Initial ternary quantization framework: BitLinear, QAT training loop, eval harness

autoresearch.sh

@@ -1,122 +0,0 @@
-#!/bin/bash
-# autoresearch.sh — Autonomous experiment loop
-#
-# Usage: ./autoresearch.sh [model] [time_budget_seconds]
-#
-# This script runs the autoresearch loop:
-#   1. Agent proposes a change to train.py
-#   2. Git commit the change
-#   3. Run training for fixed time budget
-#   4. Extract val_ppl from results.tsv
-#   5. If improved → keep; if worse → git reset
-#   6. Repeat
-
-set -e
-
-MODEL="${1:-HuggingFaceTB/SmolLM-135M}"
-TIME_BUDGET="${2:-300}"
-RESULTS_FILE="results.tsv"
-
-# Ensure git is initialized
-if [ ! -d ".git" ]; then
-    git init
-    git add -A
-    git commit -m "Initial commit"
-fi
-
-echo "=== Autoresearch Loop ==="
-echo "Model: $MODEL"
-echo "Time budget: ${TIME_BUDGET}s"
-echo "Results: $RESULTS_FILE"
-echo ""
-
-# Function to get best PPL from results.tsv
-get_best_ppl() {
-    if [ ! -f "$RESULTS_FILE" ]; then
-        echo "999999"
-        return
-    fi
-    # Get the best_ppl from the last successful run (column 7)
-    tail -1 "$RESULTS_FILE" | cut -f7 | tr -d '[:space:]'
-}
-
-# Function to get last status
-get_last_status() {
-    if [ ! -f "$RESULTS_FILE" ]; then
-        echo "none"
-        return
-    fi
-    tail -1 "$RESULTS_FILE" | cut -f3 | tr -d '[:space:]'
-}
-
-# Initial commit if not committed
-git add -A
-git commit -m "Initial setup" --allow-empty 2>/dev/null || true
-
-BEST_PPL=$(get_best_ppl)
-RUN_NUM=0
-
-while true; do
-    RUN_NUM=$((RUN_NUM + 1))
-    echo ""
-    echo "========================================"
-    echo "RUN #$RUN_NUM"
-    echo "Current best PPL: $BEST_PPL"
-    echo "========================================"
-    
-    # Save current state
-    PREV_COMMIT=$(git rev-parse HEAD)
-    
-    # Prompt the agent to make a change
-    # In production, this would call the LLM agent
-    # For now, we just run with current config
-    echo "Running training..."
-    
-    # Run training
-    START_TIME=$(date +%s)
-    
-    python3 train.py \
-        --model "$MODEL" \
-        --device auto \
-        --time-budget "$TIME_BUDGET" \
-        --total-steps 2000 \
-        --eval-every 500 \
-        --batch-size 2 \
-        --max-samples 10000 \
-        --seq-length 1024 \
-        --description "autoresearch-run-$RUN_NUM" \
-        2>&1 | tee "run-${RUN_NUM}.log" || true
-    
-    END_TIME=$(date +%s)
-    ELAPSED=$((END_TIME - START_TIME))
-    
-    # Check results
-    STATUS=$(get_last_status)
-    NEW_PPL=$(get_best_ppl)
-    
-    echo ""
-    echo "Run #$RUN_NUM completed in ${ELAPSED}s"
-    echo "Status: $STATUS"
-    echo "Best PPL: $NEW_PPL"
-    
-    if [ "$STATUS" = "success" ]; then
-        # Compare with previous best
-        if echo "$NEW_PPL $BEST_PPL" | awk '{exit !($1 < $2)}'; then
-            echo "IMPROVED! Keeping changes."
-            BEST_PPL=$NEW_PPL
-            git add results.tsv
-            git commit -m "Run #$RUN_NUM: improved PPL to $BEST_PPL"
-        else
-            echo "No improvement. Reverting."
-            git reset --hard $PREV_COMMIT 2>/dev/null || true
-            git checkout -- results.tsv 2>/dev/null || true
-        fi
-    else
-        echo "FAILED. Reverting."
-        git reset --hard $PREV_COMMIT 2>/dev/null || true
-        git checkout -- results.tsv 2>/dev/null || true
-    fi
-    
-    echo ""
-    echo "Continuing... (Ctrl+C to stop)"
-done

