react-agent
stocks_demo_react_agent_generated_train_datasetqwen3-embedding-4b_react-agent_ablation_filteredagentic-think-v2-react-hybrid
agentic-think-v2-react-hybrid
Agentic tool-calling SFT corpus, 212,396 conversations, where 36% of the episodes carry reasoning
generated in a gold-blind ReAct loop instead of after-the-fact annotation. Built by ENERZAi for
small-model (1.7B ternary) agentic SFT.
This is a drop-in successor to HBKenerzai/agentic-think-v1:
same episodes, same scale, same schema — only the <think> text differs on the replaced rows.
Why it exists
In v1 every think block was written… See the full description on the dataset page: https://huggingface.co/datasets/ENERZAiKR/agentic-think-v2-react-hybrid.qwen3-embedding-0.6b_react-agent_filteredqwen3-embedding-4b_react-agent_filteredqwen3-embedding-0.6b_react-agent_ablation_filtered