astom-M/matsuo-llm-advanced-phase-c
The astom-M/matsuo-llm-advanced-phase-c is a 7.6 billion parameter instruction-tuned language model, fine-tuned from Qwen2.5-7B-Instruct. This model specializes in agent task performance, particularly excelling in ALFWorld and DBBench scenarios. It leverages a 32768-token context length and is optimized for complex reasoning and database interaction tasks.
Loading preview...
Overview
The astom-M/matsuo-llm-advanced-phase-c is a 7.6 billion parameter language model, building upon the Qwen2.5-7B-Instruct base. It has been specifically fine-tuned to enhance performance in agent-based tasks, with a particular focus on environments like ALFWorld and DBBench. The model utilizes a substantial context window of 32768 tokens, allowing for processing longer and more complex inputs.
Key Capabilities
- Enhanced Agent Task Performance: Demonstrates improved capabilities in executing tasks within agent environments such as ALFWorld and DBBench.
- Database Interaction: Optimized for tasks involving database querying and manipulation, leveraging datasets like
dbbench_sft_dataset_react_v4,xlangai/spider, andbirdsql/bird_mini_dev. - Large Context Window: Supports a 32768-token context length, beneficial for intricate multi-turn interactions and detailed problem-solving.
Training Details
The model was fine-tuned using a curated selection of public datasets, including u-10bei/dbbench_sft_dataset_react_v4 for DBBench format alignment, xlangai/spider (CC BY-SA 4.0), and birdsql/bird_mini_dev (CC BY-SA 4.0). Importantly, no evaluation data was used during training, and no LLM was employed for data quality filtering or selection, ensuring data integrity and preventing leakage.
Good For
- Developing and deploying AI agents that require robust reasoning and interaction capabilities.
- Applications involving complex database querying and schema understanding.
- Scenarios demanding a large context window for processing extensive instructions or conversational histories.