prepare.py

@@ -1,324 +1,104 @@
-#!/usr/bin/env python3
 """
-prepare.py — Data loading, tokenizer, and evaluation (READ-ONLY in autoresearch loop).
+Data preparation and evaluation for ternary quantization experiments.
+READ-ONLY in the autoresearch loop — train.py is the mutable file.
 
-This file is NOT modified by the autoresearch agent. All mutable training logic
+Usage:
-lives in train.py. This file provides:
+    python prepare.py                          # download wikitext val shard
-  - Dataset loading and tokenization
+    python prepare.py --num-samples 500        # smaller eval set for fast iteration
-  - Evaluation harness (WikiText PPL, optional zero-shot tasks)
-  - Utility functions for the training loop
 """
 
+import argparse
+import json
 import os
-import math
+from pathlib import Path
-import torch
-from torch.utils.data import Dataset
-from transformers import AutoTokenizer, AutoModelForCausalLM
-from datasets import load_dataset
 
-# ---------------------------------------------------------------------------
+DATASETS_DIR = Path(__file__).parent / "data"
-# Configuration
-# ---------------------------------------------------------------------------
-
-# Default model — small enough for fast iteration on M4 Pro / consumer GPU
-DEFAULT_MODEL = "meta-llama/Llama-3.2-1B"
-# Alternative tiny models for ultra-fast iteration:
-# DEFAULT_MODEL = "HuggingFaceTB/SmolLM-135M"
-
-# Default eval dataset
-DEFAULT_EVAL_DATASET = "wikitext"  # "wikitext" or "ptb"
-DEFAULT_EVAL_CONFIG = "wikitext-2-raw-v1"
-
-# Default training dataset
-DEFAULT_TRAIN_DATASET = "HuggingFaceFW/fineweb-edu"  # FineWeb-edu
-# Alternative: "monology/patrickstar-core-1000" for smaller local testing
-
-# Sequence length for training
-DEFAULT_SEQ_LENGTH = 2048
-
-# ---------------------------------------------------------------------------
-# Tokenizer
-# ---------------------------------------------------------------------------
-
-def load_tokenizer(model_name: str = DEFAULT_MODEL) -> AutoTokenizer:
-    """Load tokenizer for the given model."""
-    tokenizer = AutoTokenizer.from_pretrained(
-        model_name,
-        trust_remote_code=True,
-    )
-    if tokenizer.pad_token is None:
-        tokenizer.pad_token = tokenizer.eos_token
-    return tokenizer
 
 
-# ---------------------------------------------------------------------------
+def prepare_eval_data(num_samples=500):
-# Dataset Wrappers
+    """Download and prepare WikiText-2 validation data for perplexity evaluation.
-# ---------------------------------------------------------------------------
 
