bolajiev/maxx1.5Bv2
The bolajiev/maxx1.5Bv2 is a 1.5 billion parameter Qwen2.5-based causal language model developed by bolajiev, fine-tuned for agentic tool-use and step-by-step reasoning. It features a 32768-token context length and is optimized for function calling and complex reasoning tasks. This model is specifically designed to excel in scenarios requiring structured interaction and logical progression, making it suitable for automated agent workflows.
Loading preview...
Maxx 1.5Bv2: Agentic Tool-Use and Reasoning Model
bolajiev/maxx1.5Bv2 is a 1.5 billion parameter model built upon the Qwen2.5-1.5B-Instruct base, specifically fine-tuned for enhanced agentic tool-use and step-by-step reasoning capabilities. This model leverages a 32768-token context window to handle complex interactions.
Key Capabilities
- Optimized for Agentic Workflows: Designed to perform effectively in scenarios requiring automated decision-making and tool integration.
- Enhanced Reasoning: Fine-tuned to facilitate step-by-step logical progression in responses.
- Function Calling: Specialized training for robust function calling, making it suitable for interacting with external APIs and tools.
- Efficient Training: Utilizes QLoRA SFT followed by DPO (Unsloth) for efficient fine-tuning.
Training Data
The model was trained on a diverse dataset including OpenHermes 2.5, SlimOrca, Glaive FC v2, Hermes FC, and approximately 131K synthetic ReAct tool-use trajectories for Supervised Fine-Tuning (SFT). DPO training involved around 5K synthetic preference pairs, distinguishing between good and bad tool calls.
Target Benchmarks
Maxx 1.5Bv2 aims for strong performance on the HuggingFace Open LLM Leaderboard and the Berkeley Function Calling Leaderboard (BFCL), indicating its focus on general language understanding and specialized function calling prowess.