Advance third_party/sglang to main@3424f35d, including all-target KT MoE adapter serving, the MXFP4 routed-scaling fix, and the GLM-5.3 MTP metadata fix.
* [feat](kt-kernel): add MXFP4 MoE operator with E2M1 weights × BF16 activations
Implements AMX_FP4_MOE_TP based on the RAWINT4 (k2-moe) CRTP pattern.
FP4 E2M1 weights are nibble-packed and decoded via PSHUFB LUT, then
computed with BF16 activations using _mm512_dpbf16_ps. Supports weight-only
per-kgroup scaling (group_size=32) and tensor parallelism.
Includes a Python validation test covering uniform, alternating, ramp,
and random weight patterns.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* [feat](kt-kernel): adapt MXFP4 MoE backend for DeepSeek-V4-Flash (#1950)
V4-Flash routed experts ship as native MXFP4 (E2M1 nibble + ue8m0 group
scale). Expose AMXFP4_KGroup_MOE through NativeMoEWrapper, add a loader
that handles V4's `layers.{L}.ffn.experts.{i}.{w1,w3,w2}.{weight,scale}`
naming and converts ue8m0 → bf16 via a lossless bit-cast, register the
model entry, and ship an end-to-end numerical validation script.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* [perf](kt-kernel): MXFP4 MoE add mat-mat 4×4 tile, refine mat-vec reduce (#1957)
mat_mul_kgroup previously aliased to fp4_mat_vec_kgroup, leaving large
batches stuck on the per-token path. Implement fp4_mat_mat_kgroup as a
4×4 register tile (MB=NB=4, 16 zmm accumulators) so each PSHUFB decode
of four weight rows is reused across four tokens.
Refactor fp4_mat_vec_kgroup to accumulate four N-rows in parallel and
flush them with a new reduce4 helper, removing per-row reduce_add_ps
calls from the hot loop. Mark mxfp4_to_bf16_32 always_inline.
Add bench/bench_fp4_moe.py with --routing {balanced,concentrated} and
a backend registry so future kernels can be added without changing the
runner.
Dispatch thresholds, derived_init, GeneralMOEConfig handling,
load_weights, write_weights_to_buffer and the TP_MOE specialization are
unchanged.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(loader): avoid uint16 lshift in ue8m0->bf16 conversion
PyTorch CPU has no lshift kernel for UInt16, so the previous
`(scale_t.to(torch.uint16) << 7)` raised NotImplementedError when
loading any V4-Flash MXFP4 routed-expert scale tensor on the host.
Switch to int32 for the shift (kernel exists) and narrow to int16
afterwards. The shifted value max is 255<<7 = 32640, well within
int16 range, so the narrow is lossless. The .view(bfloat16) bit
pattern is identical (bf16 sign bit is always 0 for ue8m0 values).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* docs(v4-flash): hybrid CPU/GPU recipe + bump kt-sglang submodule
Bumps third_party/sglang to kvcache-ai/sglang main (3cbd49c29) which now
contains DeepSeek V4 Flash model support + consumer-GPU (SM_120) portable
Triton/TileLang fallbacks (kt-sglang PR #38).
Adds doc/en/DeepSeek-V4-Flash.md tutorial: 8x RTX 5090 hybrid recipe with
the full launch command, OpenAI-compatible /generate + /v1/chat/completions
examples, and the kt chat CLI client.
