README.md

---
license: odc-by
task_categories:
  - text-generation
language:
  - en
pretty_name: Parameter Golf Competition Data v2
size_categories:
  - 1B<n<10B
tags:
  - parameter-golf
  - fineweb
  - language-modeling
  - competition
---

# Parameter Golf Competition Data

Pre-tokenized [FineWeb](https://huggingface.co/datasets/HuggingFaceFW/fineweb) shards for the [OpenAI Parameter Golf](https://github.com/openai/parameter-golf) competition. Multiple SentencePiece vocab sizes plus a corrected byte-exact Scylla (TokenMonster) tokenization. Free checkpoint persistence API. Zero setup friction.

---

## ⚠️ Important: Scylla v1 Deprecated

The original `fineweb_scylla/` directory uses the 998-token vocab from [PR #1143](https://github.com/openai/parameter-golf/pull/1143). That vocab's byte-accounting metadata treated TokenMonster tokens as context-free, which overcounts source bytes by ~4%. Any `val_bpb` reported through the standard pipeline on `fineweb_scylla/` is inflated by roughly the same factor.

The bug is tracked in [Issue #897](https://github.com/openai/parameter-golf/issues/897) and corrected in [PR #1314](https://github.com/openai/parameter-golf/pull/1314) (simon-marcus, "Scylla: Corrected Byte-Exact Tokenizer Path"). The corrected path uses a full byte-native TokenMonster regime (`charset=none`, `capcode=0`, `normalization=none`, explicit 0x00–0xFF byte fallback) and is byte-exact on the fixed FineWeb validation text.

**Use `fineweb_scylla_v2/` for any new work.** Old `fineweb_scylla/` is kept for reproducibility of past runs and will not be deleted, but should not be used for new BPB comparisons.

---

## How It Works (The Simple Version)

Think of it like plumbing:

1. **The reservoir** is this dataset — competition data, pre-processed and ready to flow.
2. **The pipe** is `huggingface-cli download` — one command, and data flows to your GPU pod. Fast, resumable. If the pipe breaks mid-transfer, reconnect and it picks up where it left off.
3. **Your pod** is the sink — data arrives at `/workspace/data/`, ready to use. No processing, no conversion, no waiting.
4. **The safety valve** is checkpoint persistence — every N steps, your training progress flows out to cloud storage. Pod dies? New pod picks up the flow from the last save. No lost work.

That's it. Data flows in. Checkpoints flow out. You train in between.

### Step by Step

**I just want to train. What do I do?**

```bash
# Step 1: Install huggingface-cli (if you don't have it)
pip install huggingface-hub

# Step 2: Download the competition data (SP1024 default)
huggingface-cli download LightSpeedUp/parameter-golf-data --include "fineweb_sp1024/*" --local-dir /workspace/data

# Step 3: That's it. Train.
python train_gpt.py --data_dir /workspace/data/fineweb_sp1024
```

**I want a bigger vocab:**

```bash
# SP4096 — good middle ground
huggingface-cli download LightSpeedUp/parameter-golf-data --include "fineweb_sp4096/*" --local-dir /workspace/data

# SP8192 — a step up in vocab capacity
huggingface-cli download LightSpeedUp/parameter-golf-data --include "fineweb_sp8192/*" --local-dir /workspace/data
```

**I want to save checkpoints so I don't lose work:**

```bash
# After every N steps in your training script, save:
curl -X PUT \
  -H "Authorization: Bearer YOUR_GITHUB_TOKEN" \
  --data-binary @checkpoint.pt \
  https://pgolf-api.lightspeedup.com/put/YOUR_GITHUB_USERNAME/my-run/checkpoint.pt

# On a new pod, resume:
curl -o checkpoint.pt \
  -H "Authorization: Bearer YOUR_GITHUB_TOKEN" \
  "https://pgolf-api.lightspeedup.com/download?run_id=my-run&filename=checkpoint.pt"
```

**I want the fully automated experience:**

Use our RunPod template: `matotezitanka/proteus-pytorch:community`

Set these env vars before launch:
- `PGOLF_DATA=sp1024` (or `sp4096`, `sp8192`, `sp12288`, `sp16384`, or `scylla_v2`)
- `PGOLF_SHARDS=full` (or `mini` for testing)
- `PGOLF_GITHUB_TOKEN=ghp_yourtoken` (optional, for checkpoints)
- `PGOLF_USER=yourgithubname` (optional)
- `PGOLF_RUN=my-experiment-1` (optional)

Hit deploy. SSH in when it's ready. Everything is there.

