]>
git.djapps.eu Git - pkg/ggml/sources/llama.cpp/log
uvos [Sun, 19 Apr 2026 10:59:44 +0000 (12:59 +0200)]
HIP: Remove unesscary NCCL_CHECK (#21914)
Xuan-Son Nguyen [Sun, 19 Apr 2026 09:57:21 +0000 (11:57 +0200)]
mtmd: add pos_0 to mtmd_image_tokens_get_decoder_pos (breaking change) (#22082)
* mtmd: add pos_0 to mtmd_image_tokens_get_decoder_pos
* fix build
Gaurav Garg [Sun, 19 Apr 2026 09:48:35 +0000 (15:18 +0530)]
ggml : reduce CPU overhead in meta backend (#22041)
* cache subgraph splits when cgraph is unchanged
Skip per-call subgraph construction in ggml_backend_meta_graph_compute when the same ggml_cgraph is used consecutively.
Assign uid to every sub-graph so that CUDA's fast uid check path hits too.
* Address review comments
* Keep the scope as is
* Rename last_uid and last_n_subgraphs field. Remove last_max_tmp_size field. Refactor code.
* Address review comments
* Update ggml/src/ggml-backend-meta.cpp
Co-authored-by: Johannes Gäßler <redacted>
* Update ggml/src/ggml-backend-meta.cpp
Co-authored-by: Johannes Gäßler <redacted>
---------
Co-authored-by: Johannes Gäßler <redacted>
Sigbjørn Skjæret [Sun, 19 Apr 2026 07:32:08 +0000 (09:32 +0200)]
ci : install spirv-headers for vulkan-cross (#22109)
Dowon [Sun, 19 Apr 2026 07:25:39 +0000 (16:25 +0900)]
convert : support sentence-transformer 5.4 config files (#22087)
* convert : support sentence-transformer 5.4 config files
* fix: embeddinggemma
* fix: mapping
Co-authored-by: Sigbjørn Skjæret <redacted>
* fix: pooling_mode
Co-authored-by: Sigbjørn Skjæret <redacted>
---------
Co-authored-by: Sigbjørn Skjæret <redacted>
texasich [Sun, 19 Apr 2026 07:25:05 +0000 (02:25 -0500)]
cmake: remove CMP0194 policy to restore MSVC builds (#21934)
#21630 added the CMP0194 NEW policy to silence a CMake warning, but on Windows runners it caused CMake to prefer the MinGW toolchain for ASM and broke MSVC builds.
Reverting only that policy block restores the previous working behavior. The CMake 4.1+ warning comes back, but that is cosmetic and does not break any platform.
Reported-by: oobabooga
Refs: #21630
Co-authored-by: texasich <redacted>
Sascha Rogmann [Sun, 19 Apr 2026 07:24:06 +0000 (09:24 +0200)]
server : speculative checkpointing (#19493)
* server : speculative decoding using checkpoints
* server : fix draft check with checkpoints
* server : rename spec vars
* server : log levels
* server : refactored spec logic to speculative.cpp
* server : renamed spec checkpoints option
* server : fix spec checkpoints, logging
* speculative : checkpoints with draft model, logging
* server : n_tokens_cur and create_checkpoint in draft
* server : fix server_speculative_callback (slot.id)
* spec : fix ngram-map/begin idx_last_check
* spec : init ckpt (begin() wasn't called)
* chore: update webui build output
* server : restore sampler in spec checkpoint and clear mem
* cont : avoid --spec-use-checkpoints argument
* cont : remove server_prompt_checkpoint_with_size
* spec : rename (leave_draft_state)
* cont : clean-up
* cont : do not ignore partial drafts even if the are short
* cont : spec callback owned by session
* cont : simplify
* cont : avoid empty speculative session
* cont : simplify
* cont : simplify
* cont : enable mtmd speculative decoding
* cont : keep the spec sampler alive
* cont : simplify
* cont : fix nullptr deref + draft checkpoints
* cont : remove common_speculative_accept_response
* cont : remove callback
* cont : simplify
* cont : minor
* cont : simplify
* cont : fix accepted number
---------
Co-authored-by: Georgi Gerganov <redacted>
Radoslav Gerganov [Sun, 19 Apr 2026 07:21:53 +0000 (10:21 +0300)]
rpc : refactor the RPC transport (#21998)
* rpc : refactor the RPC transport
Move all transport related code into a separate file and use the
socket_t interface to hide all transport implementation details.
* fix win32
* better socket_t construction
Cetarthoriphros [Sat, 18 Apr 2026 22:27:17 +0000 (19:27 -0300)]
server: Expose `media_tag` on /props endpoint. (#22028)
Sigbjørn Skjæret [Sat, 18 Apr 2026 18:12:00 +0000 (20:12 +0200)]
model : refactor bias tensor variable names (#22079)
* refactor bias tensor variable names
* use create_tensor_qkv for jina-bert-v2
Sigbjørn Skjæret [Sat, 18 Apr 2026 09:19:40 +0000 (11:19 +0200)]
android : libcommon -> libllama-common (#22076)
SamareshSingh [Sat, 18 Apr 2026 08:04:51 +0000 (03:04 -0500)]
ggml-backend-meta: add multi-segment read support in get_tensor (#22063)
Sigbjørn Skjæret [Sat, 18 Apr 2026 07:37:30 +0000 (09:37 +0200)]
ci : free disk space for rocm release (#22012)
Sigbjørn Skjæret [Sat, 18 Apr 2026 07:36:41 +0000 (09:36 +0200)]
convert : fix (ignore for now) typings errors (#22002)
Johannes Gäßler [Sat, 18 Apr 2026 06:16:04 +0000 (08:16 +0200)]
llama: fit ctx size for CPU only (#21568)
Reese Levine [Fri, 17 Apr 2026 16:17:11 +0000 (09:17 -0700)]
ggml-webgpu: fix compiler warnings and refactor FlashAttention encoding (#21052)
* Update workflows to remove dependence on llvmpipe
* Try setting Dawn_DIR
* remove c++20 initializers
* Move to proper guid
* Try avoiding segfaults on vulkan backend process exit
* Remove compiler warnings on parameter casting
* Fix soft_max and update reg_tile accumulation to f32 for better precision
* Refactor flash_attn a bit
* remove c++20 initializers and format
* Increase div precision for NVIDIA
* revert div precision and comment out ggml-ci node for now
* Formatting
* Try debugging on a failing CI node
* Revert "Try debugging on a failing CI node"
This reverts commit
1971e33cba919915e12bcfd5828abfbd54ca942e .
