otter-chess

Otter: Skill-Conditioned Chess Move Prediction Model

Otter is a 15.3M parameter neural network trained to predict the moves of chess players at specific skill (Elo) levels. It is conditioned jointly on the game board state, move history, time control, remaining clock time, and the Elo ratings of both players.

In addition to move prediction (policy head), Otter is trained multi-task style to predict game outcomes (value head) and move metadata (auxiliary head) such as piece types, captures, check status, and origin/destination squares.

---

Model Details

• Developed by: Peargent Labs
• Model Type: Hybrid CNN + Transformer with Cross-Attention Fusion
• License: MIT
• Weights Format: Safetensors (model.safetensors)
• Website: Otter Site
• GitHub Repository: PeargentLabs/otter-chess
• W&B Report: Otter-3M-Run
• Paper: (Arxiv)

---

Architecture Highlight

                  +-----------------------+      +---------------------------+
                  |  Board Position FEN   |      |  Last K Moves (K=20)      |
                  |  (18 x 8 x 8 Tensor)  |      |  (Canonical UCI Strings)  |
                  +-----------+-----------+      +-------------+-------------+
                              |                                |
                              v                                v
                    [ 4x ResNet Blocks ]             [ 2-Layer Transformer ]
                              |                                |
                              v                                v
                     [ Board Tokens ]                 [ History Tokens ]
                     (64 x 256 Dim)                     (20 x 256 Dim)
                              |                                |
                              +--------------+-----------------+
                                             |
                                             v
                                  [ Cross-Attention Fusion ]
                                             |
                                             v
                             [ Skill-Conditioned Self-Attn ] <--- [ 640d Conditioning Vector ]
                             (4 blocks, conditioned on Elo,       (Elo, Opponent Elo,
                              TC, Clock, and Pooled History)       TC, Clock, Pooled History)
                                             |
                                             v
                                  [ Global Average Pooling ]
                                             |
                              +--------------+--------------+
                              |              |              |
                              v              v              v
                          [Policy]        [Value]        [Auxiliary]
                           (Move)        (Outcome)       (Metadata)

• Board Encoder: CNN backbone with 4 residual blocks, 2D factored position embeddings (rank + file), and channel-wise dropout.
• History Encoder: A 2-layer Transformer encoder processing the last $K=20$ moves.
• Conditioning Module: A 640-dimensional vector concatenating embeddings of active/opponent Elo ($11$ bins each), time control category ($5$ bins), remaining clock fraction (2-layer MLP), and mean-pooled history representations.
• Attention Fusion: One skill-conditioned cross-attention layer followed by four skill-conditioned self-attention blocks.
• Heads:
  • Policy Head: $4208$-dim output over all canonical UCI moves.
  • Value Head: Tanh projection to $[-1, +1]$ (predicted game outcome).
  • Auxiliary Head: $141$-dim multi-hot output (piece type moved, captures, check status, origin/destination squares).

---

Intended Uses & Limitations

Intended uses: player modeling / style emulation at a target Elo, move recommendation for training or commentary, win-probability estimation conditioned on both players' ratings and clock time.

Limitations: trained on rated Rapid games from Lichess — move distributions may not transfer well to Blitz/Bullet/Classical. Predicts human-like move probabilities, not optimal/engine-strength play.

---

Training Data & Methodology

• Source: Lichess open database monthly PGN dumps.
• Split: Train on Jan–Dec 2024 (Rated Rapid), validate on Jan 2025.
• Optimizer: AdamW, weight decay $1\times10^{-5}$.
• LR schedule: peak $1\times10^{-4}$, linear warmup (10% of steps), cosine decay to $1\times10^{-6}$.
• Loss weights: Policy 1.0, Value 0.25, Auxiliary 0.5.

---

Performance

Validation (Jan 2025): Top-1 accuracy 55.2%, Top-5 accuracy 91.0% (mean across Elo buckets, 100,000 validation samples per bucket).

---

Move Vocabulary

Moves are UCI strings mapped to integer IDs via vocab.json, shipped in this repo:

• vocab["history"] — maps prior moves in the game window to embedding IDs (4,209 entries, PAD = 0).
• vocab["policy"] — maps policy-head output indices back to UCI moves (4,208 entries).

Indices are shifted by 1 relative to each other: history reserves 0 for PAD (e.g. "a1a2" → 1), while policy indexes moves directly from 0 (e.g. "a1a2" → 0).

import json


with open("vocab.json") as f:
    vocab = json.load(f)

history_vocab = vocab["history"]
policy_vocab = vocab["policy"]
id_to_policy_move = {v: k for k, v in policy_vocab.items()}

history_vocab.get("e2e4", 0)          # tokenize for history input
id_to_policy_move.get(277)            # decode a policy-head prediction

---

How to Use

Option 1: High-Level API (otter-chess package — Recommended)

The easiest and most reliable way to use Otter is via the official otter-chess Python package. OtterModel automatically handles board state tensor encoding, move canonicalization for White/Black perspectives, history window tracking, legal move masking, and output decoding.

pip install otter-chess
# or install from GitHub source:
# pip install git+https://github.com/PeargentLabs/otter-chess.git

from otter import OtterModel

# Initialize model (automatically downloads and caches weights from Hugging Face if not cached)
model = OtterModel(device="cpu")

