AlioLeuchtmann/ALIO-SQL-7B

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Mar 30, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

AlioLeuchtmann/ALIO-SQL-7B is a 7.6 billion parameter instruction-tuned language model based on Qwen2.5-Coder-7B, specifically fine-tuned for text-to-SQL generation. It achieves 57% on the BIRD private test set and 60% on BIRD dev greedy, outperforming larger universal models like GPT-4o and Gemini 1.5 Pro in its specialized domain. This model excels at converting natural language questions into SQLite queries, making it ideal for database interaction applications.

Loading preview...

AlioLeuchtmann/ALIO-SQL-7B: Specialized Text-to-SQL Model

AlioLeuchtmann/ALIO-SQL-7B is a 7.6 billion parameter model, fine-tuned from Qwen2.5-Coder-7B, designed exclusively for generating SQL queries from natural language. This model demonstrates strong performance on text-to-SQL tasks, particularly within the BIRD and Spider datasets.

Key Capabilities & Performance

  • Specialized Text-to-SQL: Optimized for converting natural language questions into SQLite queries.
  • Knowledge Distillation: Trained using knowledge distillation from Qwen2.5-Coder-32B, with filtered and corrected outputs from CoT prompting.
  • Benchmark Performance: Achieves 57% on the BIRD private test set, 60% on BIRD dev greedy, and 70% on BIRD dev pass4.
  • Domain-Specific Excellence: Surpasses larger, more general-purpose models like GPT-4o and Gemini 1.5 Pro in its specialized text-to-SQL domain, highlighting the effectiveness of black-box knowledge distillation.
  • Teacher Model Outperformance: After fine-tuning, this checkpoint even exceeds the performance of its teacher model, Qwen2.5-Coder-32B, through quality filtering of teacher outputs.

Usage & Limitations

  • Prompt Format: Utilizes a specific prompt template for optimal text-to-SQL generation, incorporating schema, external knowledge, and the user's question.
  • SQLite Dialect: Only trained and optimized for generating SQLite dialect SQL queries.
  • English Only: Designed for English language inputs.
  • Focused Skillset: This model's training was exclusively focused on text-to-SQL generation, and it does not retain other general LLM capabilities.

Finetunes
1 models