Aman Gupta [Fri, 17 Apr 2026 15:24:21 +0000 (23:24 +0800)]
CUDA: use LRU based eviction for cuda graphs (#21611)
* CUDA: use a ring-buffer for cuda graphs
* bump limit to 128
* use LRU eviction
* better naming
* do periodic clean-up
Yuri Khrustalev [Fri, 17 Apr 2026 09:32:24 +0000 (05:32 -0400)]
ci : add android arm64 build and release (#21647)
* server: respect the ignore eos flag
* ci: add android arm64 build and release
* patch
* pin android-setup actions to v4
* Apply suggestions from code review
Co-authored-by: Sigbjørn Skjæret <redacted>
* lf in the suggestion
---------
Co-authored-by: Sigbjørn Skjæret <redacted>
65a [Fri, 17 Apr 2026 08:48:33 +0000 (01:48 -0700)]
mtmd: add missing struct tag (#22023)
Georgi Gerganov [Fri, 17 Apr 2026 08:11:46 +0000 (11:11 +0300)]
libs : rename libcommon -> libllama-common (#21936)
* cmake : allow libcommon to be shared
* cmake : rename libcommon to libllama-common
* cont : set -fPIC for httplib
* cont : export all symbols
* cont : fix build_info exports
* libs : add libllama-common-base
* log : add common_log_get_verbosity_thold()
Eric Zhang [Fri, 17 Apr 2026 08:07:11 +0000 (16:07 +0800)]
model : Gemma4 model type detection (#22027)
* model : Gemma4 model type detection
* model : Gemma4 model type detection
lhez [Fri, 17 Apr 2026 05:28:33 +0000 (22:28 -0700)]
opencl: refactor q8_0 set_tensor and mul_mat host side dispatch for Adreno (#21938)
* opencl: refactor q8_0 gemm/gemv Adreno dispatch
* opencl: refactor q8_0 set_tensor
* opencl: fix whitespace
Sigbjørn Skjæret [Thu, 16 Apr 2026 22:12:31 +0000 (00:12 +0200)]
cli : use get_media_marker (#22017)
Xuan-Son Nguyen [Thu, 16 Apr 2026 21:25:16 +0000 (23:25 +0200)]
cmake: use glob to collect src/models sources (#22005)
nullname [Thu, 16 Apr 2026 20:48:34 +0000 (04:48 +0800)]
hexagon: optimize HMX matmul operations (#21071)
* optimize hmx_mat_mul functions by calculating row and column tiles upfront
* refactor core_dot_chunk_fp16 to use size_t for tile counts and improve readability
* wip
* set scale outside of loop
* wip
* refactor core_mma_chunk_fp16 and mat_mul_qk_0_d16a32 to use size_t for tile counts
* wip
* wip
* refactor transfer_output_chunk_fp16_to_fp32 to use size_t for dimensions
* refactor core_dot_chunk_fp16 to use size_t for tile row stride calculation
* wip
* refactor hmx_mat_mul functions to use hvx_vec_splat_f16 for column scales initialization
* refactor hmx_mat_mul_permuted_w16a32_batched to streamline scale setting and locking
* refactor core_dot_chunk_fp16 to improve tile stride calculations for output
* refactor hmx_mat_mul functions to use Q6_V_vsplat_R for column scales initialization
* fix compiling error
* wip
* optimize row and column tile indexing in core_mma_chunk_fp16 function
* wip
* Revert "wip"
This reverts commit
cde679eff79c4a28dd2d89d32f710015e09592b6 .
* Add size limit check for HAP_mmap in htp_iface_mmap and drop_mmap functions
* wip
Xuan-Son Nguyen [Thu, 16 Apr 2026 19:10:22 +0000 (21:10 +0200)]
model: using single llm_build per arch (#21970)
* model: using single llm_build per arch
* fix merge
* nits
shaofeiqi [Thu, 16 Apr 2026 19:08:33 +0000 (12:08 -0700)]
opencl: add q5_K gemm and gemv kernels for Adreno (#21595)
Pascal [Thu, 16 Apr 2026 17:46:21 +0000 (19:46 +0200)]
server: tests: fetch random media marker via /apply-template (#21962) (#21980)
* server: tests: fetch random media marker via /apply-template (#21962 fix)
* server: allow pinning media marker via LLAMA_MEDIA_MARKER env var
get_media_marker() checks LLAMA_MEDIA_MARKER at first call and uses it
as-is if set, falling back to the random marker otherwise.
Tests no longer need to fetch the marker dynamically via /apply-template:
the fixture sets LLAMA_MEDIA_MARKER=<__media__> so the hardcoded prompts
work as before.
Address review feedback from ngxson
* server: make get_media_marker() thread-safe via magic statics
Use a C++11 static local with a lambda initializer instead of a global
static with an empty-check. The runtime guarantees initialization exactly
once without explicit locking.
