HF RL Explorer

Generation and speculative-decoding paths increasingly transform key/value caches instead of treating them…

Generation and speculative-decoding paths increasingly transform key/value caches instead of treating them…: a task in MiMo-V2.6-RL-oss: Agentic RL Environments (MiMo RL release). The public DynamicCache and EncoderDecoderCache APIs need consistent bookkeeping when those transformations are…

The task

The public DynamicCache and EncoderDecoderCache APIs need consistent bookkeeping when those transformations are composed, including models whose cached activations legitimately contain zeros. Please harden these cache contracts without breaking the existing legacy/indexing and…

Part of XiaomiMiMo/MiMo-V2.6-RL-oss.