86,864 examples adapted from the hypervariance/function-calling-sharegpt dataset to fine-tune the Google gemma-2-2b-it model. DinushiTJ modified 8.49% of the examples by merging consecutive GPT responses and standardizing role names. The dataset was last updated on Hugging Face in September 2024.
Use Cases
- Fine-tuning language models for structured function calling based on the adapted conversational format.
- Training models to generate API calls or tool usage based on user instructions mentioned in the description.
- Benchmarking model performance on function-calling tasks using the standardized 'user' and 'assistant' roles.
- Studying the effect of merging consecutive model responses on instruction-following performance.
Strengths
- Based on a substantial source dataset of 86,864 examples.
- Includes specific preprocessing steps affecting 7,372 examples (8.49%) for consistency.
- Explicitly formatted for fine-tuning a specific model (gemma-2-2b-it).
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- Adapted from hypervariance/function-calling-sharegpt dataset.
- Collection Method
- Merged consecutive responses and updated role names from the original ShareGPT-style data.
- Time Range
- null
- Freshness
- Last updated 2024-09-27 21:15:31; freshness should be verified.
- Geography
- null