Address review feedback from ggerganov
* nits
* nits
PikaPikachu [Thu, 16 Apr 2026 15:41:34 +0000 (23:41 +0800)]
model : refactor QKV into common build_qkv and create_tensor_qkv helpers (#21245)
* model : refactor QKV into common build_qkv and create_tensor_qkv helpers
* model : extend build_qkv to bert/mpt/dbrx/olmo/lfm2/nemotron-h/granite-hybrid/gemma3n-iswa/t5-dec and fix wqkv_s
Sigbjørn Skjæret [Thu, 16 Apr 2026 14:51:47 +0000 (16:51 +0200)]
model : support NVFP4 tensors for Gemma4 (#21971)
* support nvfp4 tensors for Gemma4
* add wo_s to build_attn
* add wo_s to build_attn
* fix glm4
Ruben Ortlam [Thu, 16 Apr 2026 10:13:11 +0000 (12:13 +0200)]
codeowners: add team member comments (#21714)
Anav Prasad [Thu, 16 Apr 2026 10:11:45 +0000 (10:11 +0000)]
Convert: Fix NemotronH Config Parsing (#21664)
* fix NemotronH vocab loading by using trust_remote_code for unsupported config patterns
* fix NemotronH tokenizer loading by overriding set_vocab with trust_remote_code
Aman Gupta [Thu, 16 Apr 2026 09:21:28 +0000 (17:21 +0800)]
ggml: add graph_reused (#21764)
* ggml: add graph_reused
* use versioning instead of reuse flag
* increment version with atomic
* use top bits for split numbering
* add assert
* move counter to ggml.c
* set uid in split_graph only
* fix windows
* address further review comments
* get next_uid rather than doing bit manipulation
* rename + add comment about uid
Kusha Gharahi [Thu, 16 Apr 2026 08:54:37 +0000 (03:54 -0500)]
metal: Implement ROLL op (#21946)
* nix: support unified apple-sdk
* Impl roll op for Metal
* Revert "nix: support unified apple-sdk"
This reverts commit
abfa473360471532c547de8b202c780507924d4b .
* update ops.md
* update op docs
rehan-10xengineer [Thu, 16 Apr 2026 08:15:15 +0000 (13:15 +0500)]
ggml-cpu: add 128-bit RVV implementation for Quantization Vector Dot (#20633)
* ggml-cpu: add 128-bit impls for i-quants, ternary quants
* ggml-cpu: add 128-bit impls for iq2_xs, iq3_s, iq3_xxs, tq2_0
Co-authored-by: Rehan Qasim <redacted>
* ggml-cpu: refactor; add rvv checks
---------
Co-authored-by: taimur-10x <redacted>
Co-authored-by: Rehan Qasim <redacted>
rehan-10xengineer [Thu, 16 Apr 2026 08:14:26 +0000 (13:14 +0500)]
ggml : implemented simd_gemm kernel for riscv vector extension (#20627)
Co-authored-by: Rehan Qasim <redacted>
Yuannan [Thu, 16 Apr 2026 08:12:52 +0000 (08:12 +0000)]
devops : added spirv-headers to nix (#21965)
Reese Levine [Thu, 16 Apr 2026 08:12:19 +0000 (01:12 -0700)]
ggml-webgpu: compute pass batching and removing profiling overhead (#21873)
* Update register tiling matmul to use f32 accumulation
* fix profiling code
* Fix register tiling matmul for chrome, i'm blaming dawn
* Update batch tuning value for iOS
* compile fix
* Fix use of new load function
* Move to a single query set for GPU profiling
* Move to batching compute passes when not profiling
* Refactor build_multi
* remove iOS throttling now that we're batching compute passes
Ludovic Henry [Thu, 16 Apr 2026 08:11:25 +0000 (10:11 +0200)]
ci : Use ggml-org/ccache-action on RISC-V as well (#21632)
Katostrofik [Thu, 16 Apr 2026 05:34:05 +0000 (01:34 -0400)]
[SYCL] Fix Q8_0 reorder: garbage on 2nd prompt + crash on full VRAM (#21638)
* [SYCL] Fix Q8_0 reorder: add missing dequantize path for GEMM
The Q8_0 reorder optimization (#21527) was missing a reorder-aware
dequantizer for the GEMM code path used during prompt processing.
After token generation reordered Q8_0 weights (via DMMV/MMVQ), the
next prompt processing pass would read them with the standard
dequantizer, producing garbage output.
Add dequantize_block_q8_0_reorder() and wire it into both
ggml_get_to_fp16_sycl() and ggml_get_to_fp32_sycl(), matching the
pattern already used by Q4_0, Q4_K, and Q6_K.
Fixes #21589
AI (Claude) was used to assist with root cause investigation and
writing the kernel code. All code was human-reviewed and tested
on real hardware.
* SYCL: fix reorder crash when device memory is full
The reorder optimization allocates a temporary buffer the full size of
the weight tensor on the device. When VRAM is nearly full (large models
on a single GPU), this allocation fails and the subsequent memcpy crashes
on a NULL pointer.
Fix: try device allocation first, fall back to host memory if device
memory is full. The reorder kernel still works correctly reading from
host memory over PCIe. This is slower for the one-time reorder (~21 t/s
vs ~38 t/s on Intel Arc Pro B70), but the optimization is preserved for
all subsequent inference. If both device and host allocation fail, skip
the reorder and fall back to the unoptimized kernel path.
Also fixes a bug where opt_for_reorder() marked tensors as reordered
even when the reorder was skipped due to allocation failure. This caused
DMMV/MMVQ kernels to read the original AoS data as if it were SoA,
producing garbage output or NaN results.
Tested on Intel Arc Pro B70 (32GB) with Q8_0, Q4_K_M models. Coding was
AI-assisted (Claude), reviewed and tested on hardware by a human.
Fixes #20478
* SYCL: add RAII temp buffer class + macro guard for host fallback
Replace sycl_ext_malloc_with_fallback/sycl_ext_free_fallback free
functions with sycl_reorder_temp_buffer RAII class. The host_fallback
bool is now a private member, and cleanup happens automatically at
scope exit.
Add GGML_SYCL_HOST_MEM_FALLBACK cmake option (default ON) to guard
the host memory fallback code path. Device access to host memory
requires Linux kernel 6.8+ (Ubuntu 26.04+); users on older kernels
can set -DGGML_SYCL_HOST_MEM_FALLBACK=OFF to disable it.
Addresses arthw's review on PR #21638.
