Combine Qwen hybrid prompt-cache fixes on current upstream - #3
Closed
Regrad wants to merge 73 commits into
Closed
Conversation
* model: support for Mellum architecture * model: improve mellum.py formatting * model: improve mellum.py formatting once again * deps: downgrade transformers to 4.57.6 (to fix CI) * deps: remove huggingface_hub dependency * deps: remove huggingface_hub from test requirements --------- Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* hex-ops: fix profiler output (ie remove the redundant NONEs) * hex-prof: update profiling script to support tot.usec column
* tests : add support for qwen3 SSM archs * arch : add LLM_KV_ATTENTION_RECURRENT_LAYERS * cont : naming + TODOs
* cuda: reserve space for quantize kv-cache at startup * address review comments * remove forward decl Co-authored-by: Johannes Gäßler <johannesg@5d6.de> * remove assert in ggml-cuda.cu Co-authored-by: Johannes Gäßler <johannesg@5d6.de> --------- Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
…gml-org#24030) * Removes __restrict__ from PDL kernel headers due to incompatibility with PDL. Adds preprocessor directives based on arch in kernel body to add __restrict__ to retain performance on older architectures. * Simplifies new __restrict__ usage via macro * Add hopper to PDL __restrict__ fix. Co-authored-by: Oliver Simons <osimons@nvidia.com> --------- Co-authored-by: Oliver Simons <osimons@nvidia.com>
* add model * nits
* qwen35: use post-norm hidden state for MTP * rename pre_norm to nextn * fix step35
* Tidy up SYCL doc a bit - Add explicit links to referenced items - Fix spelling errors Signed-off-by: Todd Malsbary <todd.malsbary@intel.com> * Correct documented default for GGML_SYCL_GRAPH The default is ON, not OFF: $ cmake -LAH -B build | grep GGML_SYCL_GRAPH ... GGML_SYCL_GRAPH:BOOL=ON Signed-off-by: Todd Malsbary <todd.malsbary@intel.com> * Move docker instructions from SYCL.md to docker.md This makes them directly accesible from the Quick Start section of the top-level README.md. Signed-off-by: Todd Malsbary <todd.malsbary@intel.com> * Refer to intel.Dockerfile for ARGs and their defaults The defaults are always changing; this avoids accuracy errors from duplicating the information. Signed-off-by: Todd Malsbary <todd.malsbary@intel.com> * Remove mention of Nvidia in SYCL row of backend table This support was removed in 2026.02 - refer to the SYCL.md News. Signed-off-by: Todd Malsbary <todd.malsbary@intel.com> --------- Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
…2754) * ggml-cpu: add rvv 512b,1024b impls for iq4_xs * ggml-cpu: refactor; add rvv 512b, 1024b impls for q6_K, i-quants * ggml-cpu: refactor; add 512 and 1024 implementations of tq3_s, iq3_xxs, iq2_s, iq2_xs, iq2_xxs improve iq2_xs impl for rvv 256 Co-authored-by: Rehan Qasim <rehan.qasim@10xengineers.ai> --------- Co-authored-by: taimur-10x <taimur.ahmad@10xengineers.ai> Co-authored-by: Rehan Qasim <rehan.qasim@10xengineers.ai>
…rt (ggml-org#23834) * Start work on flash_attn refactor * Refactor * Split k/v quantization * Refactor and abstract quantization logic for flash_attn and mul_mat * Add quantization support to tile path * formatting * Move to functions, add a check
…#24073) * tests : refactor test-save-load-state to accept token input - Default prompt is now empty; when not provided, generate n_batch random tokens (useful for models without a tokenizer) - Tokenization happens once upfront; pass token vector to test functions - generate_tokens prints token IDs instead of decoded pieces - Use llama_model_get_vocab / llama_vocab_n_tokens API - Upgrade log level from LOG_TRC to LOG_INF for visibility Assisted-by: llama.cpp:local pi * cont : use llama_tokens alias
…ml-org#23974) The XCFramework generated by build-xcframework.sh creates a module map that manually lists public headers. That list can fall out of sync with the framework's Headers directory. The module map is currently missing ggml-opt.h, which is present in the framework headers. This can cause downstream Apple builds to fail with: Include of non-modular header inside framework module 'llama' Use the framework's Headers directory itself as the module map umbrella instead of maintaining a manual header list. This makes all public headers under the generated framework's Headers directory part of the llama module.
