SKU SFT-2105 · Sold by External
Orca AgentInstruct 1M v1
Product specifications
| SKU | SFT-2105 |
|---|---|
| Data type | SFT dataset |
| Volume | 1046410 examples |
| Size on disk | 2.21 GB (Parquet) |
| Format | Parquet (Hugging Face) |
| Access model | PUBLIC LICENSE |
| Pricing | Free · open-source license |
| Quality score | — |
| License | CDLA-Permissive-2.0 |
orca-agentinstruct-1M-v1 is a fully synthetic instruction-tuning dataset of 1,046,410 examples generated by Microsoft's AgentInstruct multi-agent 'agentic flow' pipeline (the 'agent' refers to the generation method, not tool-use content). It spans 15 task splits including open_domain_qa (272K), mcq (100K), code (100K), follow_up (99.1K), plus reading comprehension, creative writing, text modification, struct2text, analytical reasoning, and more. Each row stores a full multi-role conversation (system/user/assistant) in a serialized 'messages' field.