# --- Example position: after 1. e4 e5, White to move, 1500 vs 1600, 10+0 time control ---
res = model.predict(
    fen="rnbqkbnr/pppp1ppp/8/4p3/4P3/8/PPPP1PPP/RNBQKBNR w KQkq - 0 2",
    player_elo=1500,
    opponent_elo=1600,
    history_moves=["e2e4", "e7e5"],
    time_control="600+0",
    clock_fraction=0.5,
    top_k=5,
)

print(f"Predicted Win Probability: {res['win_probability']:.3f}")

print("\nTop Move Predictions:")
for m in res["moves"]:
    print(f"  {m['move']:<6} | Probability: {m['probability']:.1%}")

print("\nAuxiliary Head Predictions:")
aux = res["aux_predictions"]
print(f"  Moving Piece:   {aux['moving_piece']} ({aux['moving_piece_confidence']:.1%})")
print(f"  From Square:    {aux['from_square']} ({aux['from_square_confidence']:.1%})")
print(f"  To Square:      {aux['to_square']} ({aux['to_square_confidence']:.1%})")
print(f"  Check Prob:     {aux['results_in_check_probability']:.1%}")

---

Option 2: Low-Level PyTorch & fastchess Usage

If you prefer to load weights directly without installing the otter-chess package, you can interface with the PyTorch model tensors using fastchess and vocab.json.

pip install transformers torch fastchess huggingface_hub

import json
import torch
import fastchess
from transformers import AutoModel
from huggingface_hub import hf_hub_download

REPO_ID = "peargentlabs/otter-chess"
HISTORY_K = 20

# Load model and vocabulary
model = AutoModel.from_pretrained(REPO_ID, trust_remote_code=True)
model.eval()

with open(hf_hub_download(REPO_ID, "vocab.json")) as f:
    vocab = json.load(f)
history_vocab = vocab["history"]
policy_vocab = vocab["policy"]
id_to_move = {v: k for k, v in policy_vocab.items()}

# Perspective helpers
def mirror_square(sq: str) -> str: return f"{sq[0]}{9 - int(sq[1])}"
def mirror_move(mv: str) -> str: return f"{mirror_square(mv[:2])}{mirror_square(mv[2:4])}{mv[4:]}"
def canonicalize_move(mv: str, turn: int) -> str: return mirror_move(mv) if turn == fastchess.BLACK else mv

def elo_to_bucket(elo: int) -> int:
    if elo < 1100: return 0
    if elo >= 2000: return 10
    return 1 + (elo - 1100) // 100

def time_control_to_bucket(tc: str) -> int:
    if not tc or "+" not in tc: return 4
    try:
        base, inc = map(int, tc.split("+"))
        eff = base + 40 * inc
        if eff < 60: return 0
        if eff < 180: return 1
        if eff < 600: return 2
        if eff < 1800: return 3
        return 4
    except Exception: return 4

# Position setup
fen = "rnbqkbnr/pppp1ppp/8/4p3/4P3/8/PPPP1PPP/RNBQKBNR w KQkq - 0 2"
history_moves = ["e2e4", "e7e5"]
player_elo, opponent_elo_val = 1500, 1600
time_control = "600+0"
clock_fraction = 0.5

board_obj = fastchess.Board(fen)
turn = board_obj.turn

# 1. Canonicalize move history
canonical_history = []
temp_board = fastchess.Board()
for m in history_moves:
    orig_turn = temp_board.turn
    try:
        temp_board.push_uci(m)
        canonical_history.append(canonicalize_move(m, orig_turn))
    except Exception:
        canonical_history.append(canonicalize_move(m, fastchess.WHITE))

# 2. Prepare board & history tensors
board = torch.from_numpy(board_obj.to_tensor(canonical=True)).unsqueeze(0)
history_ids = torch.zeros(1, HISTORY_K, dtype=torch.long)
history_mask = torch.zeros(1, HISTORY_K, dtype=torch.bool)
window = canonical_history[-HISTORY_K:]
start = HISTORY_K - len(window)
for i, move in enumerate(window, start=start):
    history_ids[0, i] = history_vocab.get(move, 0)
    history_mask[0, i] = True

active_elo = torch.tensor([elo_to_bucket(player_elo)])
opponent_elo = torch.tensor([elo_to_bucket(opponent_elo_val)])
tc = torch.tensor([time_control_to_bucket(time_control)])
clock = torch.tensor([[clock_fraction, 0.0]])

# 3. Model forward pass (returns tuple: policy_logits, aux_logits, value_pred)
with torch.no_grad():
    policy_logits, aux_logits, value_pred = model(
        board, history_ids, history_mask, active_elo, opponent_elo, tc, clock
    )

# 4. Legal move masking & output un-canonicalization
legal_mask = torch.zeros(4208, dtype=torch.bool)
for move_uci in board_obj.legal_moves_uci():
    canon = canonicalize_move(move_uci, turn)
    if canon in policy_vocab:
        legal_mask[policy_vocab[canon]] = True

probs = torch.softmax(policy_logits[0].masked_fill(~legal_mask, -1e9), dim=-1)
top_probs, top_ids = torch.topk(probs, k=5)

print(f"Win Probability: {value_pred.item():.3f}")
for p, i in zip(top_probs.tolist(), top_ids.tolist()):
    move_canon = id_to_move[i]
    real_move = mirror_move(move_canon) if turn == fastchess.BLACK else move_canon
    print(f"  {real_move}: {p:.3f}")

---

Model Inputs Reference

If interfacing with raw model tensors directly, inputs are structured as follows:

---

Citation

@software{otter_chess_2026,
  author = {Tarun and Peargent Labs},
  title = {Otter: A Time-Aware, History-Conditioned Human Chess AI},
  year = {2026},
  url = {https://github.com/peargentlabs/otter-chess},
  version = {0.1.0}
}