From 8ba1736219f5617a20f59e0b5f94330af213e4f7 Mon Sep 17 00:00:00 2001 From: Arth Singh Date: Sat, 21 Mar 2026 03:59:32 +0530 Subject: [PATCH] Non-record: Depth Recurrence 5x3 dim=640, val_bpb=1.2716 Explores depth recurrence as a parameter-efficient approach to increase effective model depth. 5 unique layers looped 3 times = 15 effective depth at 15M params. Includes learnable loop embeddings and loop gates for per-pass conditioning. Run terminated early due to compute budget. --- .../README.md | 76 + .../submission.json | 13 + .../train.log | 65 + .../train_gpt.py | 1365 +++++++++++++++++ 4 files changed, 1519 insertions(+) create mode 100644 records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/README.md create mode 100644 records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/submission.json create mode 100644 records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/train.log create mode 100644 records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/train_gpt.py diff --git a/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/README.md b/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/README.md new file mode 100644 index 0000000000..abad04f0b1 --- /dev/null +++ b/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/README.md @@ -0,0 +1,76 @@ +# Depth Recurrence: 5 Unique Layers x 3 Loops (val_bpb=1.2716) + +**Non-record submission** exploring depth recurrence as a parameter-efficient way to increase effective model depth. + +## Core Idea + +Instead of N unique transformer blocks, use fewer unique blocks and loop through them multiple times. This gives more effective depth for the same parameter budget, freeing parameters to increase model width. + +**Config:** 5 unique layers looped 3 times = 15 effective depth at dim=640 (15M params) +**Baseline comparison:** 9 unique layers at dim=512 (17M params, 9 effective depth) + +## Architecture Changes + +Three modifications to the baseline GPT: + +1. **Shared-weight loops**: The model has `num_unique_layers=5` transformer blocks. During forward pass, it loops through all 5 blocks `num_loops=3` times, giving 15 effective layers of computation while only storing 5 layers of parameters. + +2. **Loop embeddings**: Learnable per-loop vectors (`loop_embeds`, shape `[num_loops, dim]`) added to the residual stream at the start of each loop. These let the model differentiate between passes through the same weights. Initialized to zero so the first loop behaves like vanilla. + +3. **Loop gates**: Learnable per-loop scalars (`loop_gates`) that control how much each loop contributes vs. reverting to the initial representation x0. After loop `i > 0`: `x = gate_i * x + (1 - gate_i) * x0`. Initialized to `1/num_loops`. + +The encoder-decoder skip connection pattern from the baseline was removed to keep the recurrence clean. + +## Results + +| Step | val_bpb | val_loss | +|------|---------|----------| +| 500 | 1.4720 | 2.4853 | +| 1000 | 1.3730 | 2.3182 | +| 2000 | 1.3138 | 2.2184 | +| 3000 | 1.2916 | 2.1809 | +| 4000 | 1.2775 | 2.1570 | +| 4500 | 1.2716 | 2.1471 | + +Run was terminated early at ~step 5300 due to compute budget (RunPod time limit). val_bpb was still improving. + +**Baseline comparison (9L/512dim, same optimizer settings):** +- Step 9000: val_bpb = 1.2507 + +At comparable training FLOPs, depth recurrence underperformed the baseline by ~0.02 BPB. + +## Analysis: Why It Didn't Beat Baseline + +1. **Conservative loop gating**: Gates initialized at 1/3 pull the representation back toward x0 after each loop. This effectively limits the model to ~1-1.5 loops of useful computation. The gating was intended to stabilize training through deep recurrence, but it over-regularized. + +2. **No skip connections**: The baseline's encoder-decoder skip pattern (U-Net style) provides important information shortcuts. We removed these to simplify the recurrence, but this likely hurt. + +3. **Gradient amplification through loops**: The same weights receive gradients from all 3 loops, which changes the effective learning rate. We didn't compensate for this — a lower matrix_lr might help. + +4. **Fewer unique representations**: 5 unique layers means less representational diversity per loop pass compared to 9 unique layers, even with loop conditioning. + +## What Would Improve This + +- Remove loop gating entirely, or initialize gates closer to 1.0 +- Add skip connections within each loop pass (encoder-decoder pattern per loop) +- Use 2 loops with 7 unique layers instead of 3 loops with 5 — less recurrence, more layer diversity +- Scale down learning rate proportional to num_loops (each weight gets gradient from all loops) +- Progressive loop training: start with 1 loop, add more during training +- Combine with techniques from current SOTA: sliding window eval, FP16 embeddings, int6 quantization + +## Hardware & Training + +- **Hardware**: 6x NVIDIA H200 (141GB HBM3e each), RunPod +- **Torch**: 2.8.0+cu128 +- **Training**: Distributed across 6 GPUs, ~197ms/step, FlashAttention-2, bfloat16 +- **Total training time**: ~888 seconds (14.8 minutes) for 4500 steps +- **Run was wallclock-limited, not converged** + +## How to Run + +```bash +RUN_ID=recurrent_5x3_d640 \ +NUM_UNIQUE_LAYERS=5 NUM_LOOPS=3 \ +MODEL_DIM=640 NUM_HEADS=8 NUM_KV_HEADS=4 \ +torchrun --standalone --nproc_per_node=1 train_gpt.py +``` diff --git a/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/submission.json b/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/submission.json new file mode 100644 index 0000000000..1f419d9e7d --- /dev/null +++ b/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/submission.json @@ -0,0 +1,13 @@ +{ + "name": "Arth Singh", + "github_id": "Arth-Singh", + "run_name": "Depth Recurrence 5x3 dim=640", + "val_bpb": 1.2716, + "val_loss": 2.1471, + "model_params": 15006123, + "compressed_model_bytes": null, + "train_steps": 4500, + "train_time_seconds": 888, + "hardware": "6x NVIDIA H200 (RunPod)", + "notes": "Non-record submission exploring depth recurrence: 5 unique transformer layers looped 3 times = 15 effective depth. Novel loop embeddings and learnable loop gates for per-pass conditioning. Run terminated early due to compute budget — val_bpb was still improving. See README for analysis." +} diff --git a/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/train.log b/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/train.log new file mode 100644 index 0000000000..55131ff3f0 --- /dev/null +++ b/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/train.log @@ -0,0 +1,65 @@ +logs/recurrent_5x3_d640_6gpu.txt +val_bpb:enabled tokenizer_kind=sentencepiece tokenizer_path=./data/tokenizers/fineweb_1024_bpe.model +train_loader:dataset:fineweb10B_sp1024 train_shards:80 +val_loader:shards pattern=./data/datasets/fineweb10B_sp1024/fineweb_val_*.bin tokens:62021632 +model_params:15006123 +world_size:6 grad_accum_steps:1 +sdp_backends:cudnn=False flash=True mem_efficient=False math=False +attention_mode:gqa num_heads:8 num_kv_heads:4 +tie_embeddings:True embed_lr:0.05 head_lr:0.0 matrix_lr:0.04 scalar_lr:0.04 +train_batch_tokens:786432 train_seq_len:1024 iterations:20000 warmup_steps:20 max_wallclock_seconds:2400.000 +seed:1337 +warmup_step:1/20 +warmup_step:2/20 +warmup_step:3/20 +warmup_step:4/20 +warmup_step:5/20 +warmup_step:6/20 +warmup_step:7/20 +warmup_step:8/20 +warmup_step:9/20 +warmup_step:10/20 +warmup_step:11/20 +warmup_step:12/20 +warmup_step:13/20 +warmup_step:14/20 +warmup_step:15/20 +warmup_step:16/20 +warmup_step:17/20 +warmup_step:18/20 +warmup_step:19/20 +warmup_step:20/20 +step:0/20000 val_loss:6.9442 val_bpb:4.1127 train_time:0ms step_avg:0.02ms +step:1/20000 train_loss:6.9449 train_time:96ms step_avg:96.05ms +step:2/20000 train_loss:19.7308 train_time:260ms step_avg:130.24ms +step:3/20000 train_loss:9.2896 train_time:442ms step_avg:147.40ms +step:4/20000 train_loss:6.3345 train_time:621ms step_avg:155.29ms +step:5/20000 train_loss:6.1266 train_time:799ms step_avg:159.74ms +step:6/20000 train_loss:6.1883 train_time:979ms step_avg:163.10ms +step:7/20000 train_loss:6.1662 train_time:1164ms step_avg:166.23ms +step:8/20000 train_loss:6.1170 train_time:1345ms step_avg:168.16ms +step:9/20000 train_loss:6.0177 train_time:1523ms step_avg:169.24ms +step:10/20000 train_loss:5.9612 train_time:1705ms step_avg:170.45ms +step:50/20000 train_loss:4.2888 train_time:8842ms step_avg:176.85ms +step:100/20000 train_loss:3.5005 train_time:17799ms step_avg:177.99ms +step:150/20000 train_loss:3.1377 train_time:29249ms step_avg:195.00ms +step:200/20000 train_loss:2.6386 train_time:38219ms step_avg:191.10ms +step:250/20000 train_loss:2.6223 train_time:47218ms step_avg:188.87ms +step:300/20000 train_loss:2.6910 train_time:58662ms step_avg:195.54ms +step:350/20000 train_loss:2.6512 train_time:246646ms step_avg:704.70ms +step:400/20000 train_loss:2.4294 train_time:281865ms step_avg:704.66ms +step:450/20000 train_loss:2.5145 train_time:87999ms step_avg:195.55ms +step:500/20000 train_loss:2.5140 train_time:96955ms step_avg:193.91ms +step:500/20000 val_loss:2.4853 val_bpb:1.4720 train_time:97049ms step_avg:194.10ms +step:550/20000 train_loss:2.4472 train_time:107848ms step_avg:196.09ms +step:600/20000 train_loss:2.4336 train_time:116822ms step_avg:194.70ms +step:1000/20000 train_loss:2.4097 train_time:195507ms step_avg:195.51ms +step:1000/20000 val_loss:2.3182 val_bpb:1.3730 train_time:195599ms step_avg:195.60ms +step:1500/20000 val_loss:2.2573 val_bpb:1.3369 train_time:294372ms step_avg:196.25ms +step:2000/20000 val_loss:2.2184 val_bpb:1.3138 train_time:393272ms step_avg:196.64ms +step:2500/20000 val_loss:2.1971 val_bpb:1.3012 train_time:492043ms step_avg:196.82ms +step:3000/20000 val_loss:2.1809 val_bpb:1.2916 train_time:591026ms step_avg:197.01ms +step:3500/20000 val_loss:2.1691 val_bpb:1.2847 train_time:690009ms step_avg:197.15ms +step:4000/20000 val_loss:2.1570 val_bpb:1.2775 train_time:789520ms step_avg:197.38ms +step:4500/20000 val_loss:2.1471 val_bpb:1.2716 train_time:888434ms step_avg:197.43ms +stopping_early: wallclock_cap train_time:~1046000ms step:~5300/20000 diff --git a/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/train_gpt.py b/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/train_gpt.py new file mode 100644 index 0000000000..7c53d775b0 --- /dev/null +++ b/records/track_non_record_16mb/2026-03-21_DepthRecurrence_5x3_640/train_gpt.py @@ -0,0 +1,1365 @@ +""" +The `train_gpt.py` and `train_gpt_mlx.py` scripts are intended as good launching-off points for new participants, not SOTA configs. We'll accept PRs that tune, improve, or simplify these scripts without significantly increasing complexity, but competitive submissions should stay in the `/records` folder. + +Hard stop: To keep readable for newcomers, let's make sure `train_gpt.py` and `train_gpt_mlx.py` never are longer than 1500 lines. +""" + +from __future__ import annotations + +import copy +import glob +import io +import math +import os +import random +import subprocess +import sys +import time +import uuid +import zlib +from pathlib import Path + +import numpy as np +import sentencepiece as spm +import torch +import torch.distributed as dist +import torch.nn.functional as F +from torch import Tensor, nn +from torch.nn.parallel import DistributedDataParallel as DDP + +# ----------------------------- +# HYPERPARAMETERS +# ----------------------------- +# Default Simple Baseline run: +# - 9 transformer blocks at width 512 +# - 8 attention heads with 4 KV heads (GQA) and 2x MLP expansion +# - vocab size 1024, sequence length 1024, tied embeddings +# - 524,288 train tokens per step for 20,000 iterations with a ~10 minute cap + +class Hyperparameters: + # Data paths are shard globs produced by the existing preprocessing pipeline. + data_path = os.environ.get("DATA_PATH", "./data/datasets/fineweb10B_sp1024") + train_files = os.path.join(data_path, "fineweb_train_*.bin") + val_files = os.path.join(data_path, "fineweb_val_*.bin") + tokenizer_path = os.environ.get("TOKENIZER_PATH", "./data/tokenizers/fineweb_1024_bpe.model") + run_id = os.environ.get("RUN_ID", str(uuid.uuid4())) + seed = int(os.environ.get("SEED", 1337)) + + # Validation cadence and batch size. Validation always uses the full fineweb_val split. + val_batch_size = int(os.environ.get("VAL_BATCH_SIZE", 524_288)) + val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 1000)) + train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 200)) + + # Training length. + iterations = int(os.environ.get("ITERATIONS", 20000)) + warmdown_iters = int(os.environ.get("WARMDOWN_ITERS", 1200)) + warmup_steps = int(os.environ.get("WARMUP_STEPS", 20)) + train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 524_288)) + train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 1024)) + max_wallclock_seconds = float(os.environ.get("MAX_WALLCLOCK_SECONDS", 600.0)) + qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 1.5)) + + # Model shape — depth recurrence. + vocab_size = int(os.environ.get("VOCAB_SIZE", 1024)) + num_unique_layers = int(os.environ.get("NUM_UNIQUE_LAYERS", 5)) + num_loops = int(os.environ.get("NUM_LOOPS", 3)) + num_layers = int(os.environ.get("NUM_LAYERS", 9)) # compat only + num_kv_heads = int(os.environ.get("NUM_KV_HEADS", 4)) + model_dim = int(os.environ.get("MODEL_DIM", 640)) + num_heads = int(os.environ.get("NUM_HEADS", 8)) + mlp_mult = int(os.environ.get("MLP_MULT", 2)) + tie_embeddings = bool(int(os.environ.get("TIE_EMBEDDINGS", "1"))) + rope_base = float(os.environ.get("ROPE_BASE", 10000.0)) + logit_softcap = float(os.environ.get("LOGIT_SOFTCAP", 30.0)) + + # Optimizer hyperparameters. + embed_lr = float(os.environ.get("EMBED_LR", 0.6)) + head_lr = float(os.environ.get("HEAD_LR", 0.008)) + tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.05)) + tied_embed_init_std = float(os.environ.get("TIED_EMBED_INIT_STD", 0.005)) + matrix_lr = float(os.environ.get("MATRIX_LR", 0.04)) + scalar_lr = float(os.environ.get("SCALAR_LR", 0.04)) + muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.95)) + muon_backend_steps = int(os.environ.get("MUON_BACKEND_STEPS", 5)) + muon_momentum_warmup_start = float(os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.85)) + muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 500)) + beta1 = float(os.environ.get("BETA1", 0.9)) + beta2 = float(os.environ.get("BETA2", 0.95)) + adam_eps = float(os.environ.get("ADAM_EPS", 1e-8)) + grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.0)) + + # Test-time training (LoRA) hyperparameters. + ttt_lora_rank = int(os.environ.get("TTT_LORA_RANK", 8)) + ttt_lora_lr = float(os.environ.get("TTT_LORA_LR", 0.01)) + ttt_chunk_size = int(os.environ.get("TTT_CHUNK_SIZE", 256)) + ttt_eval_seq_len = int(os.environ.get("TTT_EVAL_SEQ_LEN", 1024)) + ttt_batch_size = int(os.environ.get("TTT_BATCH_SIZE", 64)) + +# ----------------------------- +# MUON OPTIMIZER +# ----------------------------- +# +# As borrowed from modded-nanogpt +# Background on Muon: https://kellerjordan.github.io/posts/muon/ + +def zeropower_via_newtonschulz5(G: Tensor, steps: int = 10, eps: float = 1e-7) -> Tensor: + # Orthogonalize a 2D update matrix with a fast Newton-Schulz iteration. + # Muon uses this to normalize matrix-shaped gradients before applying them. + a, b, c = (3.4445, -4.7750, 2.0315) + X = G.bfloat16() + X /= X.norm() + eps + transposed = G.size(0) > G.size(1) + if transposed: + X = X.T + for _ in range(steps): + A = X @ X.T + B = b * A + c * A @ A + X = a * X + B @ X + return X.T if transposed else X + + +class Muon(torch.optim.Optimizer): + def __init__(self, params, lr: float, momentum: float, backend_steps: int, nesterov: bool = True): + super().