Modelpublic

the-acorn-ai/spiral-octothinker-3b-multi-step00192

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face

0likes14downloads

Model Card

SPIRAL OctoThinker-3B Multi-Agent Model

This model was trained using the SPIRAL (Self-Play Iterative Reinforcement learning for Adaptation and Learning) framework.

Model Details

—Base Model: OctoAI/OctoThinker-3B
—Training Framework: SPIRAL
—Checkpoint: step_00192
—Model Size: 3B parameters
—Training Date: 2025-08-26

Training Configuration

The model was trained with self-play on multiple environments:

—KuhnPoker-v1
—TicTacToe-v0
—SimpleNegotiation-v1

Training Parameters

json

{
  "learning_rate": "1e-6",
  "train_batch_size": 128,
  "num_ppo_epochs": 2,
  "temperature": 1.0,
  "max_model_len": 16384,
  "environments": [
    "KuhnPoker-v1",
    "TicTacToe-v0",
    "SimpleNegotiation-v1"
  ],
  "base_model": "OctoAI/OctoThinker-3B",
  "framework": "SPIRAL"
}

Usage

python

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

tokenizer = AutoTokenizer.from_pretrained("the-acorn-ai/spiral-octothinker-3b-multi-step00192")
model = AutoModelForCausalLM.from_pretrained(
    "the-acorn-ai/spiral-octothinker-3b-multi-step00192",
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

# Generate text
inputs = tokenizer("Your prompt here", return_tensors="pt")
outputs = model.generate(**inputs, max_length=100)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

License

This model is licensed under the Apache License 2.0.