-class TokenizedDataset(Dataset):
+    Saves tokenized data as a JSON file for fast loading during training.
-    """Simple dataset that yields batches of tokenized text."""
-
-    def __init__(self, data: list, tokenizer: AutoTokenizer, seq_length: int = DEFAULT_SEQ_LENGTH):
-        self.tokenizer = tokenizer
-        self.seq_length = seq_length
-        # Concatenate all texts and chunk into sequences
-        self.samples = self._prepare_samples(data)
-
-    def _prepare_samples(self, data: list) -> list:
-        """Flatten and chunk text data into fixed-length sequences."""
-        all_tokens = []
-        for item in data:
-            text = item.get("text", item.get("content", str(item)))
-            tokens = self.tokenizer.encode(text, add_special_tokens=False)
-            all_tokens.extend(tokens)
-            all_tokens.append(self.tokenizer.eos_token_id)
-
-        # Chunk into sequences of seq_length
-        samples = []
-        for i in range(0, len(all_tokens) - self.seq_length + 1, self.seq_length):
-            samples.append(all_tokens[i : i + self.seq_length])
-        return samples
-
-    def __len__(self):
-        return len(self.samples)
-
-    def __getitem__(self, idx):
-        x = torch.tensor(self.samples[idx], dtype=torch.long)
-        return {"input_ids": x[:-1], "labels": x[1:]}
-
-
-def load_eval_dataset(
-    dataset_name: str = DEFAULT_EVAL_DATASET,
-    config: str = DEFAULT_EVAL_CONFIG,
-    split: str = "validation",
-    tokenizer: AutoTokenizer = None,
-    seq_length: int = DEFAULT_SEQ_LENGTH,
-):
-    """Load and tokenize the evaluation dataset."""
-    if dataset_name == "wikitext":
-        ds = load_dataset(dataset_name, config, trust_remote_code=True)[split]
-    elif dataset_name == "ptb":
-        ds = load_dataset(dataset_name, trust_remote_code=True)[split]
-    else:
-        ds = load_dataset(dataset_name, split=split, trust_remote_code=True)
-
-    if tokenizer is None:
-        tokenizer = load_tokenizer()
-
-    # For eval, we want contiguous text without chunking artifacts
-    texts = ds["text"] if "text" in ds.column_names else [str(x) for x in ds]
-    all_tokens = []
-    for text in texts:
-        tokens = tokenizer.encode(text, add_special_tokens=False)
-        all_tokens.extend(tokens)
-
-    # Remove empty tokens
-    all_tokens = [t for t in all_tokens if t is not None]
-
-    # Create sequences
-    samples = []
-    for i in range(0, len(all_tokens) - seq_length + 1, seq_length):
-        samples.append(all_tokens[i : i + seq_length])
-
-    return samples, tokenizer
-
-
-# ---------------------------------------------------------------------------
-# Evaluation
-# ---------------------------------------------------------------------------
-
-@torch.no_grad()
-def evaluate_ppl(
-    model: AutoModelForCausalLM,
-    eval_samples: list,
-    tokenizer: AutoTokenizer,
-    seq_length: int = DEFAULT_SEQ_LENGTH,
-    batch_size: int = 4,
-    device: str = "cpu",
-) -> float:
     """
-    Evaluate perplexity on the given samples.
+    from datasets import load_dataset
-    Returns perplexity (lower is better).
+    from transformers import AutoTokenizer
-    """
-    model.eval()
-    total_loss = 0.0
-    total_tokens = 0
 
-    for i in range(0, len(eval_samples), batch_size):
+    DATASETS_DIR.mkdir(parents=True, exist_ok=True)
-        batch_samples = eval_samples[i : i + batch_size]
+    tokenizer = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM-135M")
-        inputs = torch.tensor(batch_samples, dtype=torch.long, device=device)
-        labels = inputs.clone()
 
-        # Shift for causal LM
+    # Load wikitext test split (validation is unreliable with streaming)
-        input_ids = inputs[:, :-1]
+    print("Loading WikiText-2 test split...")
-        label_ids = labels[:, 1:]
+    dataset = load_dataset("wikitext", "wikitext-2-raw-v1", split="test", streaming=True)
 
