Merge upstream PR #25731: TML Inkling architecture (+ upstream master sync) - #33
Merged
Conversation
…LU/NORM (ggml-org#24582) * vulkan: make SQR/SQRT/SIN/COS/CLAMP/LEAKY_RELU use unary.comp * vulkan: make NORM support noncontig * add noncontiguous row test cases for norm/l2_norm, handle this in the CPU backend and l2_norm.comp * fix supports_op for cuda and webgpu
…4913) * model : Add LFM2.5-ColBERT-350M and LFM2.5-Embedding-350M * Restore LFM2 models in README.md
…rg#24897) * chore: `npm audit fix --force` * feat: Update sidebar toggle to use Logo * refactor: Clean up favicon SVG * feat: Refactor logo component and implement theme-aware favicon generation * feat: Add configurable padding to generated PWA assets * test: Add unit tests for writeThemeFavicons * refactor: Componentization * feat: WIP * feat: WIP * feat: WIP * feat: Mobile UI * feat: add SEARCH route constant * feat: create SidebarNavigationSearchResults component * refactor: use SidebarNavigationSearchResults in conversation list * feat: enable mobile search navigation in sidebar actions * feat: add mobile search route and page * fix: prevent sidebar overflow on mobile viewports * fix: Mobile sidebar * feat: Mobile Search WIP * feat: Mobile WIP * feat: Add PWA standalone detection and refine mobile UI * feat: Improve mobile layout, sidebar handling, and chat scrolling * feat: Improve mobile sidebar visibility and iOS Safari chat spacing * fix: Disable auto-scroll on mobile * chore: Linting * fix: Wrong condition * feat: Mobile chat scroll * refactor: WIP * fix: Desktop initial scroll always working again * fix: Partial fix for mobile auto-scroll / initial scroll * fix: Desktop auto-scroll on initial load and during streaming * fix: Mobile scrolling logic * refactor: Clean up * feat: Improve start UI * feat: Add `delay` to `fadeInView` * feat: Auto-scroll button * refactor: Cleanup * refactor: Extract chat dialogs and alerts into dedicated component * refactor: Reorganize ChatScreen component structure and initialization * feat: Improve auto-scroll after sending message * feat: UI improvements * fix: Settings link * feat: UI improvements * fix: better UI spacing * fix: Remove unneeded logic * fix: Chat Processing Info UI rendering * feat: Improve mobile UI * feat: UI improvement * fix: Conditional transition delay for Chat Messages based on route from * fix: Delay mobile sidebar collapse for smoother transitions * fix: Mobile scroll down button + sidebar pointer events * fix: Mobile UI * fix: Auto scrolling * fix: Implement dynamic height calculations for chat auto-scroll positioning and UI elements * fix: Retrieve `autofocus` for Chat Form textarea * fix: Use proper class Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> * refactor: extract scroll-to-bottom logic and fix message send flow * fix: update viewport store usage and remove conflicting autofocus * feat: add accessibility labels to scroll down button * fix: correct HTML structure in sidebar empty states * fix: dynamically toggle processing info visibility * chore: remove commented exports and fix formatting * fix * fix: Mobile Chat Form Add Action Sheet interactions --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* ui: show model load progress on the selector trigger Mirror the in-dropdown stage progress as a thin bar on the selector trigger, so the active model's load percent stays visible when the menu is closed. Same status gating and composite fraction as the dropdown row, so both bars track the selected model in sync. Suggested-by: Julien Chaumond <@julien-c> * ui: show model load progress bar on the in-conversation model selector * ui: tune model load indicator to a pulsing highlight (suggested by @ngxson) Also wire the indicator onto the mobile sheet trigger, which was missing it since mobile uses the sheet instead of the dropdown. * ui: thin (@allozaur) pulsating (@ngxson) model load bar