__init__( + params, + dict(lr=lr, momentum=momentum, backend_steps=backend_steps, nesterov=nesterov), + ) + + @torch.no_grad() + def step(self, closure=None): + loss = None + if closure is not None: + with torch.enable_grad(): + loss = closure() + + distributed = dist.is_available() and dist.is_initialized() + world_size = dist.get_world_size() if distributed else 1 + rank = dist.get_rank() if distributed else 0 + + for group in self.param_groups: + params = group["params"] + if not params: + continue + lr = group["lr"] + momentum = group["momentum"] + backend_steps = group["backend_steps"] + nesterov = group["nesterov"] + + total_params = sum(int(p.numel()) for p in params) + updates_flat = torch.zeros(total_params, device=params[0].device, dtype=torch.bfloat16) + + curr = 0 + for i, p in enumerate(params): + if i % world_size == rank and p.grad is not None: + g = p.grad + state = self.state[p] + if "momentum_buffer" not in state: + state["momentum_buffer"] = torch.zeros_like(g) + buf = state["momentum_buffer"] + buf.mul_(momentum).add_(g) + if nesterov: + g = g.add(buf, alpha=momentum) + g = zeropower_via_newtonschulz5(g, steps=backend_steps) + # Scale correction from Muon reference implementations. + g *= max(1, g.size(0) / g.size(1)) ** 0.5 + updates_flat[curr : curr + p.numel()] = g.reshape(-1) + curr += p.numel() + + if distributed: + dist.all_reduce(updates_flat, op=dist.ReduceOp.SUM) + + curr = 0 + for p in params: + g = updates_flat[curr : curr + p.numel()].view_as(p).to(dtype=p.dtype) + p.add_(g, alpha=-lr) + curr += p.numel() + + return loss + + +# ----------------------------- +# TOKENIZER-AGNOSTIC EVALUATION SETUP +# ----------------------------- +# +# It's common for small models have a large fraction of their parameters be embeddings, since the 2 * d_model * d_vocab vectors can be gigantic. +# Instead of locking the tokenizer, we let you bring your own and calculate our validation metrics on the average compression of the validation set. +# We calculate BPB (bits-per-byte) instead of validation loss, so we need methods to count the number of bits per token in the tokenizer. +# Note: Submissions that edit the tokenizer will be examined more carefully, since screwing this up might unjustly improve your score. + +def build_sentencepiece_luts( + sp: spm.SentencePieceProcessor, vocab_size: int, device: torch.device +) -> tuple[Tensor, Tensor, Tensor]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("▁"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ) + + +def load_validation_tokens(pattern: str, seq_len: int) -> Tensor: + files = [Path(p) for p in sorted(glob.glob(pattern))] + if not files: + raise FileNotFoundError(f"No files found for pattern: {pattern}") + # The export pipeline writes the fixed first-50k-doc validation set to fineweb_val_*. + tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() + usable = ((tokens.numel() - 1) // seq_len) * seq_len + if usable <= 0: + raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") + return tokens[: usable + 1] + + +def eval_val( + args: Hyperparameters, + model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + grad_accum_steps: int, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, +) -> tuple[float, float]: + # Validation computes two metrics: + # - val_loss: token cross-entropy (natural log) + # - val_bpb: tokenizer-agnostic compression metric used by the challenge + local_batch_tokens = args.val_batch_size // (world_size * grad_accum_steps) + if local_batch_tokens < args.train_seq_len: + raise ValueError( + "VAL_BATCH_SIZE must provide at least one sequence per rank; " + f"got VAL_BATCH_SIZE={args.val_batch_size}, WORLD_SIZE={world_size}, " + f"GRAD_ACCUM_STEPS={grad_accum_steps}, TRAIN_SEQ_LEN={args.train_seq_len}" + ) + local_batch_seqs = local_batch_tokens // args.train_seq_len + total_seqs = (val_tokens.numel() - 1) // args.train_seq_len + seq_start = (total_seqs * rank) // world_size + seq_end = (total_seqs * (rank + 1)) // world_size + val_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + val_token_count = torch.zeros((), device=device, dtype=torch.float64) + val_byte_count = torch.zeros((), device=device, dtype=torch.float64) + + model.eval() + with torch.inference_mode(): + for batch_seq_start in range(seq_start, seq_end, local_batch_seqs): + batch_seq_end = min(batch_seq_start + local_batch_seqs, seq_end) + raw_start = batch_seq_start * args.train_seq_len + raw_end = batch_seq_end * args.train_seq_len + 1 + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64, non_blocking=True) + x = local[:-1].reshape(-1, args.train_seq_len) + y = local[1:].reshape(-1, args.train_seq_len) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + batch_loss = model(x, y).detach() + batch_token_count = float(y.numel()) + val_loss_sum += batch_loss.to(torch.float64) * batch_token_count + val_token_count += batch_token_count + prev_ids = x.reshape(-1) + tgt_ids = y.reshape(-1) + token_bytes = base_bytes_lut[tgt_ids].to(dtype=torch.int16) + token_bytes += (has_leading_space_lut[tgt_ids] & ~is_boundary_token_lut[prev_ids]).to(dtype=torch.int16) + val_byte_count += token_bytes.to(torch.float64).sum() + + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(val_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(val_token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(val_byte_count, op=dist.ReduceOp.SUM) + + val_loss = val_loss_sum / val_token_count + bits_per_token = val_loss.item() / math.log(2.0) + tokens_per_byte = val_token_count.item() / val_byte_count.item() + model.train() + return float(val_loss.item()), float(bits_per_token * tokens_per_byte) + +# ----------------------------- +# POST-TRAINING QUANTIZATION +# ----------------------------- +# +# It's silly to export our model, which is trained in bf16 and fp32, at that same precision. +# Instead, we get approximately the same model (with a small hit) by quantizing the model to int8 & zlib compressing. +# We can then decompress the model and run in higher precision for evaluation, after closing in under the size limit. + +CONTROL_TENSOR_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "CONTROL_TENSOR_NAME_PATTERNS", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,loop_embeds,loop_gates", + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_FP32_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "INT8_KEEP_FLOAT_FP32_NAME_PATTERNS", + ",".join(CONTROL_TENSOR_NAME_PATTERNS), + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_MAX_NUMEL = 65_536 +INT8_KEEP_FLOAT_STORE_DTYPE = torch.float16 +INT8_PER_ROW_SCALE_DTYPE = torch.float16 +INT8_CLIP_PERCENTILE = 99.99984 +INT8_CLIP_Q = INT8_CLIP_PERCENTILE / 100.0 + +def tensor_nbytes(t: Tensor) -> int: + return int(t.numel()) * int(t.element_size()) + +def keep_float_tensor(name: str, t: Tensor, passthrough_orig_dtypes: dict[str, str]) -> Tensor: + if any(pattern in name for pattern in INT8_KEEP_FLOAT_FP32_NAME_PATTERNS): + return t.float().contiguous() + if t.dtype in {torch.float32, torch.bfloat16}: + passthrough_orig_dtypes[name] = str(t.dtype).removeprefix("torch.") + return t.to(dtype=INT8_KEEP_FLOAT_STORE_DTYPE).contiguous() + return t + +def quantize_float_tensor(t: Tensor) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + # Matrices get one scale per row, which usually tracks output-channel + # ranges much better than a single tensor-wide scale. + clip_abs = ( + torch.quantile(t32.abs(), INT8_CLIP_Q, dim=1) + if t32.numel() + else torch.empty((t32.shape[0],), dtype=torch.float32) + ) + clipped = torch.maximum(torch.minimum(t32, clip_abs[:, None]), -clip_abs[:, None]) + scale = (clip_abs / 127.0).clamp_min(1.0 / 127.0) + q = torch.clamp(torch.round(clipped / scale[:, None]), -127, 127).to(torch.int8).contiguous() + return q, scale.to(dtype=INT8_PER_ROW_SCALE_DTYPE).contiguous() + + # Vectors / scalars use a simpler per-tensor scale. + clip_abs = float(torch.quantile(t32.abs().flatten(), INT8_CLIP_Q).item()) if t32.numel() else 0.0 + scale = torch.tensor(clip_abs / 127.0 if clip_abs > 0 else 1.0, dtype=torch.float32) + q = torch.clamp(torch.round(torch.clamp(t32, -clip_abs, clip_abs) / scale), -127, 127).to(torch.int8).contiguous() + return q, scale + +def quantize_state_dict_int8(state_dict: dict[str, Tensor]): + # Single supported clean-script export format: + # - per-row int8 for 2D float tensors + # - per-tensor int8 for other float tensors + # - exact passthrough for non-floats + # - passthrough for small float tensors, stored as fp16 to save bytes + quantized: dict[str, Tensor] = {} + scales: dict[str, Tensor] = {} + dtypes: dict[str, str] = {} + passthrough: dict[str, Tensor] = {} + passthrough_orig_dtypes: dict[str, str] = {} + qmeta: dict[str, dict[str, object]] = {} + stats = dict.fromkeys( + ("param_count", "num_tensors", "num_float_tensors", "num_nonfloat_tensors", "baseline_tensor_bytes", "int8_payload_bytes"), + 0, + ) + + for name, tensor in state_dict.items(): + t = tensor.detach().to("cpu").contiguous() + stats["param_count"] += int(t.numel()) + stats["num_tensors"] += 1 + stats["baseline_tensor_bytes"] += tensor_nbytes(t) + + if not t.is_floating_point(): + stats["num_nonfloat_tensors"] += 1 + passthrough[name] = t + stats["int8_payload_bytes"] += tensor_nbytes(t) + continue + + # Small float tensors are cheap enough to keep directly. We still downcast + # fp32/bf16 passthrough tensors to fp16 so metadata does not dominate size. + if t.numel() <= INT8_KEEP_FLOAT_MAX_NUMEL: + kept = keep_float_tensor(name, t, passthrough_orig_dtypes) + passthrough[name] = kept + stats["int8_payload_bytes"] += tensor_nbytes(kept) + continue + + stats["num_float_tensors"] += 1 + q, s = quantize_float_tensor(t) + if s.ndim > 0: + qmeta[name] = {"scheme": "per_row", "axis": 0} + quantized[name] = q + scales[name] = s + dtypes[name] = str(t.dtype).removeprefix("torch.") + stats["int8_payload_bytes"] += tensor_nbytes(q) + tensor_nbytes(s) + + obj: dict[str, object] = { + "__quant_format__": "int8_clean_per_row_v1", + "quantized": quantized, + "scales": scales, + "dtypes": dtypes, + "passthrough": passthrough, + } + if qmeta: + obj["qmeta"] = qmeta + if passthrough_orig_dtypes: + obj["passthrough_orig_dtypes"] = passthrough_orig_dtypes + return obj, stats + +def dequantize_state_dict_int8(obj: dict[str, object]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + qmeta = obj.get("qmeta", {}) + passthrough_orig_dtypes = obj.get("passthrough_orig_dtypes", {}) + for name, q in obj["quantized"].items(): + dtype = getattr(torch, obj["dtypes"][name]) + s = obj["scales"][name] + if qmeta.get(name, {}).get("scheme") == "per_row" or s.ndim > 0: + s = s.to(dtype=torch.float32) + # Broadcast the saved row scale back across trailing dimensions. + out[name] = (q.float() * s.view(q.shape[0], *([1] * (q.ndim - 1)))).to(dtype=dtype).contiguous() + else: + scale = float(s.item()) + out[name] = (q.float() * scale).to(dtype=dtype).contiguous() + for name, t in obj["passthrough"].items(): + # Restore small tensors, undoing the temporary fp16 storage cast if needed. + out_t = t.detach().to("cpu").contiguous() + orig_dtype = passthrough_orig_dtypes.get(name) + if isinstance(orig_dtype, str): + out_t = out_t.to(dtype=getattr(torch, orig_dtype)).contiguous() + out[name] = out_t + return out + + +# ----------------------------- +# DATA LOADING +# ----------------------------- + +def load_data_shard(file: Path) -> Tensor: + header_bytes = 256 * np.dtype(" None: + self.file_idx = (self.file_idx + 1) % len(self.files) + self.tokens = load_data_shard(self.files[self.file_idx]) + self.pos = 0 + + def take(self, n: int) -> Tensor: + chunks: list[Tensor] = [] + remaining = n + while remaining > 0: + avail = self.tokens.numel() - self.pos + if avail <= 0: + self._advance_file() + continue + k = min(remaining, avail) + chunks.append(self.tokens[self.pos : self.pos + k]) + self.pos += k + remaining -= k + return chunks[0] if len(chunks) == 1 else torch.cat(chunks) + + +class DistributedTokenLoader: + # Each call consumes a contiguous chunk from the shared token stream, then slices out + # one disjoint span per rank. The extra "+1" token lets us build (x, y) by shifting. + def __init__(self, pattern: str, rank: int, world_size: int, device: torch.device): + self.rank = rank + self.world_size = world_size + self.device = device + self.stream = TokenStream(pattern) + + def next_batch(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> tuple[Tensor, Tensor]: + local_tokens = global_tokens // (self.world_size * grad_accum_steps) + per_rank_span = local_tokens + 1 + chunk = self.stream.take(per_rank_span * self.world_size) + start = self.rank * per_rank_span + local = chunk[start : start + per_rank_span].to(dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + return x.to(self.device, non_blocking=True), y.to(self.device, non_blocking=True) + +# ----------------------------- +# TRANSFORMER MODULES +# ----------------------------- + +class RMSNorm(nn.Module): + def __init__(self, eps: float | None = None): + super().__init__() + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + return F.rms_norm(x, (x.size(-1),), eps=self.eps) + + +class CastedLinear(nn.Linear): + # Keep weights in fp32 for optimizer/state quality, cast at matmul time for bf16 compute. + def forward(self, x: Tensor) -> Tensor: + bias = self.bias.to(x.dtype) if self.bias is not None else None + return F.linear(x, self.weight.to(x.dtype), bias) + + +def restore_low_dim_params_to_fp32(module: nn.Module) -> None: + # Keep small/control parameters in fp32 even when the model body runs in bf16. + with torch.no_grad(): + for name, param in module.named_parameters(): + if (param.