# Hybrid GatedDeltaNet state rollback and checkpoint prefix caching > llama.cpp's hybrid memory is two sub-caches: an attention KV cache and a recurrent cache with one cell per sequence. Parent: [Mac local LLMs: Speculative decoding and MTP](https://llms-explorer.com/tree/mac-local-llms-speculative-decoding-and-mtp/) · 1 facets · 24 facts · page: https://llms-explorer.com/tree/hybrid-gateddeltanet-state-rollback-and-checkpoi/ ## Facts - llama.cpp's hybrid memory is two sub-caches: an attention KV cache and a recurrent cache with one cell per sequence. — source: `asserted` - A checkpoint stores only the part that cannot be rebuilt by truncation (`LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY`); the attention KV prefix stays in the slot. — source: `asserted` - The recurrent cache also keeps a short ring of per-token snapshots (`n_rs_seq`), used when speculative decoding rejects part of a draft. — source: `asserted` - Restore picks the latest checkpoint at or before the shared prefix and reprocesses from there. — source: `asserted` - 2026-04-26: issue 22384 shows the restore test never succeeds for recurrent state. — source: `asserted` - 2026-06: PRs 24176, 24411 and 25472 repair placement, erasure and eviction; PR 24797 (draft) targets the `seq_pos_min()` root cause. — source: `asserted` - ik_llama.cpp keeps the same logic and the same failure (issue 1762, May 2026). — source: `asserted` - Speculative decoding on hybrid models uses checkpoints: the server logs `speculative decoding will use checkpoints` at load. Partial acceptance beyond `n_rs_seq` used to fail `seq_rm` and leave the attention cache with stale entries. — source: `asserted` - The recurrent snapshot is not precise after a fall-through rollback: the model reprocesses from the last valid checkpoint. — source: `asserted` - The hybrid `seq_pos_min()` was written as `max()` of both sub-caches; the PR 24797 author argues it should return only the attention cache's minimum, as non-hybrid memory does. Whether other hybrid types (Mamba-based) should follow is untested. — source: `asserted` - Merge state and effect of PR 24797 on the residual misses reported after 24411 and 24176. — source: `asserted` - Whether Metal's behavior differs; every cited log is CUDA or ROCm. — source: `asserted` - PR 24797 states hybrid memory consists of two independent cache subsystems with different positional semantics: for the attention cache `pos_min` is the earliest cached KV position, and for the recurrent cache `pos_min == pos_max ==` the current position. — [source](https://github.com/ggml-org/llama.cpp/pull/24797) - PR 24797 says the old `seq_pos_min()` returned the `std::max()` of both caches' minimums, which made a cached prefix look absent and forced full reprocessing on any prompt variation (stripped thinking, injected timestamp). — [source](https://github.com/ggml-org/llama.cpp/pull/24797) - PR 24797 keeps `seq_pos_max` with intersection semantics because it bounds valid attention entries by where the recurrent state has advanced. — [source](https://github.com/ggml-org/llama.cpp/pull/24797) - PR 24797 notes other hybrid types with recurrent memory would probably benefit from the same `seq_pos_min()` change but were not tested. — [source](https://github.com/ggml-org/llama.cpp/pull/24797) - On hybrid models, when speculative decoding partially accepts draft tokens and the rollback distance exceeds `n_rs_seq`, the pre-24797 recurrent cache returned false from `seq_rm`, leaving the attention cache with stale entries. — [source](https://github.com/ggml-org/llama.cpp/pull/24797) - Particula states the bounded recurrent rewind succeeds only when `cell.pos - (p0 - 1)` is between 1 and `n_rs_seq` (context type `SEQ_RM_TYPE_RS`). — [source](https://particula.tech/blog/prompt-reprocessing-swa-hybrid-models-kv-cache) - Hybrid checkpoints in 24055 logs satisfy `pos_min == pos_max == n_tokens - 1` (for example 2799, 2799, 2800 tokens), so each stores one recurrent position. — [source](https://github.com/ggml-org/llama.cpp/issues/24055) - A 27B hybrid checkpoint measured 149.626 MiB at 2,800 tokens and 151.055 MiB at 364 tokens, so size is nearly independent of prompt length. — [source](https://github.com/ggml-org/llama.cpp/issues/24055) - The load log of a Qwen3.6-27B MTP model reads `speculative decoding will use checkpoints` followed by `common_speculative_init: no implementations specified for speculative decoding` when no draft type is given. — [source](https://github.com/ggml-org/llama.cpp/issues/24055) - ik_llama.cpp issue 1762 (qwen3next, Gated DeltaNet) reports the same restore-test failure and 64-token creation floor as llama.cpp 22384 and says the same fix is needed in the fork's server checkpoint code. — [source](https://github.com/ikawrakow/ik_llama.cpp/issues/1762) - Reporters in 24055 use `--spec-type draft-mtp` with `--spec-draft-n-max 3` (and 6) on hybrid Qwen3.6 models while relying on checkpoints, with 85% draft acceptance reported on one 27B run. — [source](https://github.com/ggml-org/llama.cpp/issues/24055) - Inferred: restore cost on a hybrid model is the token gap between the resume point and the nearest earlier checkpoint, so placing a checkpoint at each user-message boundary (PR 24176) bounds typical re-prefill to one turn's new text plus the last assistant turn. — source: `asserted`