* vulkan-shaders-gen: fail the build when a shader fails to compile vulkan-shaders-gen did not detect shader-compile subprocess failures, so a broken libggml-vulkan could be produced while the build reported success and the breakage only surfaced at run time. execute_command() discarded the child exit code (POSIX waitpid passed nullptr for status; the Windows branch never called GetExitCodeProcess) and string_to_spv decided success only from whether stderr was empty, so a non-zero exit with empty stderr, or a subprocess that failed to launch, was treated as success. Return the child exit code from execute_command() (WEXITSTATUS on POSIX, GetExitCodeProcess on Windows), treat a non-zero exit or non-empty stderr or a launch exception as a failure, and record it in an atomic flag. main() checks the flag after process_shaders() and returns EXIT_FAILURE before writing the output files, so the build stops instead of emitting a broken backend. Fixes ggml-org#24393 Signed-off-by: liminfei-amd <91481003+liminfei-amd@users.noreply.github.com> * vulkan-shaders-gen: simplify compile_failed access and drop unreachable return Address review feedback on ggml-org#24450: - Access the std::atomic<bool> compile_failed directly (= / implicit bool) instead of .store()/.load(); the flag stays atomic because the worker threads in process_shaders() set it concurrently. - Remove the unreachable trailing return -1 in execute_command(): on POSIX the child _exit()s after execvp and the parent returns (fork()<0 throws); on Windows the block returns the exit code. Signed-off-by: liminfei-amd <91481003+liminfei-amd@users.noreply.github.com> --------- Signed-off-by: liminfei-amd <91481003+liminfei-amd@users.noreply.github.com>
…ernel-params, cached graphs (ggml-org#24954) * hex-mm: new weight layout and fusion updates * hvx-mm: unroll the new tiled vec_dots to optimize hvx register util * hex-mm: optimize dyn.quant format for q8_0 and q8_1 to reduce overhead in vec_dots. * hvx-mm: parallel quantizer per block for large rows * hvx-mm: simplify and futher optimize dyn.quant and vec_dots * hvx-mm: keep intermediate per tile accumulators in fp16 * hmx-mm: optimize weight dequant by aligning the repacked tiles with the DMA * hmx-mm: remove qweight scratch and just use vtcm_weight * hmx-mm: remove all unused and obsolete code * hmx-mm: the new tiled repack format is here to stay -- rename all x4x2 to _tiled * hmx-mm: improve activation processing with dma prefetch * hex-mm: fix hmx/hvx fallback logic and MUL_MAT_ID allocation (unbreaks OLMoE) * hex-mm: align the weight tiles with dma just like we did in hmx-mm * hex-mm: factor out common mm bits into htp/matmul-ops.h * hex-mm: start moving mm kernel selection to the host * hex-mm: move all of the matmul param compute into the host * hmx-mm: restore pipelined mode * hmx-mm: unroll the dequant functions to optimize register usage * hmx-mm: further improve activation process * hex-mm: use vtcm_seq_alloc for all vtcm allocations and define more common functions * hex-mm: improve mm optimizer to acount for number of activation threads * hex-mm: fix matmul-id kernel params selection (unbreaks OLMoE and LFM) * hexagon: remove support for arch < v73 since HMX is now required for most use-cases * hex-mm: cleanup naming for consistency * hex-mm: make sure matmul fusion accounts for vtcm allocation * hex-mm: minor cleanup for kernel_params definition * hex-mm: replace hardcoded limits with proper checks for vtcm requirements * hex-mm: add support for non-tiled mm as a fallback option and factor out hvx kernels into separate header * hex-mm: remove unused functions * hex-mm: add shorthand for MM_SELECT in run-tool script * hvx-mm: factor out hvx/hmx microkernels and unify matmul entry and dispatch * hex-mm: further cleanup matmul fallback path * hex-mm: refactor matmul entry point and dispatch a bit further * hexagon: update cmake build to enable hmx for everything * hex-ops: optimize kernel_param updates and include summary in the logs * hex-mm: add support for GGML_HEXAGON_MM_SELECT * hex-mm: add hex-common header * hex-mm: pass correct number of tasks to workpool * hex-mm: add proper checks for no-work in dyn.quant tasks * hex-mm: convert all quantizers into a macro * hex-mm: fix hvx-flat fallback to pass all MUL_MAT tests * hex-mm: vectorize q8_1 quantizer * hex-mm: improve fused ffn mm stride handling * hex-mm: consistent use of n_threads and pipeline in kernel_params * hexagon: minor formatting * hex-mm: update MUL_MAT_ID kernel_param handling to make sure host/npu are in sync * hvx-mm: go back to accumulating in fp32 in tiled hvx kernels, more accurate and same perf * hvx-mm: unroll the loops and remove masking that is not needed for tiled accums * hmx-mm: optimize activation processing (slit loops, some unrolling, etc) * hmx-mm: minor optimization for output processing * hex-mm: consistent use of uint32_t and size_t in mm kernels * hex-mm: remove legacy restrictions for rows to be multiple of 256 * hexagon: replace sprintf with snprintf * hex-mm: relax hardcoded nrows checks and rely on VTCM size requirements * hexagon: minor alignment fix * hexagon: fix trailing