…ggml-org#24065) * webui: fix tool selector toggle/counter, key tools by stable identity Key the disabled set, counts and toggles by a stable per-tool key instead of bare function name, deduped from one canonical list. Per-tool checkboxes become presentational (single row handler, no nested button), category checkboxes drop the tristate (n/total carries partial). One getEnabledToolsForLLM keeps normalized MCP schemas and dedupes by name. * ui: use SvelteSet and SvelteMap for local tool collections to satisfy svelte/prefer-svelte-reactivity
* agents: refactor, include more guidelines * better example * rephrase a bit * add more examples * nits
…ent (ggml-org#24110) * server: avoid unnecessary checkpoint restore when new tokens are present The pos_min_thold calculation unconditionally subtracts 1 to ensure at least one token is evaluated for logits when no new tokens exist. However, when the request contains new tokens beyond the cached prefix, this -1 is overly conservative and may trigger an unnecessary checkpoint restore. Conditionally apply the -1 only when n_past >= task.n_tokens() (no new tokens), avoiding redundant KV state restoration when there is actual work to do. * cont : add ref --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
) * ggml: vectorize ggml_vec_dot_q4_1_q8_1 with WASM SIMD128 Optimize the inner loop of ggml_vec_dot_q4_1_q8_1_generic using WASM SIMD128 intrinsics, gated behind #ifdef __wasm_simd128__ so non-wasm builds are completely unaffected. Approach: - single wasm_v128_load covers all 32 packed 4-bit weights - nibbles unpacked via AND/SHR into two u8x16 registers - widened to i16 before multiply (WASM SIMD has no i8*i8 instruction) - 4x wasm_i32x4_dot_i16x8 calls accumulate all 32 element pairs - horizontal reduce via 4x wasm_i32x4_extract_lane Benchmark (node v25, emcc -O3 -msimd128, 64 blocks x QK8_1=32, 200k iterations): | impl | ns/call | speedup | |--------|---------|---------| | scalar | 880.7 | 1.00x | | simd | 257.8 | 3.42x | Correctness verified against scalar reference across 10 random seeds with exact output match. * ggml: move q4_1_q8_1 WASM SIMD implementation to wasm backend Relocate the SIMD128 implementation of ggml_vec_dot_q4_1_q8_1 to ggml/src/ggml-cpu/arch/wasm/quants.c to follow architecture-specific layout. Restore the generic implementation in ggml/src/ggml-cpu/quants.c. Move for loop in the else block. * ggml: use generic q4_1_q8_1 fallback in wasm backend
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* feat: add video support for Qwen3.5 * various clean up * revise the design * fix llava-uhd case * nits * nits 2 --------- Co-authored-by: andrewmd5 <1297077+andrewmd5@users.noreply.github.com>
…gml-org#24234) * common/chat : fix LFM2 reasoning round-trip and stray <think> leak * Gate by reasoning format and whether the template supports <think>
Prepare isolated work branch for combined Qwen3.6 hybrid prompt-cache fixes.
Preserve RAM prompt-cache entries when loading partially matching prompts so cached conversations survive switches.