Co-Authored-By: Claude Opus 4.6 (1M context) <redacted>
* SYCL: document GGML_SYCL_HOST_MEM_FALLBACK build option in SYCL.md
Co-Authored-By: Claude Opus 4.6 (1M context) <redacted>
* SYCL: add reorder-aware DMMV dequantizers for Q4_K and Q6_K
Q4_K and Q6_K had reorder support for MMVQ and GEMM paths but not
DMMV. When the DMMV path encountered reordered data it would abort.
Add DMMV kernels that read from the SOA reorder layout for both
types. Same math as the non-reorder versions, different memory
access pattern.
Co-Authored-By: Claude Opus 4.6 (1M context) <redacted>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <redacted>
Xuan-Son Nguyen [Wed, 15 Apr 2026 21:52:22 +0000 (23:52 +0200)]
server: use random media marker (#21962)
* server: use random media marker
* nits
* remove legacy <__image__> token
* revert special char in random
Ruben Ortlam [Wed, 15 Apr 2026 17:04:51 +0000 (19:04 +0200)]
vulkan: optimize im2col (#21713)
* vulkan: improve im2col memory write layout
* cap workgroups
* minimal device tuning
* use vendor_id instead of subgroup size
Pasha Khosravi [Wed, 15 Apr 2026 16:38:38 +0000 (09:38 -0700)]
cuda: Q1_0 initial backend (#21629)
* [cuda] initial Q1_0 backend
* remove unused code, fix AMD MMA guard
* attempt to support dp4a
* Apply suggestions from code review
Co-authored-by: Johannes Gäßler <redacted>
---------
Co-authored-by: Johannes Gäßler <redacted>
Reese Levine [Wed, 15 Apr 2026 16:14:40 +0000 (09:14 -0700)]
ggml-webgpu: Fix dequantization helpers to not pass in pointers (#21872)
* Fix dequantization helpers to not pass in pointers
* Increase XIELU precision
Johannes Gäßler [Wed, 15 Apr 2026 14:01:46 +0000 (16:01 +0200)]
CUDA: require explicit opt-in for P2P access (#21910)
Johannes Gäßler [Wed, 15 Apr 2026 13:58:40 +0000 (15:58 +0200)]
CUDA: manage NCCL communicators in context (#21891)
* CUDA: manage NCCL communicators in context
* add check that all backends are CUDA
* remove unused vector, limit init to > 1 GPUs
* fix warnings
* fix cuda device, cache allreduce
Valeriy Dubov [Wed, 15 Apr 2026 13:44:02 +0000 (16:44 +0300)]
rpc : add native RDMA transport for RPC backend (RoCEv2) (#20590)
Xuan-Son Nguyen [Wed, 15 Apr 2026 12:45:16 +0000 (14:45 +0200)]
docs: more extensive RoPE documentation [no ci] (#21953)
* more extensive ggml_rope documentation
* add more docs
* nits
Ruben Ortlam [Wed, 15 Apr 2026 08:55:21 +0000 (10:55 +0200)]
ci: disable test-backend-ops on Vulkan llvmpipe run and resture default timeout (#21901)
Piotr Wilkin (ilintar) [Wed, 15 Apr 2026 08:51:50 +0000 (10:51 +0200)]
autoparser: support case of JSON_NATIVE with per-call markers (test case: Reka-Edge) (#21892)
Matt [Wed, 15 Apr 2026 07:24:57 +0000 (00:24 -0700)]
read n_ctx back after making llama_context (#21939)
Yiwei Shao [Tue, 14 Apr 2026 21:09:03 +0000 (14:09 -0700)]
hexagon: optimization for HMX mat_mul (#21554)
* hexagon: add async HMX worker
Introduce hmx-worker (dedicated thread for HMX compute) to overlap HMX
matmul with HVX dequant/DMA stages in the pipeline path, replacing the
previous synchronous HMX calls that blocked the main thread.
* hexagon: cost-based VTCM chunk search for out-stationary matmul
* hexagon: fix futex race in hmx_worker_drain
Store the boolean to local variable avoid atomic load twice
* hex-mm: hmx optimize scatter/transpose and use HMX intrinsics
* hex-vmem: drop vmem limit a touch under 3GB on v73
* hexagon: add fwd declaration of htp_context
* hex-hmx: replace hmx-worker with hmx-queue that mimics dma-queue interface
Simplifies the overall implemantion, reduces thread wakeup roundtrips.
* hex-mm: add debug log to hmx work func called from hmx-queue
* Update hmx-queue.h
Co-authored-by: Max Krasnyansky <redacted>
---------
Co-authored-by: Kim-Chyan Gan <redacted>
Co-authored-by: Max Krasnyansky <redacted>
Co-authored-by: Max Krasnyansky <redacted>
Xuan-Son Nguyen [Tue, 14 Apr 2026 14:32:58 +0000 (16:32 +0200)]
ggml : remove ggml-ext.h (#21869)
* ggml: correct placement of ggml-ext.h
* ggml : remove ggml-ext.h
---------
Co-authored-by: Georgi Gerganov <redacted>
Georgi Gerganov [Tue, 14 Apr 2026 14:32:29 +0000 (17:32 +0300)]
metal : fix FA support logic (#21898)
Xuan-Son Nguyen [Tue, 14 Apr 2026 14:07:41 +0000 (16:07 +0200)]
mtmd: add mtmd_image_tokens_get_decoder_pos() API (#21851)
* mtmd: add mtmd_image_tokens_get_decoder_pos() API
* consistent naming
* fix build
Jeff Bolz [Tue, 14 Apr 2026 13:17:45 +0000 (15:17 +0200)]
vulkan: Programmatically add RoundingModeRTE to all shaders when the device supports it (#21572)
* vulkan: Programmatically add RoundingModeRTE to all shaders when the device supports it
* use FetchContent to get SPIRV-Headers
* Fetch spirv-headers unconditionally
* remove fetchcontent, rely on installed headers
* fix ubuntu job
* Update docs/build.md
Georgi Gerganov [Tue, 14 Apr 2026 12:58:09 +0000 (15:58 +0300)]
ci : re-enable mac workflows (#21894)
* ci : re-enable mac workflows
* vulkan : fix compile warning
Seyoung Jeong [Tue, 14 Apr 2026 12:43:59 +0000 (21:43 +0900)]
metal : add XIELU unary op (#20802)
Adrien Gallouët [Tue, 14 Apr 2026 11:25:09 +0000 (13:25 +0200)]
vendor : update BoringSSL to 0.