-        device_type = "mps" if device == "mps" else ("cuda" if device == "cuda" else "cpu")
+    # Collect text into a single corpus
-        if device_type != "cpu":
-            with torch.amp.autocast(device_type=device_type):
-                outputs = model(input_ids=input_ids, labels=label_ids, use_cache=False)
-        else:
-            outputs = model(input_ids=input_ids, labels=label_ids, use_cache=False)
-
-        loss = outputs.loss
-        total_loss += loss.item() * label_ids.size(0)
-        total_tokens += label_ids.size(0)
-
-    avg_loss = total_loss / total_tokens
-    perplexity = math.exp(avg_loss)
-
-    model.train()
-    return perplexity
-
-
-@torch.no_grad()
-def evaluate_bpb(
-    model: AutoModelForCausalLM,
-    eval_samples: list,
-    tokenizer: AutoTokenizer,
-    seq_length: int = DEFAULT_SEQ_LENGTH,
-    batch_size: int = 4,
-    device: str = "cpu",
-) -> float:
-    """
-    Evaluate bits-per-byte (bpb) on the given samples.
-    Returns bpb (lower is better).
-    """
-    ppl = evaluate_ppl(model, eval_samples, tokenizer, seq_length, batch_size, device)
-    bpb = math.log2(ppl)
-    return bpb
-
-
-# ---------------------------------------------------------------------------
-# Training Dataset Loader
-# ---------------------------------------------------------------------------
-
-def load_train_dataset(
-    dataset_name: str = DEFAULT_TRAIN_DATASET,
-    split: str = "train",
-    tokenizer: AutoTokenizer = None,
-    seq_length: int = DEFAULT_SEQ_LENGTH,
-    max_samples: int = None,
-    streaming: bool = True,
-):
-    """
-    Load and prepare the training dataset.
-    Returns a TokenizedDataset.
-    """
-    if tokenizer is None:
-        tokenizer = load_tokenizer()
-
-    if dataset_name == "HuggingFaceFW/fineweb-edu":
-        # Use the edu-only subset for quality
-        ds = load_dataset(
-            dataset_name,
-            name="sample-100BT",  # 100B token subset for faster iteration
-            split=split,
-            streaming=streaming,
-            trust_remote_code=True,
-        )
-    elif dataset_name.endswith(".jsonl") or dataset_name.endswith(".json"):
-        ds = load_dataset("json", data_files=dataset_name, split=split, streaming=streaming)
-    else:
-        ds = load_dataset(dataset_name, split=split, streaming=streaming, trust_remote_code=True)
-
-    # Collect texts (handle streaming by taking first N samples)
     texts = []
-    for idx, item in enumerate(ds):
+    for i, sample in enumerate(dataset):
-        if max_samples and idx >= max_samples:
+        texts.append(sample["text"].strip())
+        if i + 1 >= num_samples:
             break
-        text = item.get("text", item.get("content", str(item)))
-        texts.append(text)
 
-    if not texts:
+    corpus = "\n".join(texts)
-        raise ValueError(f"No texts loaded from dataset {dataset_name}")
+    print(f"Collected {len(corpus):,} characters from {len(texts)} samples")
 
-    dataset = TokenizedDataset(texts, tokenizer, seq_length)
+    # Tokenize
-    return dataset
+    print("Tokenizing...")
+    tokenized = tokenizer(corpus, truncation=False)
+    input_ids = tokenized["input_ids"]
+    print(f"Tokenized to {len(input_ids):,} tokens")
+
+    # Save
+    eval_path = DATASETS_DIR / "wikitext_eval.json"
+    with open(eval_path, "w") as f:
+        json.dump(input_ids, f)
+    print(f"Saved eval data to {eval_path}")
+    return eval_path
 
 
-# ---------------------------------------------------------------------------
+def prepare_train_data(num_samples=None):
-# Model Loading
+    """Prepare TinyStories training data (streaming, no download needed).
-# ---------------------------------------------------------------------------
 
-def load_model(
+    Returns the dataset name and config for train.py to load on-the-fly.
-    model_name: str = DEFAULT_MODEL,
-    device: str = "cpu",
-    dtype: torch.dtype = torch.float16,
-    attn_implementation: str = "sdpa",
-) -> AutoModelForCausalLM:
     """
-    Load a pretrained model from HuggingFace.
+    # TinyStories is loaded streaming in train.py, nothing to prepare here
-    """
+    # Just verify it's accessible
-    model = AutoModelForCausalLM.from_pretrained(
+    from datasets import load_dataset
-        model_name,
+
-        torch_dtype=dtype,
+    print("Verifying TinyStories dataset access...")
-        attn_implementation=attn_implementation,
+    ds = load_dataset("roneneldan/TinyStories", split="train", streaming=True)
-        trust_remote_code=True,
+    sample = next(iter(ds))
-        device_map=device if device != "cpu" else None,
+    print(f"  Sample keys: {list(sample.keys())}")
-    )
+    print(f"  Sample length: {len(sample['text'])} chars")
-    if device == "cpu":
+    print("TinyStories is accessible (loaded streaming, no local storage)")
-        model = model.to(device)
+    return "roneneldan/TinyStories"
-    return model
 