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS)) and param.dtype != torch.float32: + param.data = param.data.float() + + +class Rotary(nn.Module): + # Caches cos/sin tables per sequence length on the current device. + def __init__(self, dim: int, base: float = 10000.0): + super().__init__() + inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2, dtype=torch.float32) / dim)) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self._seq_len_cached = 0 + self._cos_cached: Tensor | None = None + self._sin_cached: Tensor | None = None + + def forward(self, seq_len: int, device: torch.device, dtype: torch.dtype) -> tuple[Tensor, Tensor]: + if ( + self._cos_cached is None + or self._sin_cached is None + or self._seq_len_cached != seq_len + or self._cos_cached.device != device + ): + t = torch.arange(seq_len, device=device, dtype=self.inv_freq.dtype) + freqs = torch.outer(t, self.inv_freq.to(device)) + self._cos_cached = freqs.cos()[None, None, :, :] + self._sin_cached = freqs.sin()[None, None, :, :] + self._seq_len_cached = seq_len + return self._cos_cached.to(dtype=dtype), self._sin_cached.to(dtype=dtype) + + +def apply_rotary_emb(x: Tensor, cos: Tensor, sin: Tensor) -> Tensor: + half = x.size(-1) // 2 + x1, x2 = x[..., :half], x[..., half:] + return torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + + +class CausalSelfAttention(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + rope_base: float, + qk_gain_init: float, + ): + super().__init__() + if dim % num_heads != 0: + raise ValueError("model_dim must be divisible by num_heads") + if num_heads % num_kv_heads != 0: + raise ValueError("num_heads must be divisible by num_kv_heads") + self.num_heads = num_heads + self.num_kv_heads = num_kv_heads + self.head_dim = dim // num_heads + if self.head_dim % 2 != 0: + raise ValueError("head_dim must be even for RoPE") + kv_dim = self.num_kv_heads * self.head_dim + self.c_q = CastedLinear(dim, dim, bias=False) + self.c_k = CastedLinear(dim, kv_dim, bias=False) + self.c_v = CastedLinear(dim, kv_dim, bias=False) + self.proj = CastedLinear(dim, dim, bias=False) + self.proj._zero_init = True + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rotary = Rotary(self.head_dim, base=rope_base) + + def forward(self, x: Tensor, q_delta=None, v_delta=None) -> Tensor: + bsz, seqlen, dim = x.shape + q = self.c_q(x) + (q_delta if q_delta is not None else 0) + k = self.c_k(x) + v = self.c_v(x) + (v_delta if v_delta is not None else 0) + q = q.reshape(bsz, seqlen, self.num_heads, self.head_dim).transpose(1, 2) + k = k.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim).transpose(1, 2) + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim).transpose(1, 2) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin) + k = apply_rotary_emb(k, cos, sin) + q = q * self.q_gain.to(dtype=q.dtype)[None, :, None, None] + y = F.scaled_dot_product_attention( + q, k, v, attn_mask=None, is_causal=True, + enable_gqa=(self.num_kv_heads != self.num_heads), + ) + y = y.transpose(1, 2).contiguous().reshape(bsz, seqlen, dim) + return self.proj(y) + + +class MLP(nn.Module): + # relu^2 MLP from the original modded-nanogpt setup + def __init__(self, dim: int, mlp_mult: int): + super().__init__() + hidden = mlp_mult * dim + self.fc = CastedLinear(dim, hidden, bias=False) + self.proj = CastedLinear(hidden, dim, bias=False) + self.proj._zero_init = True + + def forward(self, x: Tensor) -> Tensor: + x = torch.relu(self.fc(x)) + return self.proj(x.square()) + + +class Block(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + rope_base: float, + qk_gain_init: float, + ): + super().__init__() + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = CausalSelfAttention(dim, num_heads, num_kv_heads, rope_base, qk_gain_init) + self.mlp = MLP(dim, mlp_mult) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + + def forward(self, x: Tensor, x0: Tensor, q_delta_fn=None, v_delta_fn=None) -> Tensor: + mix = self.resid_mix.to(dtype=x.dtype) + x = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + n = self.attn_norm(x) + qd = q_delta_fn(n) if q_delta_fn is not None else None + vd = v_delta_fn(n) if v_delta_fn is not None else None + attn_out = self.attn(n, qd, vd) + x = x + self.attn_scale.to(dtype=x.dtype)[None, None, :] * attn_out + x = x + self.mlp_scale.to(dtype=x.dtype)[None, None, :] * self.mlp(self.mlp_norm(x)) + return x + + +class GPT(nn.Module): + # Depth recurrence: N unique blocks looped L times = N*L effective depth. + # Loop embeddings let the model differentiate passes. Loop gates control contribution. + def __init__( + self, + vocab_size: int, + num_unique_layers: int, + num_loops: int, + model_dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + tie_embeddings: bool, + tied_embed_init_std: float, + logit_softcap: float, + rope_base: float, + qk_gain_init: float, + ): + super().__init__() + if logit_softcap <= 0.0: + raise ValueError(f"logit_softcap must be positive, got {logit_softcap}") + self.tie_embeddings = tie_embeddings + self.tied_embed_init_std = tied_embed_init_std + self.logit_softcap = logit_softcap + self.num_unique_layers = num_unique_layers + self.num_loops = num_loops + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.blocks = nn.ModuleList( + [ + Block(model_dim, num_heads, num_kv_heads, mlp_mult, rope_base, qk_gain_init) + for _ in range(num_unique_layers) + ] + ) + self.loop_embeds = nn.Parameter(torch.zeros(num_loops, model_dim, dtype=torch.float32)) + self.loop_gates = nn.Parameter(torch.ones(num_loops, dtype=torch.float32) * (1.0 / num_loops)) + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + if self.lm_head is not None: + self.lm_head._zero_init = True + self._init_weights() + + def _init_weights(self) -> None: + if self.tie_embeddings: + nn.init.normal_(self.tok_emb.weight, mean=0.0, std=self.tied_embed_init_std) + for module in self.modules(): + if isinstance(module, nn.Linear) and getattr(module, "_zero_init", False): + nn.init.zeros_(module.weight) + + def forward(self, input_ids: Tensor, target_ids: Tensor, lora=None) -> Tensor: + x = self.tok_emb(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x0 = x + + for loop_idx in range(self.num_loops): + loop_emb = self.loop_embeds[loop_idx].to(dtype=x.dtype)[None, None, :] + gate = self.loop_gates[loop_idx].to(dtype=x.dtype) + x = x + loop_emb + for i, block in enumerate(self.blocks): + qd = lora.q_loras[i] if lora else None + vd = lora.v_loras[i] if lora else None + x = block(x, x0, qd, vd) + if loop_idx > 0: + x = gate * x + (1.0 - gate) * x0 + + x = self.final_norm(x) + if self.tie_embeddings: + logits = F.linear(x, self.tok_emb.weight) + else: + logits = self.lm_head(x) + logits = logits + (lora.lm_head_lora(x) if lora else 0) + logits = self.logit_softcap * torch.tanh(logits / self.logit_softcap) + if lora: + bsz, sl, V = logits.shape + return F.cross_entropy( + logits.float().reshape(-1, V), target_ids.reshape(-1), reduction="none").reshape(bsz, sl) + return F.cross_entropy(logits.float().reshape(-1, logits.size(-1)), target_ids.reshape(-1), reduction="mean") + + +# ----------------------------- +# TEST-TIME TRAINING (LoRA) +# ----------------------------- +# +# At evaluation time, we adapt per-document low-rank adapters on the validation data. +# Each document gets its own adapter, so there is no inter-document dependency. + +BOS_ID = 1 + +class BatchedLinearLoRA(nn.Module): + """LoRA for a linear layer, with independent weights per batch element. + Computes x @ Aᵀ @ Bᵀ = x @ (BA)ᵀ, i.e. the LoRA delta is ΔW = BA.""" + def __init__(self, bsz: int, in_features: int, out_features: int, rank: int): + super().