spaces * hex-mm: relax padding from 256 to 128 (leftovers) * hex-mm: remove redundant checks for weight align to 128 we always use 2D dma for the weights and align them properly * hmx-mm: MUL_MAT_ID better work distribution between hvx threads and hmx tracing * hex-mm: specialize per-token mmid activation handling * hex-profile: update python scripts to handle kernel-params section in the logging output * hex-mm: move n_prefetch (aka dma_depth) into kernel params and remove unused fields * hex-trace: use easier to parse format, simply and fix post-proc scripts * hmx-mm: relax 32 row limit for output processing which helps utilization * hmx-mm: use start-chunk idx for tracing info * hmx-mm: parameterize activation dma pipeline * hexagon: add support for simple graph caching to avoid recomputing kernel-params * hex-mm: remove left-over repack functions * hex-mm: tighten n_prefetch asserts * hex-mm: remove duplicate round/align_up helper * hexagon: cleanup common header used in host/npu * hexagon: update early wakeup threshold * hmx-mm: define cost constants and update solver to assume that repacked ne[1] is padded to 32 * hmx-mm: make precompute_matmul a bit more readable (split into smaller functions, etc) * hex-mm: remove n_threads constraint * hex-mm: minor formatting updates * hex-mm: remove obsolete profiling logs * hex-mm: restore hardcode gate to refuse lm-head to avoid repacking that tensor
* Sycl tp stage1 (#1) * SYCL: tensor parallelism (--split-mode tensor) for dual-GPU Adds the comm_init/comm_free/comm_allreduce_tensor trio that the meta-backend queries via get_proc_address to enable backend-specific all-reduce, mirroring the pattern used by ggml-cuda.cu. For N=2 (the common dual-GPU case) implements a degenerate ring all-reduce with two size-branched paths: * Small (nelem < 32768): FP32 direct memcpy + per-device ADD kernel chained via depends_on(memcpy_event). 4 SYCL submissions/call. * Large (nelem >= 32768): BF16-compressed. Each device compresses FP32 -> BF16 in a local outbox, cross-device memcpys to the peer's inbox (HALF the PCIe bytes), then decompresses + adds into the local FP32 partial. 6 SYCL submissions/call but PCIe bytes halved -- wins for any tensor where PCIe dominates kernel time. Threshold and BF16 path pattern mirror the CUDA NCCL allreduce. Storage: ONE persistent uint8_t buffer per device, 4 * nelem bytes (matches both path layouts: FP32 nelem floats; BF16 outbox+inbox = 2 * nelem uint16_t each). Single alloc+free per device keeps the SYCL pool's strict-LIFO invariant trivial. Initial impl handles N=2 FP32 contiguous tensors. Other cases return false, causing the meta-backend to use its generic butterfly fallback. Per-call sync is intentionally omitted. SYCL in-order queue semantics ensure that the meta-backend's next compute on the same per-device queue waits for our final ADD, and the next allreduce's first op on the same persistent buffer waits via the same queue. Only comm_free does an explicit final wait. OneCCL is NOT used: OneCCL 2021.17 hardcodes single-device-per-process in communicator_impl.hpp:47 (condition devices.size() == 1), which is incompatible with llama.cpp's single-process multi-GPU model. Measured on dual Intel Arc Pro B70 (NEO 26.05.x, oneAPI 2025.3 + DPC++ nightly): Llama-3.3-70B Q4_K_M, -sm tensor -fa 1 -ctk f16 -ctv f16: pp512 = 377.08 t/s (vs 313.65 layer mode = +20.2%) tg128 = 17.40 t/s (vs 9.74 layer mode = +78.6%) Qwen3-Coder-Next-80B-A3B Q3_K_M (MoE): pp512 = 216.56 t/s (vs 156.58 meta-backend butterfly = +38.3%) tg128 = 17.60 t/s (vs 14.31 meta-backend butterfly = +23.0%) Qwen3-4B Q4_K_M: pp64 = 984.51 t/s, tg16 = 49.29 t/s Llama-3.3-70B in SYCL TP now comfortably beats production layer mode on both prefill and decode. Coder-Next-80B-A3B (MoE) also wins on both — the BF16 path is what unlocks the many-medium-allreduces prefill pattern. Build/CMake: no changes. No new dependencies. ~210 lines added across ggml-sycl.h and ggml-sycl.cpp. * Fix comments * documentation update to address PR feedback * Bring over my device-to-device memcpy chagnes * move the dev2dev_memcpy calls to the upstream 7-parameter variety * Fix a typo and remove a trailing whitespace