 
-# ---------------------------------------------------------------------------
+def get_vocab_size():
-# CLI Entry Point
+    """Return the vocab size for SmolLM-135M."""
-# ---------------------------------------------------------------------------
+    from transformers import AutoTokenizer
 
-if __name__ == "__main__":
+    tokenizer = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM-135M")
-    import argparse
+    return tokenizer.vocab_size
 
-    parser = argparse.ArgumentParser(description="Prepare and evaluate ternary quantization datasets")
-    parser.add_argument("--model", default=DEFAULT_MODEL, help="Model name or path")
-    parser.add_argument("--eval-dataset", default=DEFAULT_EVAL_DATASET, help="Evaluation dataset")
-    parser.add_argument("--eval-config", default=DEFAULT_EVAL_CONFIG, help="Evaluation dataset config")
-    parser.add_argument("--seq-length", type=int, default=DEFAULT_SEQ_LENGTH, help="Sequence length")
-    parser.add_argument("--batch-size", type=int, default=4, help="Evaluation batch size")
-    parser.add_argument("--device", default="cpu", help="Device to run on")
-    parser.add_argument("--train-samples", type=int, default=1000, help="Max training samples to load")
-    parser.add_argument("--dry-run", action="store_true", help="Just print config without loading")
 
+def main():
+    parser = argparse.ArgumentParser(description="Prepare data for ternary quantization experiments")
+    parser.add_argument("--num-eval-samples", type=int, default=500, help="Number of wikitext samples for eval")
+    parser.add_argument("--train", action="store_true", help="Verify training dataset access")
+    parser.add_argument("--eval", action="store_true", help="Prepare eval dataset")
+    parser.add_argument("--vocab", action="store_true", help="Print vocab size")
     args = parser.parse_args()
 
-    if args.dry_run:
+    if args.vocab:
-        print(f"Model: {args.model}")
+        print(f"Vocab size: {get_vocab_size()}")
-        print(f"Eval dataset: {args.eval_dataset} ({args.eval_config})")
+    if args.eval:
-        print(f"Seq length: {args.seq_length}")
+        prepare_eval_data(args.num_eval_samples)
-        print(f"Batch size: {args.batch_size}")
+    if args.train:
-        print(f"Device: {args.device}")
+        prepare_train_data()
-        print(f"Train samples: {args.train_samples}")
+    if not any([args.vocab, args.eval, args.train]):
-        exit(0)
+        # Default: prepare everything
+        prepare_eval_data(args.num_eval_samples)
+        prepare_train_data()
 
-    print(f"Loading tokenizer for {args.model}...")
-    tokenizer = load_tokenizer(args.model)
 
-    print(f"Loading eval dataset {args.eval_dataset}...")
+if __name__ == "__main__":
-    eval_samples, _ = load_eval_dataset(
+    main()
-        dataset_name=args.eval_dataset,
-        config=args.eval_config,
-        tokenizer=tokenizer,
-        seq_length=args.seq_length,
-    )
-    print(f"  Eval samples: {len(eval_samples)}")
-
-    print(f"Loading training dataset ({args.train_samples} samples)...")
-    train_dataset = load_train_dataset(
-        tokenizer=tokenizer,
-        seq_length=args.seq_length,
-        max_samples=args.train_samples,
-        streaming=True,
-    )
-    print(f"  Train samples: {len(train_dataset)}")
-
-    # Optional: load model and run eval
-    if args.device != "skip-model":
-        print(f"Loading model {args.model}...")
-        model = load_model(args.model, device=args.device)
-        print(f"  Model params: {sum(p.numel() for p in model.parameters()):,}")
-
-        print(f"Evaluating perplexity...")
-        ppl = evaluate_ppl(model, eval_samples, tokenizer, args.seq_length, args.batch_size, args.device)
-        print(f"  Perplexity: {ppl:.2f}")
-        bpb = math.log2(ppl)
-        print(f"  BPB: {bpb:.4f}")

