lyrain2001/Auto-Fill-Qwen3-8B-Knowledge
The lyrain2001/Auto-Fill-Qwen3-8B-Knowledge model is an 8 billion parameter Qwen3-based specialist language model, part of the Auto-Fill ensemble, designed to predict missing values in tables. It focuses on filling cells requiring world knowledge or pattern recognition within the table, such as entity attributes, codes, names, dates. This model is specifically fine-tuned for direct, non-chain-of-thought responses to table-filling tasks, providing a single predicted value.
Loading preview...
Auto-Fill Qwen3-8B Knowledge Specialist
This model is the knowledge specialist component of the Auto-Fill system, designed to accurately predict missing values in tabular data. It is built upon the Qwen3-8B base model and fine-tuned for direct, non-chain-of-thought responses. The model excels at inferring missing cell values that depend on world knowledge or discernible patterns within the table itself, such as entity attributes, codes, names, or dates.
Key Capabilities
- Missing Value Prediction: Specializes in filling
[MISSING]cells in tables. - Direct Output: Provides the predicted value directly in JSON format, without additional reasoning text.
- Table Serialization: Processes tables serialized as Markdown pipe tables.
- Ensemble Integration: Designed to work in parallel with other Auto-Fill specialists (Reasoning and Coding) for comprehensive missing value imputation, with confidence calibration.
- Training Data: Trained on 30,000 examples of
(table, value)pairs from public sources, with one masked cell per table.
When to Use This Model
This specialist is ideal for scenarios where missing table cells can be inferred from:
- World Knowledge: Common facts, names, dates, or codes.
- Table Patterns: Attributes that follow clear patterns within the given table context.
It is recommended to use this model as part of the full Auto-Fill ensemble for optimal performance and calibrated confidence, especially when dealing with diverse missing value types. The model was trained and evaluated on English-language tables, with a maximum input length of 40,960 tokens.