QuadConnect2.5-1.5B is a specialized language model trained to master the game of Connect Four. Built on Qwen 2.5 (1.5B parameter base), this model uses GRPO (Group Relative Policy Optimization) to learn the strategic intricacies of Connect Four gameplay.

Status: Early training experiments (v0.0.9b) - Reward functions still evolving

🔍 Model Details

Developed by: Lyte
Model type: Small Language Model (SLM)
Language: English
Base model: unsloth/qwen2.5-1.5b-instruct-unsloth-bnb-4bit
Training method: TRL's GRPO
Training data: Lyte/ConnectFour-T10

🚀 Quick Start

Option 1: Using Transformers

python

from transformers import pipeline

SYSTEM_PROMPT = """You are a master Connect Four strategist whose goal is to win while preventing your opponent from winning. The game is played on a 6x7 grid (columns a–g, rows 1–6 with 1 at the bottom) where pieces drop to the lowest available spot.

Board:
- Represented as a list of occupied cells in the format: <column><row>(<piece>), e.g., 'a1(O)'.
- For example: 'a1(O), a2(X), b1(O)' indicates that cell a1 has an O, a2 has an X, and b1 has an O.
- An empty board is shown as 'Empty Board'.
- Win by connecting 4 pieces in any direction (horizontal, vertical, or diagonal).

Strategy:
1. Identify taken positions, and empty positions.
2. Find and execute winning moves.
3. If There isn't a winning move, then block your opponent's potential wins.
4. Control the center and set up future moves.

Respond in XML:
<reasoning>
Explain your thought process, focusing on your winning move, how you block your opponent, and your strategic plans.
</reasoning>
<move>
Specify the column letter (a–g) for your next move.
</move>
"""

board = {
    "empty": "Game State:\n- You are playing as: X\n- Your previous moves: \n- Opponent's moves: \n- Current board state: Empty Board\n- Next available position per column:  \nColumn a: a1, a2, a3, a4, a5, a6  \nColumn b: b1, b2, b3, b4, b5, b6  \nColumn c: c1, c2, c3, c4, c5, c6  \nColumn d: d1, d2, d3, d4, d5, d6  \nColumn e: e1, e2, e3, e4, e5, e6  \nColumn f: f1, f2, f3, f4, f5, f6  \nColumn g: g1, g2, g3, g4, g5, g6\n\nMake your move.",
    "one_move": "Game State:\n- You are playing as: X\n- Your previous moves: \n- Opponent's moves: b1\n- Current board state: b1(O)\n- Next available position per column:  \nColumn a: a1, a2, a3, a4, a5, a6  \nColumn b: b2, b3, b4, b5, b6  \nColumn c: c1, c2, c3, c4, c5, c6  \nColumn d: d1, d2, d3, d4, d5, d6  \nColumn e: e1, e2, e3, e4, e5, e6  \nColumn f: f1, f2, f3, f4, f5, f6  \nColumn g: g1, g2, g3, g4, g5, g6\n\nMake your move.",
    "four_moves": "Game State:\n- You are playing as: X\n- Your previous moves: a1, a2\n- Opponent's moves: d1, a3\n- Current board state: a1(X), d1(O), a2(X), a3(O)\n- Next available position per column:  \nColumn a: a4, a5, a6  \nColumn b: b1, b2, b3, b4, b5, b6  \nColumn c: c1, c2, c3, c4, c5, c6  \nColumn d: d2, d3, d4, d5, d6  \nColumn e: e1, e2, e3, e4, e5, e6  \nColumn f: f1, f2, f3, f4, f5, f6  \nColumn g: g1, g2, g3, g4, g5, g6\n\nMake your move.",
}

generator = pipeline("text-generation", model="Lyte/QuadConnect2.5-1.5B-v0.1.0b", device="cuda")

# use 'empty', 'one_move' or 'four_moves' in board['']
output = generator([
    {"role": "system", "content": SYSTEM_PROMPT}, 
    {"role": "user", "content": board['empty']}
], max_new_tokens=10245, return_full_text=False)[0]

print(output["generated_text"])

Option 2: Using GGUF

Download the Quantized GGUF (Q8_0) and use it in your favorite GGUF inference engine (e.g., LMStudio).