github-actions
Bot
added
documentation
Regrad
pushed a commit
that referenced
this pull request
Jul 26, 2026
…rg#24908) * Add preliminary MiniMax-M3 support Text-only port that re-uses existing components: MiniMax-M2 style GQA with per-head QK-norm and partial rotary, DeepSeek-V3 style leading-dense and routed/shared experts, and swigluoai activation. Sparse attention is not yet supported (dense fallback); vision tower and MTP heads are dropped. * MiniMax-M3 vision tower (mmproj + clip graph) * Delete m3_vision_ref.py * Update clip.cpp * MSA * Update constants.py * Update minimax.py * Cache creation. Working withotu flash attention * Added flash attention for sparse layers * Decomposed slow cpu OP into GPU + CPU ops. Massive speedup over long ctx * Rewrote indexer op to be cuda native. Modified flash attention to match per group block picking * Implement sparse attention calc out of stock ops. * Fix a cache allocation and cont issue * Fixed -fa auto crash, flagged debug spots * Delete vocab.json * Delete model.safetensors.index.json * Delete generation_config.json * Delete Minimax directory * Handled multi stream case to fall back on Dense Attention * Development scaffolding cleanup. No functional change to the decode or 4-way paths. Full debug harness remains at <8136a9c68ed7a5eb009aa67bba3fda8062f4648f> for reproducing the selection-parity validation. * Remove redundant comment from minimax-m3.cpp * Changed 3 Gelu Ops for vision into Gelu_erf ops * Assert that n_kv is multiple of 128 * Rename MSA index tensors to indexer convention Note: All GGUFs generated before this change will need to be regenerated. * Fix incorrect Assert * Review driven changes (#3) * Remove comment from conversion minimax.py Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> * Remove whitespaces from constants.py Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> * Tighten comment in minimax.py Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> * inherit MiniMax-M3 from MiniMax-M2 * drop dead text_config fallbacks * Add indexer writer methods * Reuse LLM_FFN_SWIGLU_OAI_MOE * Remove duplicate indexer setters, add only block_size/local_blocks, follow value naming convention * Fix conversion error /gguf_writer.py Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> * Update gguf-py/gguf/gguf_writer.py Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> * Update gguf-py/gguf/tensor_mapping.py Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> * Update conversion/minimax.py Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> * Update conversion/minimax.py Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> * Remove whitespace in src/llama-kv-cache.cpp Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> * Remove Whitespace in Update src/llama-model.h Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> * Remove whitespace in src/llama-hparams.h Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> * remove multimodal code upon maintainer request. Will be made as a separate PR * Whitespace clean in tensor_mapping.py * Log cache size on launch, block ctx shift, support prompt caching Log indexer cache size on launch Disallow ctx shift Support prompt caching * Update minimax-m3.cpp * Optimize implementation, add multi stream support. Fully rewrote minimax-m3.cpp for speed and buffer size gains: Unified the 4-way + decode, 1 FA call per layer instead of 4, with the groups mapped onto ne[3] Custom CPU op now emits block-level mask, expanded on GPU, which causes CPU to GPU transfer to shrinks at prefill Decode: ~25 nodes/layer vs ~50, no per-group concats/conts Unified selection semantics, so both regimes rank bs + local bias (position-anchored local force), which means prefill/decode can no longer disagree on selection can_reuse on the MSA bias input. Graph reuse at decode restored (was rebuilding the full graph every token) In-place mask adds, shrinking compute buffer ~6.8 to ~4.2 GiB at ub2048/62k Multi-stream: MSA now runs with -np N when kv_unified=false. Decode stays batched across streams (still 1 FA call), prefill loops per stream. dense fallback only for --kv-unified + multi-seq Measured effect on expert offload bound setup: decode 6.2(4WAY)–7.15(MSA_decode) -> 7.7~7.8 t/s, flat from 5k to 60k+. prefill around 10% faster. buffer about 20% smaller, multi-user support. * set default cache type to F32 * Fix potential DSA double indexer cache allocation bug, only allocate in-cache k_idx for archs that opt in * remove F16 downcasts in MSA attention, force F32 indexer score accum * Add Minimax eos to llama vocab * Guard edge case where idx cache can become stale after a tail trim * Update llama-kv-cache.h * Update llama-kv-cache.cpp * Update llama-kv-cache.cpp * Update llama-kv-cache.h * Update llama-kv-cache.cpp * Review driven changes * style fix * indexer hparams are required * fix tests * fix lint --------- Co-authored-by: Daniel Han <danielhanchen@gmail.com> Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com> Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What is included
ggml-org/llama.cppmaster, including merged fix server: avoid unnecessary checkpoint restore when new tokens are present ggml-org/llama.cpp#24110.fix/qwen-hybrid-checkpoint-reuse.Why
Qwen3.6 hybrid/recurrent models can repeatedly fall back to full prompt re-processing. This work branch combines the available fixes without modifying
master.Notes
LLAMA_SERVER_SLOTS_DEBUG=1logging is retained. It already prints the old and new token sequences around the first mismatch.