---------
Co-authored-by: ouqingliang <1692110604@qq.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
* [feat]: simplify sglang installation with submodule, auto-sync CI, and version alignment
- Add kvcache-ai/sglang as git submodule at third_party/sglang (branch = main)
- Add top-level install.sh for one-click source installation (sglang + kt-kernel)
- Add sglang-kt as hard dependency in kt-kernel/pyproject.toml
- Add CI workflow to auto-sync sglang submodule daily and create PR
- Add CI workflow to build and publish sglang-kt to PyPI
- Integrate sglang-kt build into release-pypi.yml (version.py bump publishes both packages)
- Align sglang-kt version with ktransformers via SGLANG_KT_VERSION env var injection
- Update Dockerfile to use submodule and inject aligned version
- Update all 13 doc files, CLI hints, and i18n strings to reference new install methods
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [build]: bump version to 0.5.2
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [build]: rename PyPI package from kt-kernel to ktransformers
Users can now `pip install ktransformers` to get everything
(sglang-kt is auto-installed as a dependency).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Revert "[build]: rename PyPI package from kt-kernel to ktransformers"
This reverts commit e0cbbf6364.
* [build]: add ktransformers meta-package for PyPI
`pip install ktransformers` now works as a single install command.
It pulls kt-kernel (which in turn pulls sglang-kt).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [fix]: show sglang-kt package version in kt version command
- Prioritize sglang-kt package version (aligned with ktransformers)
over sglang internal __version__
- Update display name from "sglang" to "sglang-kt"
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [fix]: improve sglang-kt detection in kt doctor and kt version
Recognize sglang-kt package name as proof of kvcache-ai fork installation.
Previously both commands fell through to "PyPI (not recommended)" for
non-editable local source installs. Now version.py reuses the centralized
check_sglang_installation() logic.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [build]: bump version to 0.5.2.post1
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: move legacy code to archive/ directory
- Moved ktransformers, csrc, third_party, merge_tensors to archive/
- Moved build scripts and configurations to archive/
- Kept kt-kernel, KT-SFT, doc, and README files in root
- Preserved complete git history for all moved files
* refactor: restructure repository to focus on kt-kernel and KT-SFT modules
* fix README
* fix README
* fix README
* fix README
* docs: add performance benchmarks to kt-kernel section
Add comprehensive performance data for kt-kernel to match KT-SFT's presentation:
- AMX kernel optimization: 21.3 TFLOPS (3.9× faster than PyTorch)
- Prefill phase: up to 20× speedup vs baseline
- Decode phase: up to 4× speedup
- NUMA optimization: up to 63% throughput improvement
- Multi-GPU (8×L20): 227.85 tokens/s total throughput with DeepSeek-R1 FP8
Source: https://lmsys.org/blog/2025-10-22-KTransformers/
This provides users with concrete performance metrics for both core modules,
making it easier to understand the capabilities of each component.
* refactor: improve kt-kernel performance data with specific hardware and models
Replace generic performance descriptions with concrete benchmarks:
- Specify exact hardware: 8×L20 GPU + Xeon Gold 6454S, Single/Dual-socket Xeon + AMX
- Include specific models: DeepSeek-R1-0528 (FP8), DeepSeek-V3 (671B)
- Show detailed metrics: total throughput, output throughput, concurrency details
- Match KT-SFT presentation style for consistency
This provides users with actionable performance data they can use to evaluate
hardware requirements and expected performance for their use cases.
* fix README
* docs: clean up performance table and improve formatting
* add pic for README
* refactor: simplify .gitmodules and backup legacy submodules
- Remove 7 legacy submodules from root .gitmodules (archive/third_party/*)
- Keep only 2 active submodules for kt-kernel (llama.cpp, pybind11)
- Backup complete .gitmodules to archive/.gitmodules
- Add documentation in archive/README.md for researchers who need legacy submodules
This reduces initial clone size by ~500MB and avoids downloading unused dependencies.
* refactor: move doc/ back to root directory
Keep documentation in root for easier access and maintenance.
* refactor: consolidate all images to doc/assets/
- Move kt-kernel/assets/heterogeneous_computing.png to doc/assets/
- Remove KT-SFT/assets/ (images already in doc/assets/)
- Update KT-SFT/README.md image references to ../doc/assets/
- Eliminates ~7.9MB image duplication
- Centralizes all documentation assets in one location
* fix pic path for README
github-actions[bot]