Sign in to view source links and access this dataset
Description
Capybara is a dataset of over 10,000 multi-turn conversational examples. It was created by LDJnr and released on Hugging Face, with a last recorded update on 2024-06-07. The dataset is described as the culmination of synthesis techniques like Evol-instruct, Alpaca, Orca, Vicuna, Lamini, and FLASK.
Use Cases
Fine-tuning language models for multi-turn dialogue based on the described synthesis techniques.
Benchmarking conversational AI systems on logical robustness and diversity.
Studying the effects of instruction synthesis methods like Evol-instruct and FLASK.
Training models to follow complex, multi-step instructions derived from vetted seed datasets.
Strengths
Contains over 10,000 multi-turn conversational examples.
Synthesized using multiple advanced techniques including Evol-instruct, Alpaca, Orca, Vicuna, Lamini, and FLASK.
Seed datasets were personally vetted by the author and noted for diversity and logical robustness.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown beyond the 'over 10,000' figure, which may limit suitability assessment.
The description references a full description on an external page, requiring navigation for complete context.
Provenance
Source
LDJnr on Hugging Face.
Collection Method
Synthesized from vetted seed datasets using techniques like Evol-instruct, Alpaca, Orca, Vicuna, Lamini, and FLASK.
Freshness
Last updated 2024-06-07 20:15:36.
License information is unknown and should be verified before use.