20260413 .0 (#21881)
Signed-off-by: Adrien Gallouët <redacted>
Richard Davison [Tue, 14 Apr 2026 11:23:45 +0000 (13:23 +0200)]
ggml : fix ARM NEON nvfp4 dot product on non-dotprod targets (#21559)
texasich [Tue, 14 Apr 2026 10:47:56 +0000 (05:47 -0500)]
cmake: fix CMP0194 warning on Windows with MSVC (#21630)
* cmake: fix CMP0194 warning on Windows with MSVC
Set CMP0194 policy to NEW before project() call in ggml/CMakeLists.txt to suppress the "MSVC is not an assembler for language ASM" warning introduced in CMake 4.1.
The ggml project enables ASM globally for Metal (macOS) and KleidiAI (ARM) backends. On Windows/MSVC, no assembler sources are used, but CMake 4.1+ warns because cl.exe is not a valid ASM compiler.
This follows the same pattern used in ggml-vulkan (CMP0114, CMP0147).
Closes ggml-org/llama.cpp#20311
* cmake: apply cisc's formatting suggestion
---------
Co-authored-by: texasich <redacted>
Reese Levine [Tue, 14 Apr 2026 10:46:41 +0000 (03:46 -0700)]
ggml-webgpu: Update register tiling matmul to use f32 accumulation (#21644)
* Update register tiling matmul to use f32 accumulation
* fix profiling code
* Fix register tiling matmul for chrome, i'm blaming dawn
* Update batch tuning value for iOS
* compile fix
* Fix use of new load function
Berk Idem [Tue, 14 Apr 2026 10:43:06 +0000 (06:43 -0400)]
common: skip reasoning budget sampler when no budget is requested (#21870)
* common: skip reasoning budget sampler when no budget is requested
After I added thinking_start_tag / thinking_end_tag for gemma4 in #21697, the reasoning budget sampler gets unconditionally created even when no budget is configured (the default -1). The same applies to kimi_k2, lfm2, lfm2_5, and ministral_3 which also set these tags. The budget gets converted to INT_MAX, so the sampler never actually forces any tokens but still runs per-token checks (start tag matching in IDLE state, token-to-piece conversion + UTF-8 checks in COUNTING state).
More importantly, the mere existence of the sampler (non-null rbudget) disables backend sampling. Backend sampling lets the GPU select tokens directly, avoiding a full logits transfer from GPU to CPU every token. This could explain the 30% speed regression reported in #21784 (98 t/s to 70 t/s on Vulkan).
So I added a reasoning_budget_tokens >= 0 check to the sampler creation condition. When the budget is unlimited, the sampler is not created, backend sampling stays enabled, and no per-token overhead is added. When a budget is explicitly set (0, 128, 1024, etc.), the sampler is created and works as before.
* common: preserve rbudget when grammar is lazy
Following up on the review feedback on #21870: keep the reasoning budget sampler when grammar_lazy is true, so the thinking-block grammar suppression from #20970 still works when tools are in use. This way, we only skip the sampler when both no budget is set AND grammar is not lazy.
Jeff Bolz [Tue, 14 Apr 2026 09:34:23 +0000 (11:34 +0200)]
vulkan: Support GGML_TYPE_NVFP4 (#21455)
This adds nvfp4 support for get_rows, dequant, and mul_mat(_id). For
mul_mat, it does not add support for the dp4/q8_1 path, it's all via
fp16/fp32.
Xuan-Son Nguyen [Tue, 14 Apr 2026 09:09:52 +0000 (11:09 +0200)]
server: support OAI /v1/audio/transcriptions API (#21863)
* server: support OAI /v1/audio/transcriptions API
* address autoreview comments
* correct default response_format value
Aldehir Rojas [Mon, 13 Apr 2026 23:18:18 +0000 (18:18 -0500)]
common/gemma4 : handle parsing edge cases (#21760)
Xuan-Son Nguyen [Mon, 13 Apr 2026 20:28:17 +0000 (22:28 +0200)]
docs: listing qwen3-asr and qwen3-omni as supported (#21857)
* docs: listing qwen3-asr and qwen3-omni as supported
* nits
Piotr Wilkin (ilintar) [Mon, 13 Apr 2026 20:23:53 +0000 (22:23 +0200)]
chat: dedicated DeepSeek v3.2 parser + "official" template (#21785)
Christian Kastner [Mon, 13 Apr 2026 17:18:44 +0000 (19:18 +0200)]
ci: Also exempt 'security' tag from auto-close (#21844)
Ruben Ortlam [Mon, 13 Apr 2026 12:21:31 +0000 (14:21 +0200)]
vulkan: Flash Attention DP4A shader for quantized KV cache (#20797)
* use integer dot product for quantized KV flash attention
* small improvements
* fix SHMEM_STAGING indexing
* add missing KV type quants
* fixes
* add supported quants to FA tests
* readd fast paths for <8bit quants
* fix mmq gate and shmem checks
Adrien Gallouët [Mon, 13 Apr 2026 09:18:23 +0000 (11:18 +0200)]
common : add download cancellation and temp file cleanup (#21813)
Signed-off-by: Adrien Gallouët <redacted>
Gaspard Petit [Mon, 13 Apr 2026 09:14:42 +0000 (05:14 -0400)]
server: Expose build_info in router mode (#21835)
Oliver Simons [Mon, 13 Apr 2026 09:14:06 +0000 (11:14 +0200)]
CUDA: Limit DeviceSegmentedSort to immediate mode (#21718)
* CUDA: Limit DeviceSegmentedSort to immediate mode
DeviceSegmentedSort is currently not capturable in a cuda graph. Hence,
we have to go for the slower DeviceSegmentedRadixSort in that case.