Option 3: Using Hugging Face Space

Visit the QuadConnect Space to interact with the model directly. You can also duplicate the space or download its code for local use.

📊 Evaluation Results

Model performance was evaluated on the Lyte/ConnectFour-T10 validation split with various temperature settings.
Note: It appears 0.8 is the ideal temperature, and that reward functions for game strategy are harder to balance.

Summary Metrics Comparison

Metric	v0.0.6b (Temp 0.6)	v0.0.8b (Temp 0.6)	v0.0.9b (Temp 0.6)	v0.0.9b (Temp 0.8)	v0.0.9b (Temp 1.0)	v0.1.0b (Temp 0.8)
Total games evaluated	5082	5082	5082	5082	5082	5082
Correct predictions	518	394	516	713	677	809
Accuracy	10.19%	7.75%	10.15%	14.03%	13.32%	15.92%
Most common move	d (41.14%)	d (67.61%)	a (38.72%)	a (31.01%)	a (26.99%)	e (20.89%)
Middle column usage	75.05%	99.53%	29.08%	35.43%	39.49%	51.38%

Move Distribution by Column

Column	v0.0.6b (Temp 0.6)	v0.0.8b (Temp 0.6)	v0.0.9b (Temp 0.6)	v0.0.9b (Temp 0.8)	v0.0.9b (Temp 1.0)	v0.1.0b (Temp ?)
a	603 (19.02%)	3 (0.12%)	1447 (38.72%)	1547 (31.01%)	1351 (26.99%)	803 (16.71%)
b	111 (3.50%)	4 (0.16%)	644 (17.23%)	924 (18.52%)	997 (19.92%)	223 (4.64%)
c	785 (24.76%)	463 (17.96%)	648 (17.34%)	1003 (20.11%)	985 (19.68%)	844 (17.57%)
d	1304 (41.14%)	1743 (67.61%)	101 (2.70%)	202 (4.05%)	306 (6.11%)	621 (12.92%)
e	290 (9.15%)	360 (13.96%)	338 (9.04%)	562 (11.27%)	686 (13.70%)	1004 (20.89%)
f	50 (1.58%)	3 (0.12%)	310 (8.30%)	408 (8.18%)	354 (7.07%)	459 (9.55%)
g	27 (0.85%)	2 (0.08%)	249 (6.66%)	342 (6.86%)	327 (6.53%)	851 (17.71%)

🔧 Training Details

Data Preparation

Started with Leon-LLM/Connect-Four-Datasets-Collection
Filtered for clean, complete entries
Further filtered to include only games with 10 or fewer turns
Split into train and validation sets
Final dataset: Lyte/ConnectFour-T10

Evaluation Parameters

Temperature: 0.6, 0.8, 1.0 (compared)
Top-p: 0.95
Max tokens: 1024

Framework Versions

TRL: 0.15.1
Transformers: 4.49.0
PyTorch: 2.5.1+cu121
Datasets: 3.2.0
Tokenizers: 0.21.0

📚 Citations

For GRPO:

bibtex

@article{zhihong2024deepseekmath,
    title        = {{DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models}},
    author       = {Zhihong Shao and Peiyi Wang and Qihao Zhu and Runxin Xu and Junxiao Song and Mingchuan Zhang and Y. K. Li and Y. Wu and Daya Guo},
    year         = 2024,
    eprint       = {arXiv:2402.03300},
}

For TRL:

bibtex

@misc{vonwerra2022trl,
    title        = {{TRL: Transformer Reinforcement Learning}},
    author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
    year         = 2020,
    journal      = {GitHub repository},
    publisher    = {GitHub},
    howpublished = {\url{https://github.com/huggingface/trl}}
}