Model Overview

The astom-M/matsuo-llm-advanced-phase-e2a is a 7.6 billion parameter language model, fine-tuned from the Qwen/Qwen2.5-7B-Instruct base. Its core specialization lies in agent tasks, specifically demonstrating enhanced performance on benchmarks like DBBench and ALFWorld.

Key Capabilities

• Agentic Task Proficiency: Optimized for complex, multi-step agent workflows, including database querying and interactive environment control.
• Database Interaction: Fine-tuned using datasets such as u-10bei/dbbench_sft_dataset_react_v4, xlangai/spider, and birdsql/bird_mini_dev, making it adept at understanding and generating SQL-like queries.
• Interactive Environment Navigation: Utilizes the official ALFWorld v5 dataset to improve its ability to reason and act within simulated environments.
• Compliance: Ensures no evaluation data was used in training, no LLM was used for data quality filtering, and inference code remains unmodified.

Good For

• Developing AI Agents: Ideal for applications requiring autonomous agents that can interact with databases or navigate digital environments.
• Database Query Generation: Suitable for tasks involving natural language to SQL conversion or database schema understanding.
• Complex Reasoning Tasks: Excels in scenarios demanding sequential decision-making and problem-solving within defined contexts.

This model is compatible with vLLM v0.13.0+ for efficient inference.