Perf numbers on RTX Pro 6000 Blackwell Max-Q:
DeviceSegmentedRadixSort in graph mode (i.e. CUDA Graphs)
ARGSORT(type=f32,ne=[2048,512,1,1],order=1): 12291 runs - 105.94 us/run - 8192 kB/run - 73.75 GB/s
ARGSORT(type=f32,ne=[4096,512,1,1],order=1): 10245 runs - 115.08 us/run - 16384 kB/run - 135.77 GB/s
ARGSORT(type=f32,ne=[8192,512,1,1],order=1): 5125 runs - 221.22 us/run - 32768 kB/run - 141.26 GB/s
ARGSORT(type=f32,ne=[16384,512,1,1],order=1): 2565 runs - 430.98 us/run - 65536 kB/run - 145.02 GB/s
ARGSORT(type=f32,ne=[32768,512,1,1],order=1): 1028 runs - 1185.83 us/run - 131072 kB/run - 105.41 GB/s
ARGSORT(type=f32,ne=[65536,512,1,1],order=1): 387 runs - 2748.62 us/run - 262144 kB/run - 90.95 GB/s
DeviceSegmentedSort in immediate mode
ARGSORT(type=f32,ne=[2048,512,1,1],order=1): 16388 runs - 71.17 us/run - 8192 kB/run - 109.78 GB/s
ARGSORT(type=f32,ne=[4096,512,1,1],order=1): 12294 runs - 81.38 us/run - 16384 kB/run - 192.00 GB/s
ARGSORT(type=f32,ne=[8192,512,1,1],order=1): 5125 runs - 240.81 us/run - 32768 kB/run - 129.77 GB/s
ARGSORT(type=f32,ne=[16384,512,1,1],order=1): 2565 runs - 406.60 us/run - 65536 kB/run - 153.71 GB/s
ARGSORT(type=f32,ne=[32768,512,1,1],order=1): 1285 runs - 873.23 us/run - 131072 kB/run - 143.15 GB/s
ARGSORT(type=f32,ne=[65536,512,1,1],order=1): 516 runs - 2288.46 us/run - 262144 kB/run - 109.24 GB/s
* Add test case for dispatch to DeviceSegmentedRadixSort
We currently lack a way to force graph mode in CUDA, patch callback to
invoke ggml_backend_compare_graph_backend twice to enforce each test to
run in graph mode
Xuan-Son Nguyen [Mon, 13 Apr 2026 07:47:55 +0000 (09:47 +0200)]
mtmd: use causal attn for gemma 4 audio (#21824)
Rohan Jain [Mon, 13 Apr 2026 07:30:46 +0000 (13:00 +0530)]
webui: add setting for first-line chat titles (#21797)
* webui: add setting for first-line chat titles
Add an opt-in setting (`titleGenerationUseFirstLine`) to use the first
non-empty line of a prompt as the generated conversation title.
Previously, the complete multi-line prompt was being used, which created
long titles for complex queries. Coupled with
"Ask for confirmation before changing conversation title", the dialog
would overflow.
* Update tools/server/webui/src/lib/utils/text.ts
Co-authored-by: Aleksander Grygier <redacted>
* Update tools/server/webui/src/lib/utils/text.ts
Co-authored-by: Aleksander Grygier <redacted>
* webui: Run build to update the bundle
As requested in:
https://github.com/ggml-org/llama.cpp/pull/21797#pullrequestreview-
4094935065
* webui: Fix missing import for NEWLINE_SEPARATOR
---------
Co-authored-by: Aleksander Grygier <redacted>
Aleksander Grygier [Mon, 13 Apr 2026 05:58:38 +0000 (07:58 +0200)]
webui: MCP Diagnostics improvements (#21803)
* Add MCP Connection diagnostics and CORS hint to web-ui
* tidy up test
* webui: Refactor and improve MCP diagnostic logging
---------
Co-authored-by: evalstate <redacted>
Masashi Yoshimura [Mon, 13 Apr 2026 03:13:04 +0000 (12:13 +0900)]
Remove extra conditional check on debug mode. (#21798)
Akarshan Biswas [Mon, 13 Apr 2026 01:44:58 +0000 (07:14 +0530)]
sycl: disable Q1_0 in backend and cleanup unused variables (#21807)
Sergiu [Sun, 12 Apr 2026 21:59:21 +0000 (00:59 +0300)]
mtmd: fix crash when sending image under 2x2 pixels (#21711)
Xuan-Son Nguyen [Sun, 12 Apr 2026 21:57:25 +0000 (23:57 +0200)]
mtmd: qwen3 audio support (qwen3-omni and qwen3-asr) (#19441)
* add qwen3a
* wip
* vision ok
* no more deepstack for audio
* convert ASR model ok
* qwen3 asr working
* Apply suggestions from code review
Co-authored-by: Sigbjørn Skjæret <redacted>
* nits
* Apply suggestions from code review
Co-authored-by: Sigbjørn Skjæret <redacted>
* fix bad merge
* fix multi inheritance
---------
Co-authored-by: Sigbjørn Skjæret <redacted>
Sigbjørn Skjæret [Sun, 12 Apr 2026 17:22:29 +0000 (19:22 +0200)]
convert : force f16 or f32 on step3-vl conv weights (#21646)
Xuan-Son Nguyen [Sun, 12 Apr 2026 14:29:03 +0000 (16:29 +0200)]
mtmd: add gemma 4 test (vision + audio) [no ci] (#21806)
* mtmd: add gemma 4 test (vision + audio)
* add to docs
Stephen Cox [Sun, 12 Apr 2026 12:15:26 +0000 (00:15 +1200)]
mtmd: add Gemma 4 audio conformer encoder support (#21421)
* mtmd: add Gemma 4 audio conformer encoder support
Add audio processing for Gemma 4 E2B/E4B via a USM-style Conformer.