__init__() + self.in_features = in_features + self.A = nn.Parameter(torch.empty(bsz, rank, in_features)) # down-projection + self.B = nn.Parameter(torch.zeros(bsz, out_features, rank)) # up-projection + self.reset() + + def forward(self, x: Tensor) -> Tensor: + return (x @ self.A.transpose(1, 2)) @ self.B.transpose(1, 2) # (bsz, T, out) + + def reset(self) -> None: + bound = 1.0 / math.sqrt(self.in_features) + with torch.no_grad(): + self.A.uniform_(-bound, bound) # kaiming-uniform + self.B.zero_() + +class BatchedTTTLoRA(nn.Module): + """All LoRA adapters for one batch: LM head and Q/V per block.""" + def __init__(self, bsz: int, model: GPT, rank: int): + super().__init__() + dim = model.tok_emb.embedding_dim + vocab = model.tok_emb.num_embeddings + self.lm_head_lora = BatchedLinearLoRA(bsz, dim, vocab, rank) + self.q_loras = nn.ModuleList() + self.v_loras = nn.ModuleList() + for block in model.blocks: + self.q_loras.append(BatchedLinearLoRA(bsz, dim, block.attn.c_q.weight.shape[0], rank)) + self.v_loras.append(BatchedLinearLoRA(bsz, dim, block.attn.c_v.weight.shape[0], rank)) + + def reset(self) -> None: + for m in self.modules(): + if isinstance(m, BatchedLinearLoRA): + m.reset() + +def _reset_ttt_optimizer(opt): + for group in opt.param_groups: + for p in group['params']: + s = opt.state.get(p) + if not s: # Fresh state. + continue + s['exp_avg'].zero_() + s['exp_avg_sq'].zero_() + s['step'].fill_(0) + +def _build_ttt_optimizer(lora, args: Hyperparameters): + return torch.optim.Adam(lora.parameters(), lr=args.ttt_lora_lr, betas=(args.beta1, args.beta2), eps=1e-10) + +def _find_docs(all_tokens: Tensor, include_next_bos: bool = True) -> list[tuple[int, int]]: + """Return (start_offset, length) for each document, identified by BOS boundaries. + + If include_next_bos is True, include next document's BOS (to match continuous-stream + eval token count exactly). + """ + bos_positions = (all_tokens == BOS_ID).nonzero(as_tuple=True)[0].numpy() + docs = [] + for i in range(len(bos_positions)): + start = int(bos_positions[i]) + end = int(bos_positions[i + 1]) if i + 1 < len(bos_positions) else all_tokens.numel() + if include_next_bos and i + 1 < len(bos_positions): + end += 1 + assert end - start >= 2 + docs.append((start, end - start)) + return docs + +def _compute_chunk_window(ci: int, pred_len: int, num_chunks: int, chunk_size: int, eval_seq_len: int): + """Return (win_start, win_len, chunk_offset, chunk_len) for chunk `ci` of a doc.""" + chunk_start = ci * chunk_size + chunk_end = pred_len if ci == num_chunks - 1 else (ci + 1) * chunk_size + win_start = max(0, chunk_end - eval_seq_len) + win_len = chunk_end - win_start + chunk_offset = chunk_start - win_start + chunk_len = chunk_end - chunk_start + return win_start, win_len, chunk_offset, chunk_len + +def _accumulate_bpb( + ptl: Tensor, x: Tensor, y: Tensor, + batch_i: int, chunk_offset: int, chunk_len: int, + base_bytes_lut: Tensor, has_leading_space_lut: Tensor, is_boundary_token_lut: Tensor, + loss_sum: Tensor, byte_sum: Tensor, token_count: Tensor, +): + """Add one doc-chunk's contribution to the running BPB accumulators.""" + lbl = ptl[batch_i, chunk_offset:chunk_offset + chunk_len].to(torch.float64) + prev = x[batch_i, chunk_offset:chunk_offset + chunk_len] + tgt = y[batch_i, chunk_offset:chunk_offset + chunk_len] + tok_bytes = base_bytes_lut[tgt].to(torch.float64) + tok_bytes += has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev] + loss_sum += lbl.sum() + byte_sum += tok_bytes.sum() + token_count += chunk_len + +def eval_val_ttt_lora( + args: Hyperparameters, + base_model: GPT, + rank: int, + world_size: int, + device: torch.device, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, +) -> tuple[float, float]: + """Evaluate with batched LoRA test-time training. Returns (val_loss, val_bpb).""" + # Load validation tokens and find document boundaries + files = sorted(glob.glob(args.val_files)) + all_tokens = torch.cat([load_data_shard(Path(f)) for f in files]) + docs = _find_docs(all_tokens) + + # Each rank takes a contiguous slice of documents + rank_docs = docs[(len(docs) * rank) // world_size : (len(docs) * (rank + 1)) // world_size] + chunk_size = args.ttt_chunk_size + eval_seq_len = args.ttt_eval_seq_len + batch_size = args.ttt_batch_size + lora_rank = args.ttt_lora_rank + + rank_docs.sort(key=lambda d: (d[1] - 2) // chunk_size) + + base_model.eval() + for p in base_model.parameters(): + p.requires_grad_(False) + + lora = BatchedTTTLoRA(batch_size, base_model, lora_rank).to(device) + opt = _build_ttt_optimizer(lora, args) + + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + byte_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + + for bi in range(0, len(rank_docs), batch_size): + batch = rank_docs[bi:bi + batch_size] + bsz = len(batch) + + if bsz == batch_size: + cur_lora, cur_opt = lora, opt + cur_lora.reset() + _reset_ttt_optimizer(cur_opt) + else: + cur_lora = BatchedTTTLoRA(bsz, base_model, lora_rank).to(device) + cur_opt = _build_ttt_optimizer(cur_lora, args) + + pred_lens = [doc_len - 1 for _, doc_len in batch] + num_chunks = [(pl + chunk_size - 1) // chunk_size for pl in pred_lens] + max_nc = max(num_chunks) + + for ci in range(max_nc): + chunk_stats = _compute_chunk_window(ci, (ci + 1) * chunk_size, ci + 1, chunk_size, eval_seq_len) + context_size, chunk_offset = chunk_stats[1], chunk_stats[2] + + active = [ci < nc for nc in num_chunks] + needs_train = any(ci < nc - 1 for nc in num_chunks) + + x = torch.zeros(bsz, context_size, dtype=torch.int64, device=device) + y = torch.zeros(bsz, context_size, dtype=torch.int64, device=device) + doc_info = [] # (chunk_offset, chunk_len) per doc + for b in range(bsz): + if not active[b]: + doc_info.append((0, 0)) + continue + ds, dl = batch[b] + ws, wl, co, cl = _compute_chunk_window(ci, pred_lens[b], num_chunks[b], chunk_size, eval_seq_len) + chunk = all_tokens[ds + ws: ds + ws + wl + 1] + toks = chunk.to(dtype=torch.int64, device=device) + x[b, :wl] = toks[:-1] + y[b, :wl] = toks[1:] + doc_info.append((co, cl)) + + # Forward pass (keep grad graph alive only when we need to train) + if needs_train: + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + ptl = base_model(x, y, lora=cur_lora) + else: + with torch.no_grad(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + ptl = base_model(x, y, lora=cur_lora) + + # Score: accumulate loss and byte counts for BPB (before training on chunk) + with torch.no_grad(): + for b in range(bsz): + if not active[b]: + continue + co, cl = doc_info[b] + _accumulate_bpb( + ptl, x, y, b, co, cl, base_bytes_lut, has_leading_space_lut, + is_boundary_token_lut, loss_sum, byte_sum, token_count) + + # Train: one Adam step on the LoRA params using this chunk's loss + if needs_train: + mask = torch.tensor([float(ci < num_chunks[b] - 1) for b in range(bsz)], device=device) + per_doc = ptl[:, chunk_offset:chunk_offset + chunk_size].mean(dim=-1) + cur_opt.zero_grad() + (per_doc * mask).sum().backward() + cur_opt.step() + + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + + val_loss = float(loss_sum.item() / token_count.item()) + val_bpb = float((loss_sum.item() / math.log(2.0)) / byte_sum.item()) + return