Adds an opt-in LLAMA_BUILD_MTMD CMake option so build-xcframework.sh can link libmtmd.a into the framework binary without pulling in the rest of tools/ (which doesn't cross-build cleanly to iOS/tvOS/visionOS). - CMakeLists.txt: new option, default OFF. When on with LLAMA_BUILD_TOOLS=OFF, only the tools/mtmd subdir is added. Useful for any binding that wants just libmtmd (Apple XCFramework, WASM). - tools/mtmd/CMakeLists.txt: gate the CLI exe targets on LLAMA_BUILD_TOOLS. Gating on LLAMA_BUILD_COMMON is not enough — it defaults ON in standalone builds and visionOS xcodebuild then fails with "install TARGETS given no BUNDLE DESTINATION for MACOSX_BUNDLE executable target 'llama-mtmd-cli'". - build-xcframework.sh: turn the option on, pass -DLLAMA_BUILD_MTMD, add libmtmd.a to combine_static_libraries, and copy mtmd.h and mtmd-helper.h into the framework Headers dir. The umbrella module map then exposes them, so Swift / Obj-C consumers can import the mtmd C API directly. After this, nm on ios-arm64/llama.framework/llama shows 52 _mtmd_ symbols. Verified end-to-end: a Swift target links the produced framework and calls mtmd_default_marker, mtmd_bitmap_init, etc. without a shim on macos / iphoneos / iphonesimulator / xros slices. Co-authored-by: Abraham Gonzalez <abraham@theabecaster.com>
…4935) Assisted-by: Claude
…gml-org#24706) * ggml : address integer overflows in binary ops CUDA implementation * ggml : add size_t casts to avoid integer overflows * ggml : add more asserts checking integer overflows in binary ops CUDA implementation --------- Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
* app : add the download command (with llama-download) Signed-off-by: Adrien Gallouët <angt@huggingface.co> * Remove llama-download tool for now Signed-off-by: Adrien Gallouët <angt@huggingface.co> --------- Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* eagle3: accept Eagle3LlamaForCausalLM draft checkpoints * docs: add eagle3 speculative decoding section * docs: address eagle3 review comments * docs: add more angelslim eagle3 models * docs: add gpt-oss eagle3 models and link to pr 18039
* common: refactor models handling * remote preset * cont * rm skip_download option * missing header * fix plan.model_files * fix --offline case * move hf_plan to download * refactor * rm redundant curr_ex, add comments * adapt
* misc: update lables * bring back examples, add mtmd
* server: use status code 403 for disabled features * cont * fix test case
* Add failing test-case to test-backend-ops Extracted from ggml-org#24072 * Minimize repro with help of AI N = 8 * (65535 - 1) + 1 = 524273 * Port and adjust workaround from LostRuins@0ba7983 Fall-back should share code, also relax y-z constraint to be inclusive * Add test-case + fallback also for y dim * Fix x-guards which is 2^{31}-1, so inlusive of INT_MAX * Fix overflow problems for transposed copy kernel
* [SYCL] F16 (default) Flash Attention with XMX engine via oneDNN graph API; Qwen3.6-27b-Q8_0 prefill speed up x1.21 at p=512 and x4.26 at p=80k * [SYCL] Address review on FA oneDNN path. Result: llama-bench---pp512; 32% increase with fa1; llama-perplexity---0.11% difference; tested model: mradermacher/Meta-Llama-3.1-8B-Instruct-Q8_0.gguf * PR-25222 revision v2: addressed audits * [SYCL] flash-attn oneDNN SDPA KV F16 rev 3.0: add BMG gate + multi-device sync. Narrow the scrope of this PR to Battlemage only (bmg; Xe2). Other archs (e.g., alchemist) fall back to existing FA kernel. When device_count >1, apply stream -> wait_and_throw(), validated working path for multi-gpu sync fix by @maxious. Co-authored-by: maxious <81432+maxious@users.noreply.github.com> * updated comment on bmg gate, noted the issue --------- Co-authored-by: scientist3 <scientist.3@users.noreply.github.com> Co-authored-by: hmscider <hmscider@users.noreply.github.com> Co-authored-by: maxious <81432+maxious@users.noreply.github.com>
…rg#25525) Raise the threshold for minimum buffer size from 1 GiB to 4 GiB, based on real-world experiments of overcommitting device memory with model weights larger than available VRAM, for example Qwen3.5-35B-A3B-Q8 running on a B70. Also add a debug message to better track USM system allocations. Signed-off-by: Francois Dugast <francois.dugast@intel.com>
Hybrid attention model: 55 sliding-window plus 11 global layers, banded content-dependent relative position bias instead of RoPE, per-layer short convolution state, fine-grained MoE (256 experts top-6 plus 2 shared), attention log-scaling past 128K, 1M context. Includes the GGML_OP_FLASH_ATTN_EXT_BANDED operator (CPU and CUDA, fused into the MMA flash attention kernel with an fp16 accumulator overflow guard), HF to GGUF conversion, chat template with typed content block parsing (interleaved thinking, narration and tool calls), mmproj vision and audio support, and backend op tests at production shapes.
