Convex Markets / Datasets / SFT dataset / WizardLM Evol-Instruct V2 196k
SKU SFT-2102 · Sold by External

WizardLM Evol-Instruct V2 196k

Product specifications

SKUSFT-2102
Data typeSFT dataset
Volume143000 examples
Size on disk162 MB (Parquet auto-conversion)
FormatJSON (Parquet auto-conversion available)
Access modelPUBLIC LICENSE
PricingFree · open-source license
Quality score
LicenseMIT
WizardLM_evol_instruct_V2_196k is the Evol-Instruct V2 training mixture used to fine-tune WizardLM models, produced by iteratively evolving instructions to increase complexity and breadth across reasoning, code, math, and general conversation. The current Hugging Face train split contains 143,000 multi-turn conversation rows; the card notes the full ~196K figure requires merging with the original ShareGPT source. Each row stores a ShareGPT-style 'conversations' list of {from, value} turns.