DeepSpec · Issues· 65 open

[架构提案] 针对递归自我改进（RSI）的硬件门控遏制框架（Génesis 协议 V5.0）

Can I train dspark for deepseek-v4-flash-0731?

can I train dspark for qwen3.8-flash-next?

Does this framework currently support only the three Qwen3 models listed in the documentation? Does it support training for the Qwen3.5 and Qwen3.6 model series? If not, are there any plans to add support for them in the future?

Eagle3 peaks at ~4.6 epochs then degrades; also 6% short of the released checkpoint (DSpark/DFlash reproduce)

DSpark在多模态场景适用吗，例如qwen3-vl模型

Target feature selection for DSpark

Support online training without precomputing the target cache

[Off-topic] Question for Damai Dai / Chengqi Deng: has profile-based MoE routing been explored?

[Question] Why DFlash is worse than EAGLE3 for Gemma4-12B in paper's experiment

DSpark Qwen3 draft modeling silently ignores `partial_rotary_factor`, producing checkpoints whose config mismatches the trained weights (breaks vLLM serving AL)

feat: support Multi-Token Prediction (MTP) training for LLM and speech modalities

一个关于生成范式的远期思考（非Bug报告）

Release speculators used for ablations

请问deepseek官方是否有针对新版本模型写作文笔能力修正？