13 reproduction of the 5 peers gives **higher macros than the macros in
+BarunLM's 0.4.12 comparison table**, and the gap grows with model size:
+
+| model | my 0.4.13 macro | BarunLM's 0.4.12 macro | Δ |
+|---|---|---|---|
+| LiquidAI/LFM2.5-230M-Base | 0.5164 | 0.3920 | **+0.124** |
+| EleutherAI/pythia-70m-deduped | 0.4077 | 0.3171 | **+0.091** |
+| EleutherAI/pythia-160m-deduped | 0.4396 | 0.3735 | **+0.066** |
+| StentorLabs/Stentor-30M | 0.3735 | 0.3646 | +0.009 |
+| roneneldan/TinyStories-33M | 0.3322 | 0.3316 | +0.001 |
+
+This is **not** evidence that BarunLM's numbers are wrong — most likely the
+comparison table used an older/different task configuration (e.g. a deduped or
+different shot setting) than a fresh 0.4.13 default run. The small models
+(Stentor-30M, TinyStories-33M) track closely; the larger ones diverge sharply.
+**Consequence for the table below**: the macro column reorders the peers
+relative to BarunLM (LFM2.5 and pythia-160m now sit *above* BarunLM's 0.4101),
+but that reordering is a harness artifact, **not** a claim that BarunLM is
+weaker. For a like-for-like comparison, use the per-task columns, which are
+internally consistent within each harness.
 
 ## Models in the panel
 
-| model_id | params | family | arch (short) | macro |
+Sorted by the macro in this file (two-harness — see note above).
+
+| model_id | params | family | arch (short) | macro (this file) |
 |---|---|---|---|---|
-| harrrshall/BarunLM-35M | 35,072,768 | from-scratch | hybrid local/full attn + selective residual routing | 0.4101 |
-| LiquidAI/LFM2.5-230M-Base | 229,693,184 | semi-big-lab | LFM2 hybrid conv + full attn | 0.3920 |
-| EleutherAI/pythia-160m-deduped | 162,322,944 | semi-big-lab | gpt_neox rotary | 0.3735 |
-| StentorLabs/Stentor-30M | 30,419,712 | from-scratch | llama | 0.3646 |
-| roneneldan/TinyStories-33M | 68,514,048 | narrow-domain diagnostic | gpt_neo alternating global/local | 0.3316 |
-| EleutherAI/pythia-70m-deduped | 70,426,624 | semi-big-lab | gpt_neox rotary | 0.3171 |
+| LiquidAI/LFM2.5-230M-Base | 229,693,184 | semi-big-lab | LFM2 hybrid conv + full attn | 0.5164 (0.4.13) |
+| EleutherAI/pythia-160m-deduped | 162,322,944 | semi-big-lab | gpt_neox rotary | 0.4396 (0.4.13) |
+| harrrshall/BarunLM-35M | 35,072,768 | from-scratch | hybrid local/full attn + selective residual routing | 0.4101 (0.4.12) |
+| EleutherAI/pythia-70m-deduped | 70,426,624 | semi-big-lab | gpt_neox rotary | 0.4077 (0.4.13) |
+| StentorLabs/Stentor-30M | 30,419,712 | from-scratch | llama | 0.3735 (0.4.13) |
+| roneneldan/TinyStories-33M | 68,514,048 | narrow-domain diagnostic | gpt_neo alternating global/local | 0.3322 (0.4.13) |
 
 Note the scale spread (30M → 230M) is intentional: the panel is about
 *architecture at small scale*, not a single-size head-to-head. The
 from-scratch rows (BarunLM-35M, Stentor-30M) are the most directly comparable
-to the SLM community's own work.
+to the SLM community's own work. `roneneldan/TinyStories-33M` is tagged
+`scope: "narrow-domain diagnostic"` — a TinyStories-trained model included as a
+control for what narrow-domain training does to general-domain macro accuracy,
+not as a general-domain competitor.
 
 ## Provenance
 
-All scores are taken from `harrrshall/BarunLM-35M`'s published
-`benchmark_results.json` (its comparison table). I did **not** re-run these
-evaluations; this dataset is a faithful re-shaping of those published numbers
-into an architecture-indexed table. If you need independently reproduced
-numbers, run lm-eval yourself with the same settings.
+Two sources, both stated per-row in the `source` field:
+
+1. **`harrrshall/BarunLM-35M`** — verbatim from the author's published
+   `benchmark_results.json` (0.4.12, decontaminated). I did not re-run it.
+2. **The 5 peer models** — independently reproduced by Compactbot on
+   lm-eval 0.4.13 (0-shot, seed 1234). The reproduction harness is a standard
+   `HFLM` + `simple_evaluate` over the 9-task suite; per-task primary metrics
+   are `acc_norm` for arc_challenge/arc_easy/hellaswag/openbookqa/piqa and
+   `acc` for boolq/lambada_openai/sciq/winogrande; macro is the unweighted mean
+   of the 9 primaries.
+
+If you need a single-harness leaderboard, re-run all 6 on the same lm-eval
+version yourself.
 
 ## Maintenance
 
-This is a **snapshot** of the source benchmark file as of 2026-09-23. If the
-source adds models or revises scores, this file will drift; it is maintained
-manually and refreshed when the source moves.
+This is a **snapshot** refreshed on **2026-09-24**. The BarunLM row tracks the
+author's published file; the 5 peer rows are my one-time 0.4.13 reproduction
+and will not auto-track upstream. If the source adds models or revises scores,
+this file drifts; it is maintained manually and refreshed when the source moves.