Architecture:
- 12-layer Conformer: FFN → Self-Attention → Causal Conv1D → FFN → Norm
- Subsampling Conv Projection: 2x Conv2D(stride=2) with LayerNorm
- Full self-attention with sinusoidal RPE and sliding window mask (24)
- Logit softcapping at 50.0, ClippableLinear clamping
- Output: 1024 → 1536 → RMSNorm → multimodal embedder
Mel preprocessing (dedicated mtmd_audio_preprocessor_gemma4a):
- HTK mel scale, 128 bins, magnitude STFT, mel_floor=1e-3
- Standard periodic Hann window (320 samples), zero-padded to FFT size
- Semicausal left-padding (frame_length/2 samples)
- Frame count matched to PyTorch (unfold formula)
- No pre-emphasis, no Whisper-style normalization
- Mel cosine similarity vs PyTorch: 0.9998
Key fixes:
- Tensor loading dedup: prevent get_tensor() from creating duplicate
entries in ctx_data. Fixed with std::set guard.
- ClippableLinear clamp_info loading moved after per-layer tensors.
- Sliding window mask (24 positions) matching PyTorch context_size.
- Skip Whisper normalization for Gemma4 mel output.
Tested on E2B and E4B with CPU and Vulkan backends.
Transcribes: "Glad to see things are going well and business is starting
to pick up" (matching ground truth).
Ref: #21325
Aleksander Grygier [Sun, 12 Apr 2026 11:08:11 +0000 (13:08 +0200)]
fix: Proper messages rendering for "Show raw output" (#21672)
Xuan-Son Nguyen [Sun, 12 Apr 2026 11:02:38 +0000 (13:02 +0200)]
docs: add guide on how to add multimodal support (#21778)
* docs: add guide on how to add multimodal support
* nits
Johannes Gäßler [Sat, 11 Apr 2026 16:52:11 +0000 (18:52 +0200)]
CUDA: skip compilation of superfluous FA kernels (#21768)
Sirui He [Sat, 11 Apr 2026 12:15:48 +0000 (20:15 +0800)]
mtmd : add MERaLiON-2 multimodal audio support (#21756)
* mtmd : add MERaLiON-2 multimodal audio support
Adds support for A*STAR's MERaLiON-2 audio-language model (3B and 10B)
to the multimodal framework.
Architecture:
- Whisper large-v2 encoder for audio feature extraction
- Gated MLP adaptor: ln_speech -> frame stack (x15) -> Linear+SiLU -> GLU -> out_proj
- Gemma2 3B / 27B decoder
The mmproj GGUF is generated via convert_hf_to_gguf.py --mmproj on the full
MERaLiON-2 model directory (architecture: MERaLiON2ForConditionalGeneration).
The decoder is converted separately as a standard Gemma2 model after stripping
the text_decoder. weight prefix.
New projector type: PROJECTOR_TYPE_MERALION
Supports tasks: speech transcription (EN/ZH/MS/TA), translation, spoken QA.
Model: https://huggingface.co/MERaLiON/MERaLiON-2-3B
https://huggingface.co/MERaLiON/MERaLiON-2-10B
* simplify comments in meralion adaptor
* meralion: use format_tensor_name, ascii arrows in comments
shaofeiqi [Sat, 11 Apr 2026 08:46:19 +0000 (01:46 -0700)]
opencl: add basic support for q5_k (#21593)
* opencl: add general q5_k mv
* opencl: add flattened Q5_K mv and general Q5_K mm
* opencl: fix Q5_K unit tests
Johannes Gäßler [Sat, 11 Apr 2026 07:23:42 +0000 (09:23 +0200)]
TP: fix Qwen 3 Next data split (#21732)
Sigbjørn Skjæret [Sat, 11 Apr 2026 06:45:00 +0000 (08:45 +0200)]
ggml : fix a few instances of missing GGML_TYPE_Q1_0 cases (#21716)
Bartowski [Sat, 11 Apr 2026 06:44:15 +0000 (02:44 -0400)]
py : Bump typer to latest to fix huggingface_hub issue (#21701)
Aman Gupta [Sat, 11 Apr 2026 02:30:30 +0000 (10:30 +0800)]
CUDA: also store node->src ne/nb for graph equality (#21736)
Galunid [Fri, 10 Apr 2026 23:26:36 +0000 (01:26 +0200)]
fix: Fix broken structured output when using $refs in json_schema (#21699)
Todor Boinovski [Fri, 10 Apr 2026 22:57:23 +0000 (15:57 -0700)]
hexagon: add support for linux on snapdragon (#21707)
* hexagon: add support for debian on ex2
* hexagon: add -fvectotize to c/c++ cmake flags
* hexagon: remove trailing white space
* update onboarding steps
* hexagon: update linux setup documentation
* hexagon: update intallation scripts
* Hexagon: update docs
* hexagon: update onboarding scripts
---------
Co-authored-by: Zack Li <redacted>
Max Krasnyansky [Fri, 10 Apr 2026 22:47:43 +0000 (15:47 -0700)]
hexagon: improved Op queuing, buffer and cache management (#21705)
* hexagon: introduce op request batching and rewrite buffer managment
The host now prepares batches of requests and dispatches them via a single dspqueue message.
Buffers are mapped explicitly by NPU while processing batches.
* hex-dma: disable l2 bypass since to work around new issue due to no flushes between Ops
* hex-utils: add explicit l2flush and l2clear helpers
* hex-opreq: use fine-grain per tensor l2 management
* hex-opreq: avoid redundant invalidates for tensors we already flushed
* hex-opreq: update debug messages
* htp-opreq: reuse ops_context
* hex-opreq: do not flush or invalidate cache lines beyond buffer boundry
* hex-opreq: fix errors in log message
* Revert "hex-opreq: do not flush or invalidate cache lines beyond buffer boundry"
This reverts commit
8b7f0a55a750a6430ce4eb1874c7feb3d720056d .