---

## Comprehensive Guide

### Available Data

| Tokenizer | Vocab | Size | Use case |
|-----------|-------|------|----------|
| **SP1024** | 1024 tokens | ~15 GB | Competition default. Most PRs use this. |
| **SP4096** | 4096 tokens | ~12 GB | Larger vocab, shorter sequences per doc. |
| **SP8192** | 8192 tokens | ~11 GB | Common for bigram/mixer submissions. |
| **SP12288** | 12288 tokens | ~10 GB | Explores the 8k–16k vocab range. |
| **SP16384** | 16384 tokens | ~9 GB | Largest SentencePiece variant we publish. |
| **byte260** | 260 tokens | ~40 GB | Pure-byte tokenization. Bytes `0x00..0xFF` → ids `0..255` directly. Reserved specials: `pad=256, bos=257, eos=258, unk=259`. Encoding is `[257] + list(text.encode("utf-8"))`. No SentencePiece model involved. ~195 train + 2 val shards (byte density is ~4× SP1024). |
| **Scylla v2** | 1254 tokens | ~17 GB | **Corrected** TokenMonster byte-exact tokenizer (PR #1314). Leaderboard-comparable BPB. |
| ~~Scylla v1~~ | ~~998 tokens~~ | ~~~11 GB~~ | **Deprecated** — buggy byte accounting (Issue #897). Kept for reproducibility only. |

Each directory contains 80 training shards + 1 validation shard + tokenizer models.

### Download Options

```bash
# === DATA SELECTION ===

# SP1024 — competition default (~15 GB)
huggingface-cli download LightSpeedUp/parameter-golf-data --include "fineweb_sp1024/*" --local-dir /workspace/data

# SP4096 — 4k vocab (~12 GB)
huggingface-cli download LightSpeedUp/parameter-golf-data --include "fineweb_sp4096/*" --local-dir /workspace/data

# SP8192 — 8k vocab (~11 GB)
huggingface-cli download LightSpeedUp/parameter-golf-data --include "fineweb_sp8192/*" --local-dir /workspace/data

# SP12288 — 12k vocab (~10 GB)
huggingface-cli download LightSpeedUp/parameter-golf-data --include "fineweb_sp12288/*" --local-dir /workspace/data

# SP16384 — 16k vocab (~9 GB)
huggingface-cli download LightSpeedUp/parameter-golf-data --include "fineweb_sp16384/*" --local-dir /workspace/data

# byte260 — UTF-8 bytes + 4 reserved specials, no tokenizer training (~40 GB)
huggingface-cli download LightSpeedUp/parameter-golf-data --include "fineweb_byte260/*" --local-dir /workspace/data

# Scylla v2 — corrected TokenMonster 1254-token vocab (~17 GB)
huggingface-cli download LightSpeedUp/parameter-golf-data --include "fineweb_scylla_v2/*" --local-dir /workspace/data

# Legacy Scylla v1 — 998-token vocab, deprecated (use v2 instead)
huggingface-cli download LightSpeedUp/parameter-golf-data --include "fineweb_scylla/*" --local-dir /workspace/data

# All datasets + all tokenizers (~70 GB)
huggingface-cli download LightSpeedUp/parameter-golf-data --local-dir /workspace/data

# Just tokenizer models (tiny, < 2 MB)
huggingface-cli download LightSpeedUp/parameter-golf-data --include "tokenizers/*" --local-dir /workspace/data


# === SHARD SUBSETS (save time/bandwidth) ===

# Mini — 10 shards for smoke tests (~2 GB)
huggingface-cli download LightSpeedUp/parameter-golf-data \
    --include "fineweb_sp1024/fineweb_train_00000?.bin" \
    --include "fineweb_sp1024/fineweb_val*" \
    --local-dir /workspace/data

# Half — 40 shards (~7 GB)
huggingface-cli download LightSpeedUp/parameter-golf-data \
    --include "fineweb_sp1024/fineweb_train_0000[0-3]?.bin" \
    --include "fineweb_sp1024/fineweb_val*" \
    --local-dir /workspace/data

# Val only — just validation data (~200 MB)
huggingface-cli download LightSpeedUp/parameter-golf-data \
    --include "fineweb_sp1024/fineweb_val*" \
    --local-dir /workspace/data
```