val_loss, val_bpb + +# ----------------------------- +# TRAINING +# ----------------------------- + +def main() -> None: + global zeropower_via_newtonschulz5 + + code = Path(__file__).read_text(encoding="utf-8") + args = Hyperparameters() + zeropower_via_newtonschulz5 = torch.compile(zeropower_via_newtonschulz5) + + # ----------------------------- + # DISTRIBUTED + CUDA SETUP + # ----------------------------- + + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + rank = int(os.environ.get("RANK", "0")) + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + if world_size <= 0: + raise ValueError(f"WORLD_SIZE must be positive, got {world_size}") + total_accum = 8 + if total_accum % world_size != 0: + total_accum = world_size # 1 accum step per rank when world_size doesn't divide 8 + grad_accum_steps = total_accum // world_size + grad_scale = 1.0 / grad_accum_steps + if not torch.cuda.is_available(): + raise RuntimeError("CUDA is required") + device = torch.device("cuda", local_rank) + torch.cuda.set_device(device) + if distributed: + dist.init_process_group(backend="nccl", device_id=device) + dist.barrier() + master_process = rank == 0 + + # Fast math knobs + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + from torch.backends.cuda import enable_cudnn_sdp, enable_flash_sdp, enable_math_sdp, enable_mem_efficient_sdp + + enable_cudnn_sdp(False) + enable_flash_sdp(True) + enable_mem_efficient_sdp(False) + enable_math_sdp(False) + + logfile = None + if master_process: + os.makedirs("logs", exist_ok=True) + logfile = f"logs/{args.run_id}.txt" + print(logfile) + + def log0(msg: str, console: bool = True) -> None: + if not master_process: + return + if console: + print(msg) + if logfile is not None: + with open(logfile, "a", encoding="utf-8") as f: + print(msg, file=f) + + log0(code, console=False) + log0("=" * 100, console=False) + log0(f"Running Python {sys.version}", console=False) + log0(f"Running PyTorch {torch.__version__}", console=False) + log0( + subprocess.run(["nvidia-smi"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=False).stdout, + console=False, + ) + log0("=" * 100, console=False) + + # ----------------------------- + # TOKENIZER + VALIDATION METRIC SETUP + # ----------------------------- + + random.seed(args.seed) + np.random.seed(args.seed) + torch.manual_seed(args.seed) + torch.cuda.manual_seed_all(args.seed) + + if not args.tokenizer_path.endswith(".model"): + raise ValueError(f"Script only setup for SentencePiece .model file: {args.tokenizer_path}") + sp = spm.SentencePieceProcessor(model_file=args.tokenizer_path) + if int(sp.vocab_size()) != args.vocab_size: + raise ValueError( + f"VOCAB_SIZE={args.vocab_size} does not match tokenizer vocab_size={int(sp.vocab_size())}" + ) + dataset_dir = Path(args.data_path).resolve() + actual_train_files = len(list(dataset_dir.glob("fineweb_train_*.bin"))) + val_tokens = load_validation_tokens(args.val_files, args.train_seq_len) + base_bytes_lut, has_leading_space_lut, is_boundary_token_lut = build_sentencepiece_luts( + sp, args.vocab_size, device + ) + log0(f"val_bpb:enabled tokenizer_kind=sentencepiece tokenizer_path={args.tokenizer_path}") + log0(f"train_loader:dataset:{dataset_dir.name} train_shards:{actual_train_files}") + log0(f"val_loader:shards pattern={args.val_files} tokens:{val_tokens.numel() - 1}") + + # ----------------------------- + # MODEL + OPTIMIZER SETUP + # ----------------------------- + + base_model = GPT( + vocab_size=args.vocab_size, + num_unique_layers=args.num_unique_layers, + num_loops=args.num_loops, + model_dim=args.model_dim, + num_heads=args.num_heads, + num_kv_heads=args.num_kv_heads, + mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, + tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, + rope_base=args.rope_base, + qk_gain_init=args.qk_gain_init, + ).to(device).bfloat16() + for module in base_model.modules(): + if isinstance(module, CastedLinear): + module.float() + if isinstance(module, Rotary): + module.inv_freq.data = module.inv_freq.data.float() + restore_low_dim_params_to_fp32(base_model) + compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) + model: nn.Module = DDP(compiled_model, device_ids=[local_rank], broadcast_buffers=False) if distributed else compiled_model + + # Optimizer split: + # - token embedding (Adam) uses EMBED_LR + # - untied lm_head (Adam) uses HEAD_LR + # - matrix params in transformer blocks use MATRIX_LR via Muon + # - vectors/scalars use SCALAR_LR via Adam + block_named_params = list(base_model.blocks.named_parameters()) + matrix_params = [ + p + for name, p in block_named_params + if p.ndim == 2 and not any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + scalar_params = [ + p + for name, p in block_named_params + if p.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + scalar_params.append(base_model.loop_embeds) + scalar_params.append(base_model.loop_gates) + token_lr = args.tied_embed_lr if args.tie_embeddings else args.embed_lr + optimizer_tok = torch.optim.Adam( + [{"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + fused=True, + ) + optimizer_muon = Muon( + matrix_params, + lr=args.matrix_lr, + momentum=args.muon_momentum, + backend_steps=args.muon_backend_steps, + ) + for group in optimizer_muon.param_groups: + group["base_lr"] = args.matrix_lr + optimizer_scalar = torch.optim.Adam( + [{"params": scalar_params, "lr": args.scalar_lr, "base_lr": args.scalar_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + fused=True, + ) + optimizers: list[torch.optim.Optimizer] = [optimizer_tok, optimizer_muon, optimizer_scalar] + if base_model.lm_head is not None: + optimizer_head = torch.optim.Adam( + [{"params": [base_model.lm_head.weight], "lr": args.head_lr, "base_lr": args.head_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + fused=True, + ) + optimizers.insert(1, optimizer_head) + + n_params = sum(p.numel() for p in base_model.parameters()) + log0(f"model_params:{n_params}") + log0(f"world_size:{world_size} grad_accum_steps:{grad_accum_steps}") + log0("sdp_backends:cudnn=False flash=True mem_efficient=False math=False") + log0(f"attention_mode:gqa num_heads:{args.num_heads} num_kv_heads:{args.num_kv_heads}") + log0( + f"tie_embeddings:{args.tie_embeddings} embed_lr:{token_lr} " + f"head_lr:{args.head_lr if base_model.lm_head is not None else 0.0} " + f"matrix_lr:{args.matrix_lr} scalar_lr:{args.scalar_lr}" + ) + log0( + f"train_batch_tokens:{args.train_batch_tokens} train_seq_len:{args.train_seq_len} " + f"iterations:{args.iterations} warmup_steps:{args.warmup_steps} " + f"max_wallclock_seconds:{args.max_wallclock_seconds:.3f}" + ) + log0(f"seed:{args.seed}") + + # ----------------------------- + # DATA LOADER & MODEL WARMUP + # ----------------------------- + + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + + def zero_grad_all() -> None: + for opt in optimizers: + opt.zero_grad(set_to_none=True) + + max_wallclock_ms = 1000.0 * args.max_wallclock_seconds if args.max_wallclock_seconds > 0 else None + + def lr_mul(step: int, elapsed_ms: float) -> float: + if args.warmdown_iters <= 0: + return 1.0 + if max_wallclock_ms is None: + warmdown_start = max(args.iterations - args.warmdown_iters, 0) + return max((args.iterations - step) / max(args.warmdown_iters, 1), 0.0) if warmdown_start <= step < args.iterations else 1.0 + step_ms = elapsed_ms / max(step, 