EvoTrain-180K

EvoTrain-180K is a diverse dataset designed for the joint optimization of latent memory and long-context retrieval, presented in the paper EvoEmbedding: Evolvable Representations for Long-Context Retrieval and Agentic Memory.

🔗 GitHub Repository | 🏠 Project Page | 📚 Paper

Dataset Structure

The dataset uses an intuitive, chat-style fine-tuning format designed for joint SFT (Supervised Fine-Tuning) and retrieval optimization. Each training instance is a self-contained context window structured with the following key components:

• messages: Standard alternating user/assistant turns representing the SFT generation target.
• meta.turns: Formatted turn-level strings (concatenating User and Assistant messages) processed sequentially by the retriever path to simulate dynamic context streaming.
• meta.evidence_turns: Zero-based indices pointing to the specific historical turn(s) containing the ground-truth evidence (serving as positive targets $v^+$ for contrastive loss).

Structural Example

{
    "messages": [
        {
            "role": "user",
            "content": "[Target Evidence] I’m a retired cop... Any good local diners? No seafood, please."
        },
        {
            "role": "assistant",
            "content": "Try Mama’s Diner on Main St. No seafood on the menu."
        },
        
        ... (Multiple irrelevant dialogue turns / long context chunks omitted) ...
        
        {
            "role": "user",
            "content": "[Query] What type of restaurant did I say not to recommend earlier?"
        },
        {
            "role": "assistant",
            "content": "[Ground-truth Answer] Seafood."
        }
    ],
    "meta": {
        "evidence_turns": [
            0
        ],
        "turns": [
            "User: [Target Evidence] I’m a retired cop... Any good local diners? No seafood, please.
Assistant: Try Mama’s Diner on Main St. No seafood on the menu.",
            "... (Multiple irrelevant dialogue turns / long context chunks omitted) ...",
            "[Query] What type of restaurant did I say not to recommend earlier?"
        ]
    }
}

Citation

If you find this dataset or our methodology useful, please cite our paper:

@article{nie2026evoembedding,
  title={EvoEmbedding: Evolvable Representations for Long-Context Retrieval and Agentic Memory},
  author={Nie, Chang and Fu, Chaoyou and Feng, Junlan and Shan, Caifeng},
  journal={arXiv preprint arXiv:2606.21649},
  year={2026}
}