**No HuggingFace account required.** This is a public dataset. No login, no token, no signup.

**Downloads are resumable.** If your connection drops, re-run the same command and it picks up where it left off.

### Checkpoint Persistence API

Save and resume training across pod preemptions. Your GitHub token is your identity — no accounts to create.

```bash
# === CHECKPOINT API (https://pgolf-api.lightspeedup.com) ===

# Upload a checkpoint
curl -X POST https://pgolf-api.lightspeedup.com/upload \
  -H "Authorization: Bearer ghp_yourtoken" \
  -H "Content-Type: application/json" \
  -d '{"run_id": "my-run", "filename": "checkpoint_step500.pt"}'

# Then PUT the file
curl -X PUT https://pgolf-api.lightspeedup.com/put/yourusername/my-run/checkpoint_step500.pt \
  -H "Authorization: Bearer ghp_yourtoken" \
  --data-binary @checkpoint_step500.pt

# Download a checkpoint
curl -o checkpoint.pt \
  -H "Authorization: Bearer ghp_yourtoken" \
  "https://pgolf-api.lightspeedup.com/download?run_id=my-run&filename=checkpoint_step500.pt"

# List your checkpoints
curl -H "Authorization: Bearer ghp_yourtoken" \
  "https://pgolf-api.lightspeedup.com/list?run_id=my-run"

# Delete a run's checkpoints
curl -X DELETE -H "Authorization: Bearer ghp_yourtoken" \
  "https://pgolf-api.lightspeedup.com/clean?run_id=my-run"
```

**Limits:** 10 checkpoints per user, 2 GB max each. Auto-deleted after 7 days.

### Docker Image

```bash
docker pull matotezitanka/proteus-pytorch:community
```

Includes: PyTorch 2.11.0 + CUDA 12.8 + Flash Attention 3 + all competition deps (brotli, tokenmonster, sentencepiece) + tools (cpu_test.py, retokenizer, swap_pytorch.sh) + automated boot script.

Works on RunPod, Vast.ai, or any Docker host with NVIDIA GPUs.

### Data Integrity

Every file has a SHA256 checksum. After downloading:

```bash
cd /workspace/data
sha256sum -c SHA256SUMS.txt
```

If any checksum fails, re-download that file. The download is resumable — you won't re-download files that are already correct.

### Dataset Structure

```
parameter-golf-data/
├── fineweb_sp1024/        # 80 train + 1 val, SentencePiece BPE 1024
├── fineweb_sp4096/        # 80 train + 1 val, SentencePiece BPE 4096
├── fineweb_sp8192/        # 80 train + 1 val, SentencePiece BPE 8192
├── fineweb_sp12288/       # 80 train + 1 val, SentencePiece BPE 12288
├── fineweb_sp16384/       # 80 train + 1 val, SentencePiece BPE 16384
├── fineweb_byte260/       # ~195 train + 2 val shards, pure-byte (PureByteTokenizer: pad=0, bos=1, eos=2, unk=3, bytes 4..259)
├── fineweb_scylla_v2/     # 80 train + 1 val, corrected TokenMonster 1254-token (PR #1314)
├── fineweb_scylla/        # DEPRECATED — 998-token buggy vocab, kept for reproducibility
├── tokenizers/
│   ├── fineweb_1024_bpe.model           (SP1024 SentencePiece)
│   ├── fineweb_4096_bpe.model           (SP4096 SentencePiece)
│   ├── fineweb_8192_bpe.model           (SP8192 SentencePiece)
│   ├── fineweb_12288_bpe.model          (SP12288 SentencePiece)
│   ├── fineweb_16384_bpe.model          (SP16384 SentencePiece)
│   ├── scylla_v2/scylla.yaml            (TokenMonster 1254, corrected)
│   ├── scylla_v2/scylla.meta.npz        (byte LUTs, byte-exact)
│   ├── scylla/candidate.vocab           (TokenMonster 998, deprecated)
│   └── scylla/candidate.meta.npz        (old byte LUTs, overcounts)
├── SHA256SUMS.txt
└── PATENTS.md
```

---

## Security & Privacy

We believe in transparency. Here's exactly what we can and can't see.

### What we CAN access (technically)
- **Your checkpoint files** — they're stored in our Cloudflare R2 bucket. We have admin access to the bucket. We don't look at them, but we could.
- **Your checkpoint metadata** — filenames, sizes, upload timestamps. This is visible in the R2 dashboard.
- **Request logs** — Cloudflare logs request metadata (IP addresses, timestamps, URLs) by default. We do not add any additional logging.
- **Your GitHub username** — extracted from your token to scope your storage namespace.

### What we CANNOT access
- **Your training code** — it runs on your pod, never touches our infrastructure.
- **Your model weights** (unless you upload them as a checkpoint).
- **Your GitHub token contents** — the token transits our Worker to call GitHub's API, but it is NOT stored, NOT logged, and NOT persisted anywhere. It's used once per request for authentication and discarded.
- **Other users' data** — the Worker enforces namespace isolation. Your GitHub username is your storage prefix. You cannot read, list, or delete another user's checkpoints.

### What we DO NOT do
- We do NOT sell, share, or analyze your data.
- We do NOT train models on your checkpoints.
- We do NOT log your GitHub token value.
- We do NOT track your usage beyond standard Cloudflare request metrics.

### What we disclose
- The checkpoint API Worker code is in our private repo. We plan to open-source it.
- Cloudflare's privacy policy applies to request metadata: https://www.cloudflare.com/privacypolicy/
- Checkpoints are automatically deleted after 7 days. We do not keep backups.

