Code Instruct Final: Curated Code Instruction Datasets for LLM Tuning

A curated collection of five established code instruction datasets formatted for LLM training. The datasets, including Magicoder-OSS-Instruct-75K and glaive-code-assistant-v3, have been processed into the LLAMA chat format with markdown for code snippets. It was created by MaLA-LM and last updated in July 2024.

Use Cases

• Fine-tuning code generation models based on curated instruction-response pairs.
• Training models to follow complex coding instructions based on the formatted datasets.
• Benchmarking model performance on code-related tasks using established instruction-tuning sources.
• Developing multilingual code assistants based on the subset of programming languages selected.

Strengths

• Integrates five established instruction-tuning datasets, including Magicoder-Evol-Instruct-110K.
• Data is formatted consistently using the LLAMA chat format and markdown for code snippets.
• Subset is filtered for specific programming languages using guesslang, targeting continued pretraining.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and total size are unknown, which may limit suitability assessment.
• Last updated 2024-07-09 03:24:42; freshness should be verified.

Provenance

Source

Aggregated from five Hugging Face datasets including ise-uiuc/Magicoder-OSS-Instruct-75K and glaiveai/glaive-code-assistant-v3.

Collection Method

Curated, reformatted, and filtered from source datasets.

Freshness

Last updated 2024-07-09 03:24:42.