add arc1 README

arc1/README.md

@@ -0,0 +1,68 @@
+# FPRM — ARC-1-concept (single-z, conv1d)
+
+Fixed-Point Tiny Recursive Model (FPRM / FPTRM single-z) trained on **ARC-1-concept**
+(`arc1concept-aug-1000`, the 1000-task augmented build). Evaluated with two attempts it
+reaches **47.5% pass@2**.
+
+Original run name: `fptrm_singlez_arc1_lr1e-3_wd0.01_det_a0.75_0.25_iter8` (step_518071 = final).
+
+## Results
+
+| metric | value |
+|---|---|
+| **test pass@2** | **47.5%** |
+| eval budget | `max_iter_eval=1000` (stepsize_decay 0.9, patience 5) |
+| checkpoint | `step_518071` (final, EMA) |
+
+## Architecture
+
+| | value |
+|---|---|
+| model | FPTRM single-z (`fp_trm_singlez`) |
+| hidden size / heads / expansion | 512 / 8 / 4 |
+| H-cycles / L-cycles | 3 / 6 |
+| H-layers / L-layers | 0 / 2 |
+| n_backwards_L | 6 |
+| positional enc / puzzle-emb len | RoPE / 16 |
+| halting | fixed-point + Q-head (ACT) |
+| max_iter (train) | 8 (deterministic) |
+| max_iter (eval) | 1000 |
+| stepsize-decay / patience | 0.9 / 5 |
+| fp_thresh | 0.1 |
+| norm type / placement | pre-norm / none |
+| residual scaling | input-independent (α₁, α₂ init = 0.75, 0.25) |
+| conv branch | conv1d (kernel 4) |
+| loss | stablemax cross-entropy, deep supervision, q_loss_coeff 0.5 |
+
+## Optimizer
+
+| | value |
+|---|---|
+| optimizer | adam_atan2 (β = 0.9, 0.95) |
+| learning rate | 1e-3 (constant, no warm-up) |
+| weight decay | 0.01 |
+| **puzzle-emb LR** | **1e-2 (decoupled)** |
+| puzzle-emb WD | 1.0 |
+| batch size | 768 |
+| epochs | 100000 |
+| EMA | enabled (rate 0.999) |
+| seed | 0 |
+
+## Files
+
+- `step_<N>` — EMA-averaged eval checkpoints (~1.7 GB each), N = 51807 … **518071** (final).
+  Because `ema=True`, `step_<N>` is the EMA copy — **these are the weights to load for
+  inference/eval**.
+- `all_config.yaml` — the exact training configuration.
+- `fp_trm_singlez.py`, `losses.py` — model and loss source for this checkpoint.
+- `evaluator_ARC_step_0/submission.json` — ARC evaluation submission (predictions) for this run.
+
+> Note: this folder ships **eval checkpoints only** — the `step_<N>_train_state.pt` training
+> states (optimizer + live weights, for resuming) are not included.
+
+## Reproduce the 47.5%
+
+1. **Train** with `all_config.yaml` (FPTRM single-z, conv1d/k4, pre-norm / placement none,
+   input-independent residual scaling, `adam_atan2`, lr 1e-3, decoupled puzzle-emb LR 1e-2,
+   100000 epochs) on the **`arc1concept-aug-1000`** build. Produces `step_518071`.
+2. **Eval at `max_iter=1000`** on the final EMA checkpoint and report **pass@2** (two attempts).