* hexagon: limit l2 flushes to 1MB which covers l2 cache
* hex-opreq: limit cache flush to 4MB
Looks like 4MB cont. vitual space should cover the 1MB cache.
* hexagon: drop cache flush size to 2MB
* hex-opreq: start reworking opreq packing
* hex-opreq: introduce new way of packing opbatch where tensors are stored separately
* hex-opreq: add a simple fastrpc call to force unmap all buffers
* hex-l2flush: somehow 2MB does not seem robust, also cleanup step size to use line-size
* hex-opreq: bump opreq batch size to 256
* hex-mm: place src1 spad at the top of vtcm for easy reuse
* hex-ops: introduce internal types and disable src1 reuse for now
Nothing new just formalizing the repack / qyn.quant types we've been using.
* htp-opreq: use tensor pointers instead of copies
* hex-opreq: introduce more robust way for tracking vtcm/spad reuse
This removes the SKIP_QUANTIZE flag that became fragile with the addition of HMX and other ops.
* hex-cumsum: fix error post opreq merge
* hex-opreq: move request batch handling into the session
Prepping everything for using dspqueue buffers and doing that inside the session is much cleaner.
* hex-mm: yet another fix for src1 reuse when we're mixing hmx/hvx
* hex-bufs: introduce pinned mmapings and use non-pinned ones for model buffers
* hex-buf: add support for allocating shared/pinned buffer for opreqs
* hex-opbatch: make opbatches configurable
* hex-naming: better name for ggml_hexagon_shared_buffer
* hex-naming: add session->c_name() helper
* hex-opbatch: start using shm but still copy for now
* hex-opbatch: use shared buffer for packing opbatch
* hex-opbatch: beter naming for opbatch related classes and code
* hex-opbatch: reuse batched tensors with same data/dims/strides
* hex-opbatch: update logging
* hex-opbatch: add support for vmem limit for op batching
* hex-opbatch: update htp side to properly support dynamic mmap/unmap
* hex-opbatch: add OB and OQ params for run-completion script and fix the asserts in batch processing
* hex-opbatch: fixed src1 handling in act ops
* hex-act: fix empty src1 handling in swiglu and friends
Simplify preamble macro while at it
* hex-mm: minor fix vtcm and dma handling in matmul
cleaning up some left-overs from merges
* hex-opbatch: allocate extra 1KB for dspqueue overhead
* hexagon: fix softmax for non-aligned tensors and cleanup vtcm alloc
* hex-mm: properly handle hmx_disabled flag
* hex-ops: update comments
* hex-ops: add debug output for get/set-rows
* hex-mmap: optimize un/mapping of buffers
* hex-opreq: global cache flush and invalidate beyond 128KB threshold
* hex-ops: add super simple opfilter regex for debugging
If an Op matches the regex hex backend will reject it.
* hex-opbatch: wireup newer ops missed in merge and update main switch to detect this in future
* hexagon: improved vtcm acquision to remove inter-op overhead
Fully compatible with QNN-HTP coex
* hex-mm: fixed hvx fallback path
* hex-mm: lower the vmem threshold a bit further to ~3GB
* hexagon: update debug & error logs
This also fixes an issue with newer llvm merging repack and non-repack
functions. We use those pointer to distinguish between buffer types.
* hexagon: move ops context into main context
Just a cleanup. We don't need separate contexts at this point.
* hex-opbatch: cleanup naming and headers for opbatch and related descriptors
* hex-fa: it's now better to enable FA during TG to reduce graph splits
* hexagon: remove GGML_HEXAGON_EXPERIMENTAL env var
It's no longer useful. Please use more flexible GGML_HEXAGON_OPFILTER to disable Ops
if needed for debugging or validation.
* hexagon: fixed editorconfig check
* Update ggml/src/ggml-hexagon/ggml-hexagon.cpp
Co-authored-by: Sigbjørn Skjæret <redacted>
---------
Co-authored-by: Trivikram Reddy <redacted>
Co-authored-by: Sigbjørn Skjæret <redacted>
Aldehir Rojas [Fri, 10 Apr 2026 21:12:53 +0000 (16:12 -0500)]
common : better align to the updated official gemma4 template (#21704)
Adrien Gallouët [Fri, 10 Apr 2026 20:17:00 +0000 (22:17 +0200)]
common : add callback interface for download progress (#21735)
Signed-off-by: Adrien Gallouët <redacted>
MoonRide303 [Fri, 10 Apr 2026 19:45:50 +0000 (21:45 +0200)]
model : make Gemma 4 shared-KV tail attn_k tensors optional on load (#21739)
Rithik Sharma [Fri, 10 Apr 2026 17:52:38 +0000 (10:52 -0700)]
ggml-webgpu: support non-square subgroup matrix configs for Intel GPUs (#21669)
Chen Yuan [Fri, 10 Apr 2026 17:52:01 +0000 (13:52 -0400)]
ggml-webgpu: address quantization precision and backend lifecycle managment (#21521)
* ggml(webgpu): fix the busy-polls in Emscripten in the waitAny after #20618, and remove the busy webgpu log
* Merge with upstream
* Fix GET_ROWS packed integer NaN when using f16 as memory buffer in shader quants
* Update Unary wgsl EXP and EXPM1 for f16 stability
* Fix GET_ROWS IQ4_XS strcut for NaN f16 canonicalization
* Fix numerical percision for unary sqrt when working with f16
* Fix NaN canonicalization for packed integers using f16
* Update err threshold for binary div ops when using f16
* backend: Keep one Dawn/WebGPU instance alive for the lifetime of the static backend
* clean: uncomment existing code logs
* clean: clean the unncessary debug info
* Refactor and generalize dequant helpers
* Remove deprecated quant structs
* Refactor shader defines to reduce repetition
* Remove error override for F16 type
* fix: fix the accidential removal of the proper initialization of ctx
* clean: clean legacy and format code
* fix: did not modify tests ops
---------
Co-authored-by: Jeremy J. Hartmann <redacted>