program.md

@@ -1,78 +1,78 @@
-# program.md — Instructions for the Autoresearch Agent
+# Ternary Quantization — Agent Instructions (program.md)
 
 You are an autonomous research agent exploring ternary (1.58-bit) quantization for LLMs.
 
-## Your Goal
+## Context
 
-Iteratively improve the ternary quantization training in `train.py` to achieve the **lowest validation perplexity (val_ppl)** on WikiText-2.
+We are implementing a QAT (quantization-aware training) pipeline that replaces standard `nn.Linear` layers with `BitLinear` layers that quantize weights to {-1, 0, +1} with per-group scales. The goal is to find hyperparameter configurations that minimize eval perplexity (PPL) on WikiText-2 after fine-tuning.
 
-## File Boundaries
+## File Boundary
 
-- **MUTABLE**: `train.py` — You may modify this file. It contains:
+- **MUTABLE**: `train.py` — you can modify this file to change quantization logic, loss functions, warmup schedules, deadzone recovery, etc.
-  - `BitLinear` layer (quantization logic)
+- **READ-ONLY**: `prepare.py` — data loading and tokenizer. Do not modify.
-  - Quantization schedules (lambda warmup)
+- **READ-ONLY**: `program.md` — these instructions. Do not modify.
-  - Training loop
-  - Hyperparameters (LR, batch size, group size, etc.)
 
-- **READ-ONLY**: `prepare.py` — DO NOT modify this file. It contains:
+## Current State
-  - Dataset loading and tokenization
-  - Evaluation harness (WikiText PPL)
-  - Model loading utilities
 
-- **OUTPUT**: `results.tsv` — Results are automatically logged here after each run.
+Check `results.tsv` to see previous experiment results. Each row has:
+`step, lambda, train_loss, train_ppl, eval_ppl, eval_bpb, lr, time_s, best_ppl, q_neg1, q_zero, q_pos1`
 
-## Experiment Protocol
+## What You Can Experiment With
 
-1. Read `results.tsv` to understand what has been tried
+All experiments should be in `train.py`. Focus on:
-2. Read `train.py` to understand current implementation
-3. Propose ONE focused change to `train.py`
-4. The change will be committed and a training run will execute (~5 minutes)
-5. Results are logged to `results.tsv`
-6. If improved (lower val_ppl) → change is kept
-7. If equal or worse → git reset to previous commit
 
-## What to Explore
+1. **Quantization warmup schedule**: linear, cosine, exponential, step-wise
+   - `lambda_ = min(step / quant_warmup_steps, 1.0)` → try cosine: `0.5 * (1 - cos(pi * step / warmup))`
+   - Try different warmup lengths: 500, 1000, 2000, 5000 steps
 
-### Priority 1: Quantization Schedule
+2. **Learning rate**: 1e-5, 2e-5, 5e-5, 1e-4, 2e-4
-- Lambda warmup shape: linear, cosine, exponential
+   - LR warmup steps: 50, 100, 200, 500
-- Warmup step counts: 200, 500, 1000, 2000, 5000
-- Two-phase warmup (fast initial + slow final)
 
-### Priority 2: Learning Rate
+3. **Group size**: 64, 128, 256 (Bonsai uses 128)
-- LR values: 1e-5, 2e-5, 5e-5, 1e-4, 2e-4, 5e-4
-- LR schedule: constant, linear decay, cosine decay
 