### If you don't trust us
That's fair. You can:
1. **Use just the HF dataset** — no account, no tokens, no interaction with our API. Just `huggingface-cli download`.
2. **Save checkpoints locally** — skip the API entirely. Save to `/workspace/` and accept the risk of losing work on preemption.
3. **Inspect the Worker** — we'll open-source it. Until then, the API surface is 5 endpoints, ~150 lines of JavaScript, zero dependencies.

---

## Provenance

- **Source:** [FineWeb](https://huggingface.co/datasets/HuggingFaceFW/fineweb) (CommonCrawl-derived, by Hugging Face). Our docs come from the `willdepueoai/parameter-golf` competition mirror.
- **SP1024 tokenization:** SentencePiece BPE, 1024 tokens — from the [openai/parameter-golf](https://github.com/openai/parameter-golf) competition repo, `data/tokenizer_specs.json`.
- **SP4096 / SP8192 / SP12288 / SP16384 tokenizations:** SentencePiece BPE models trained by LightSpeedUp on 1 M FineWeb docs decoded byte-exactly from the canonical SP1024 shards (SP1024 uses `byte_fallback=True`, so the decode is lossless). Trainer settings mirror the competition's `build_sentencepiece_tokenizer`: `character_coverage=0.999`, `byte_fallback=True`, `split_digits=True`, `normalization_rule_name=nmt_nfkc`, `add_dummy_prefix=False`, `hard_vocab_limit=False`.
- **byte260 tokenization:** direct UTF-8 byte mapping. Vocab 260 = 256 bytes + 4 reserved specials. Bytes `0x00..0xFF` map to ids `0..255` directly (no offset). Reserved specials: `pad_id=256, bos_id=257, eos_id=258, unk_id=259`. Encoding is `[257] + list(text.encode("utf-8"))`; `append_eos=False`. No SentencePiece model involved. Byte-accounting LUT is trivial: `base_bytes_lut[tok] = 1 if tok < 256 else 0`. **Note on convention:** this differs from the `openai/parameter-golf` `PureByteTokenizer` defaults (which put specials at 0..3 and offset bytes to 4..259). When loading these shards into the canonical `train_gpt.py --variant byte260` path, use a loader that matches the convention above.
- **Scylla v2 tokenization:** corrected byte-exact TokenMonster vocab (1254 tokens, logical 1178, `bos_id=1253`) from [@simon-marcus](https://github.com/simon-marcus)'s [PR #1314](https://github.com/openai/parameter-golf/pull/1314). Regime: `charset=none`, `capcode=0`, `normalization=none`, explicit full byte fallback, latin-1 byte interpretation, synthetic zero-byte BOS per doc.
- **Scylla v1 tokenization (deprecated):** original 998-token TokenMonster vocab from [@simon-marcus](https://github.com/simon-marcus)'s [PR #1143](https://github.com/openai/parameter-golf/pull/1143). Superseded by v2 due to the byte-accounting issue in [Issue #897](https://github.com/openai/parameter-golf/issues/897).
- **No modification** to token sequences on any variant — byte-identical to what you'd produce by running the same pipeline yourself.

### Attribution Chain

CommonCrawl (CC-BY) → Hugging Face FineWeb (ODC-By 1.0) → willdepueoai/parameter-golf (ODC-By 1.0) → This dataset (ODC-By 1.0)

## License

**Data:** [Open Data Commons Attribution License (ODC-By 1.0)](https://opendatacommons.org/licenses/by/1-0/)

**Code & Tools:** Apache 2.0 — see [PATENTS.md](PATENTS.md)

---

## Related Community Resources

- **[`sproos/parameter-golf-tokenizers`](https://huggingface.co/sproos/parameter-golf-tokenizers)** — a complementary community mirror that publishes `fineweb10B_{sp1024, sp2048, sp4096, sp8192}/` pre-tokenized shards plus the corresponding SentencePiece `.model` / `.vocab` files. If you only need the SP variants in the 1K–8K vocab range, sproos is a direct source. Our `LightSpeedUp/parameter-golf-data` drop is complementary: we add the larger SP variants (12288, 16384), the corrected Scylla v2, and byte260, and we bundle the free checkpoint-persistence API + Docker image for end-to-end training on free-tier GPUs.

## Roadmap

- Cloudflare R2 mirror for HF-rate-limited users (coming soon)
- Automated checkpoint save/resume in the boot script
- Open-source the CF Worker code

## Community

- [The Agora](https://matotezitanka.github.io/parameter-golf) — live leaderboard + compliance tracker
- [Issue #942](https://github.com/openai/parameter-golf/issues/942) — compute resources discussion
- [Issue #140](https://github.com/openai/parameter-golf/issues/140) — competition discussion thread