# Conflicts: # docs/speculative.md # ggml/include/ggml-rpc.h # ggml/include/ggml.h # ggml/src/ggml-cpu/ggml-cpu.c # ggml/src/ggml-cpu/ops.h # ggml/src/ggml-cuda/fattn.cu # ggml/src/ggml-cuda/ggml-cuda.cu # ggml/src/ggml-metal/ggml-metal.metal # ggml/src/ggml-vulkan/ggml-vulkan.cpp # ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp # ggml/src/ggml.c # gguf-py/gguf/constants.py # include/llama.h # src/llama-kv-cache.cpp # src/llama-model-loader.cpp # src/llama-vocab.h # tests/test-backend-ops.cpp # tests/test-quantize-fns.cpp # tools/server/server-context.cpp # tools/server/server-models.cpp # tools/server/server-models.h # tools/server/server.cpp
github-actions
Bot
added
documentation
…merge The definitions of ggml_cuda_host_staging_pool and ggml_cuda_copy_across_devices lived inside the old ggml_cuda_op_mul_mat split infrastructure block that upstream deleted; the merge kept the forward declarations and call sites but dropped the bodies. Restore the pool and the 1D copy helper, and drop the now-unreferenced 2D variant (its only caller was the deleted split mul_mat path).
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.
Merges the upstream draft PR ggml-org#25731 ("Add TML Inkling architecture") into the fork, together with the upstream master history it was based on (~388 commits). Full per-file conflict log lives in MERGE_NOTES.md.
What comes in
inklingarchitecture: 975B MoE (256 experts top-6 + 2 shared), 66 layers, hybrid 55 SWA / 11 global attention, banded content-dependent relative position bias (no RoPE), per-layer short conv state, attention log-scaling past 128K, 1M contextGGML_OP_FLASH_ATTN_EXT_BANDED(CPU + CUDA, fused MMA with fp16-accumulator overflow guard) andGGML_OP_LIGHTNING_INDEXERGGML_TYPE_Q2_0, HF→GGUF conversion, Inkling chat template (typed content blocks), mmproj vision/audio, backend op tests at production shapesKey merge decisions
src/llama-context.cpp): turbo2/3/4 KV cache is not implemented for the banded-attention path, so forLLM_ARCH_INKLINGthe context warns and falls back to the standard f16 KV cache. All existing architectures keep full turbo cache, TQ3_1S/TQ4_1S weights, and MTP/NextN.ggml_cuda_mul_mat(old split-buffer multi-GPU infra deleted upstream —-sm rowis gone); TQ fused-dp4a / fWHT-hint / TQ4_1S-cuBLAS hooks ported into the new skeleton. Kept our non-power-of-2 GQA dispatch fix infattn.cu./v1/models, cache-key slot affinity, mmproj-draft mirroring).Verification
test-quantize-fns(tq3_1s, tq4_1s, q2_0),test-quant-type-selection(snapshots regenerated for upstream's intentionaln_layer_allNextN change)FLASH_ATTN_EXT_BANDEDat production shapes; numeric banded tests need a GPU box (test harness compares backends against CPU)test-llama-archslaguna-dense failure is pre-existing (loader/fixture identical pre-merge), not a regressionBefore release (see MERGE_NOTES.md → "Still required")
test-backend-ops -o FLASH_ATTN_EXT_BANDED+ full suite on a GPU machine-sm rowmulti-GPU split support