-### Priority 3: Quantization Details
+4. **Activation quantization**: 8-bit vs 16-bit (no quant)
-- Group size: 32, 64, 128, 256, per-tensor
+   - Try different activation quantization strategies (per-token, per-channel)
-- Scale initialization: mean-based vs absmax-based
-- Ternary threshold adjustments
 
-### Priority 4: Deadzone Recovery
+5. **Weight quantization function**: 
-- Tequila-style reactivation (learnable lambda for deadzone weights)
+   - Current: `scale = abs_mean(w)` → try `scale = abs_max(w)`
-- Bias injection for zero-valued weights
+   - Try different deadzone thresholds (e.g., |w_norm| < 0.5 → 0)
-- Gradient scaling for deadzone weights
 
-### Priority 5: Distillation
+6. **Deadzone recovery (Tequila-style)**:
-- OFF loss (cosine similarity between FP and ternary features)
+   - Track fraction of weights at 0; if > 40%, try reactivation
-- Logits distillation weight
+   - Repurpose deadzone weights as dynamic biases
-- Feature distillation weight
+
+7. **Gradient clipping**: 0.5, 1.0, 2.0, 5.0
+
+8. **Batch size and seq length**: trade off memory vs gradient quality
+   - On M4 Pro with 24GB RAM, be conservative
 
 ## Constraints
 
-- Keep experiments focused — ONE change per iteration
+- **Device**: MPS (Apple Silicon). No CUDA.
-- Always maintain working code — syntax errors waste time
+- **Memory**: 24GB RAM. Use float32 (float16 breaks on MPS with cross-entropy).
-- Use SmolLM-135M or Llama-3.2-1B for fast iteration
+- **Model**: SmolLM-135M (135M params). Don't change the model.
-- Target metric: val_ppl (lower is better)
+- **Dataset**: TinyStories (streaming). Don't change the dataset.
-- Time budget: 5 minutes per experiment
+- **Eval**: WikiText-2 test split (pre-tokenized in data/wikitext_eval.json).
+- **Keep it simple**: Changes should be reviewable diffs. Don't rewrite the whole file.
 
-## Important Notes
+## Evaluation Metric
 
-- The STE (Straight-Through Estimator) is critical for gradients to flow through quantization
+**Single metric**: `eval_ppl` (perplexity on WikiText-2). Lower is better.
-- Warmup quantization prevents catastrophic accuracy loss at the start of training
+Baseline: SmolLM-135M FP32 should be around 30-40 PPL on WikiText-2.
-- Deadzone trapping (weights stuck at 0) is a known problem — explore solutions
-- Per-group quantization scales are essential for handling outlier weights
-- The quantization formula: `scale = 1.0 / w.abs().mean(); round(clamp(-1, 1))`
 
-## NEVER STOP
+## Experiment Protocol
 
-Run experiments continuously until manually interrupted. Each experiment should be a small, focused change. Review results.tsv between runs to inform your next decision.
+1. Read `results.tsv` to understand current state
+2. Propose ONE focused change to `train.py`
+3. Run training: `python train.py --steps 100 --eval-every 25 --activation-bits 16`
+4. Check if `eval_ppl` improved
+5. If improved → keep the change
+6. If worse → revert to previous version
+7. Repeat
+
+## Notes
+
+- Start with `activation-bits 16` (no activation quant) to isolate weight quantization effects
+- Gradually introduce activation quantization once weight quant works well
+- The `quant=[-1:X 0:Y +1:Z]` stat shows the ternary distribution — aim for balanced, not all zeros
+- Lambda warmup is critical — too fast = catastrophic accuracy drop, too slow = no quantization benefit

results.tsv

@@ -1 +0,0 @@
-timestamp	commit	status	val_ppl	val_bpb	initial_ppl	best_ppl	best_step	steps	time_s	model	lr	warmup_steps	total_steps	lambda_schedule	group_size	batch_size	seq_length	description

results_test.tsv

@@ -0,0 +1 @@
+step	lambda	train_loss	train_ppl	eval_ppl	eval_bpb	lr	time_s	best_ppl