1) + warmdown_ms = args.warmdown_iters * step_ms + remaining_ms = max(max_wallclock_ms - elapsed_ms, 0.0) + return remaining_ms / max(warmdown_ms, 1e-9) if remaining_ms <= warmdown_ms else 1.0 + + # Warmup primes the compiled forward/backward/optimizer paths, then we restore the + # initial weights/optimizer state so measured training starts from the true init. + if args.warmup_steps > 0: + initial_model_state = {name: tensor.detach().cpu().clone() for name, tensor in base_model.state_dict().items()} + initial_optimizer_states = [copy.deepcopy(opt.state_dict()) for opt in optimizers] + model.train() + for warmup_step in range(args.warmup_steps): + zero_grad_all() + for micro_step in range(grad_accum_steps): + if distributed: + model.require_backward_grad_sync = micro_step == grad_accum_steps - 1 + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + warmup_loss = model(x, y) + (warmup_loss * grad_scale).backward() + for opt in optimizers: + opt.step() + zero_grad_all() + if args.warmup_steps <= 20 or (warmup_step + 1) % 10 == 0 or warmup_step + 1 == args.warmup_steps: + log0(f"warmup_step:{warmup_step + 1}/{args.warmup_steps}") + base_model.load_state_dict(initial_model_state, strict=True) + for opt, state in zip(optimizers, initial_optimizer_states, strict=True): + opt.load_state_dict(state) + zero_grad_all() + if distributed: + model.require_backward_grad_sync = True + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + + # ----------------------------- + # MAIN TRAINING LOOP + # ----------------------------- + + training_time_ms = 0.0 + stop_after_step: int | None = None + torch.cuda.synchronize() + t0 = time.perf_counter() + + step = 0 + while True: + last_step = step == args.iterations or (stop_after_step is not None and step >= stop_after_step) + + should_validate = last_step or (args.val_loss_every > 0 and step % args.val_loss_every == 0) + if should_validate: + torch.cuda.synchronize() + training_time_ms += 1000.0 * (time.perf_counter() - t0) + val_loss, val_bpb = eval_val( + args, + model, + rank, + world_size, + device, + grad_accum_steps, + val_tokens, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + ) + log0( + f"step:{step}/{args.iterations} val_loss:{val_loss:.4f} val_bpb:{val_bpb:.4f} " + f"train_time:{training_time_ms:.0f}ms step_avg:{training_time_ms / max(step, 1):.2f}ms" + ) + torch.cuda.synchronize() + t0 = time.perf_counter() + + if last_step: + if stop_after_step is not None and step < args.iterations: + log0( + f"stopping_early: wallclock_cap train_time:{training_time_ms:.0f}ms " + f"step:{step}/{args.iterations}" + ) + break + + elapsed_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + scale = lr_mul(step, elapsed_ms) + zero_grad_all() + train_loss = torch.zeros((), device=device) + for micro_step in range(grad_accum_steps): + if distributed: + model.require_backward_grad_sync = micro_step == grad_accum_steps - 1 + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + loss = model(x, y) + train_loss += loss.detach() + (loss * grad_scale).backward() + train_loss /= grad_accum_steps + + frac = min(step / args.muon_momentum_warmup_steps, 1.0) if args.muon_momentum_warmup_steps > 0 else 1.0 + muon_momentum = (1 - frac) * args.muon_momentum_warmup_start + frac * args.muon_momentum + for group in optimizer_muon.param_groups: + group["momentum"] = muon_momentum + + for opt in optimizers: + for group in opt.param_groups: + group["lr"] = group["base_lr"] * scale + + if args.grad_clip_norm > 0: + torch.nn.utils.clip_grad_norm_(base_model.parameters(), args.grad_clip_norm) + for opt in optimizers: + opt.step() + zero_grad_all() + + step += 1 + approx_training_time_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + should_log_train = ( + args.train_log_every > 0 + and (step <= 10 or step % args.train_log_every == 0 or stop_after_step is not None) + ) + if should_log_train: + log0( + f"step:{step}/{args.iterations} train_loss:{train_loss.item():.4f} " + f"train_time:{approx_training_time_ms:.0f}ms step_avg:{approx_training_time_ms / step:.2f}ms" + ) + + # Needed to sync whether we've reached the wallclock cap. + reached_cap = max_wallclock_ms is not None and approx_training_time_ms >= max_wallclock_ms + if distributed and max_wallclock_ms is not None: + reached_cap_tensor = torch.tensor(int(reached_cap), device=device) + dist.all_reduce(reached_cap_tensor, op=dist.ReduceOp.MAX) + reached_cap = bool(reached_cap_tensor.item()) + if stop_after_step is None and reached_cap: + stop_after_step = step + + log0( + f"peak memory allocated: {torch.cuda.max_memory_allocated() // 1024 // 1024} MiB " + f"reserved: {torch.cuda.max_memory_reserved() // 1024 // 1024} MiB" + ) + + # ----------------------------- + # SERIALIZATION + ROUNDTRIP VALIDATION + # ----------------------------- + # Save the raw state (useful for debugging/loading in PyTorch directly), then always produce + # the compressed int8+zlib artifact and validate the round-tripped weights. + + if master_process: + torch.save(base_model.state_dict(), "final_model.pt") + model_bytes = os.path.getsize("final_model.pt") + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model: {model_bytes} bytes") + log0(f"Code size: {code_bytes} bytes") + log0(f"Total submission size: {model_bytes + code_bytes} bytes") + + quant_obj, quant_stats = quantize_state_dict_int8(base_model.state_dict()) + quant_buf = io.BytesIO() + torch.save(quant_obj, quant_buf) + quant_raw = quant_buf.getvalue() + quant_blob = zlib.compress(quant_raw, level=9) + quant_raw_bytes = len(quant_raw) + if master_process: + with open("final_model.int8.ptz", "wb") as f: + f.write(quant_blob) + quant_file_bytes = os.path.getsize("final_model.int8.ptz") + code_bytes = len(code.encode("utf-8")) + ratio = quant_stats["baseline_tensor_bytes"] / max(quant_stats["int8_payload_bytes"], 1) + log0( + f"Serialized model int8+zlib: {quant_file_bytes} bytes " + f"(payload:{quant_stats['int8_payload_bytes']} raw_torch:{quant_raw_bytes} payload_ratio:{ratio:.2f}x)" + ) + log0(f"Total submission size int8+zlib: {quant_file_bytes + code_bytes} bytes") + + if distributed: + dist.barrier() + with open("final_model.int8.ptz", "rb") as f: + quant_blob_disk = f.read() + quant_state = torch.load(io.BytesIO(zlib.decompress(quant_blob_disk)), map_location="cpu") + base_model.load_state_dict(dequantize_state_dict_int8(quant_state), strict=True) + torch.cuda.synchronize() + t_qeval = time.perf_counter() + q_val_loss, q_val_bpb = eval_val( + args, + model, + rank, + world_size, + device, + grad_accum_steps, + val_tokens, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0( + f"final_int8_zlib_roundtrip val_loss:{q_val_loss:.4f} val_bpb:{q_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_qeval):.0f}ms" + ) + log0(f"final_int8_zlib_roundtrip_exact val_loss:{q_val_loss:.8f} val_bpb:{q_val_bpb:.8f}") + + # LoRA test-time training evaluation (the competition score) + torch._dynamo.reset() + torch.cuda.synchronize() + t_ttt = time.perf_counter() + ttt_val_loss, ttt_val_bpb = eval_val_ttt_lora( + args, base_model, rank, world_size, device, + base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0( + f"final_int8_ttt_lora val_loss:{ttt_val_loss:.4f} val_bpb:{ttt_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_ttt):.0f}ms" + ) + + if distributed: + dist.destroy_process_group() + + +if __name__ == "__main__": + main()