freetoken 0.1.1__tar.gz

freetoken-0.1.1/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 FreeToken Authors
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

freetoken-0.1.1/PKG-INFO

@@ -0,0 +1,88 @@
+Metadata-Version: 2.4
+Name: freetoken
+Version: 0.1.1
+Summary: FreeToken inference runtime
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/FlashML-org/FreeToken
+Project-URL: Repository, https://github.com/FlashML-org/FreeToken
+Project-URL: Issues, https://github.com/FlashML-org/FreeToken/issues
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: apache-tvm-ffi==0.1.13.post3
+Requires-Dist: einops<1,>=0.8
+Requires-Dist: fastapi<1,>=0.115
+Requires-Dist: flashlib==0.3.0
+Requires-Dist: gguf<1,>=0.19
+Requires-Dist: huggingface_hub<2,>=1.5
+Requires-Dist: msgpack<2,>=1.1
+Requires-Dist: modelscope<2,>=1.37
+Requires-Dist: numpy<2.5,>=2.0
+Requires-Dist: openai<3,>=2.0
+Requires-Dist: partial-json-parser<1,>=0.2
+Requires-Dist: prompt_toolkit<4,>=3.0
+Requires-Dist: pydantic<3,>=2.9
+Requires-Dist: pyzmq<28,>=27
+Requires-Dist: safetensors<1,>=0.6
+Requires-Dist: torch<2.12,>=2.11
+Requires-Dist: tqdm<5,>=4.66
+Requires-Dist: transformers<6,>=5.5
+Requires-Dist: triton==3.6.0
+Requires-Dist: uvicorn<1,>=0.30
+Provides-Extra: dev
+Requires-Dist: pytest>=6.0; extra == "dev"
+Provides-Extra: fi
+Requires-Dist: flashinfer-python[cu13]<0.7,>=0.6; extra == "fi"
+Provides-Extra: sgl
+Requires-Dist: sglang-kernel==0.4.5; extra == "sgl"
+Provides-Extra: accel
+Requires-Dist: freetoken[fi,sgl]; extra == "accel"
+Dynamic: license-file
+
+<div align="center">
+  <picture>
+    <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/FlashML-org/FreeToken/main/assets/freetoken-logo-dark.svg">
+    <source media="(prefers-color-scheme: light)" srcset="https://raw.githubusercontent.com/FlashML-org/FreeToken/main/assets/freetoken-logo-light.svg">
+    <img alt="FreeToken" src="https://raw.githubusercontent.com/FlashML-org/FreeToken/main/assets/freetoken-logo.svg" width=55%>
+  </picture>
+  
+  [![Slack](https://img.shields.io/badge/Slack-join%20the%20community-4A154B?logo=slack&logoColor=white)](https://join.slack.com/t/flashml/shared_invite/zt-3zpdh5j10-9dwTXrgLiqpVxizhA9KVbA)
+[![Discord](https://img.shields.io/badge/Discord-join%20the%20community-5865F2?logo=discord&logoColor=white)](https://discord.gg/xzwSnMdsX)
+</div>
+
+
+
+A local, MoE-offload inference runtime with an OpenAI- and Anthropic-compatible
+HTTP API — Run DeepSeek-V4-Flash on your 5090 with 20+ TPS.
+
+
+
+## Quick start
+
+See [docs/install.md](https://github.com/FlashML-org/FreeToken/blob/main/docs/install.md) for requirements and installation.
+
+```bash
+ft serve --model ~/models/Qwen3.6-35B-A3B   # API server on http://127.0.0.1:1919
+ft launch claude                            # point an agent at it (codex / opencode / openclaw)
+ft shell                                    # or chat in the terminal
+```
+
+## Documentation
+
+- [Install](https://github.com/FlashML-org/FreeToken/blob/main/docs/install.md) — requirements and setup
+- [Supported models](https://github.com/FlashML-org/FreeToken/blob/main/docs/models.md) — model × quantization
+- [CLI reference](https://github.com/FlashML-org/FreeToken/blob/main/docs/cli.md) — `ft` commands and environment variables
+
+## Acknowledgment
+
+FreeToken was deeply inspired by [mini-sglang](https://github.com/sgl-project/mini-sglang), and
+learned the design and reused code from the following projects:
+[SGLang](https://github.com/sgl-project/sglang),
+[vLLM](https://github.com/vllm-project/vllm),
+[FlashInfer](https://github.com/flashinfer-ai/flashinfer),
+[flash-linear-attention](https://github.com/fla-org/flash-linear-attention),
+[LightLLM](https://github.com/ModelTC/lightllm) and [llama.cpp](https://github.com/ggml-org/llama.cpp).
+
+## License
+
+[Apache License 2.0](https://github.com/FlashML-org/FreeToken/blob/main/LICENSE).

freetoken-0.1.1/README.md

@@ -0,0 +1,47 @@
+<div align="center">
+  <picture>
+    <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/FlashML-org/FreeToken/main/assets/freetoken-logo-dark.svg">
+    <source media="(prefers-color-scheme: light)" srcset="https://raw.githubusercontent.com/FlashML-org/FreeToken/main/assets/freetoken-logo-light.svg">
+    <img alt="FreeToken" src="https://raw.githubusercontent.com/FlashML-org/FreeToken/main/assets/freetoken-logo.svg" width=55%>
+  </picture>
+  
+  [![Slack](https://img.shields.io/badge/Slack-join%20the%20community-4A154B?logo=slack&logoColor=white)](https://join.slack.com/t/flashml/shared_invite/zt-3zpdh5j10-9dwTXrgLiqpVxizhA9KVbA)
+[![Discord](https://img.shields.io/badge/Discord-join%20the%20community-5865F2?logo=discord&logoColor=white)](https://discord.gg/xzwSnMdsX)
+</div>
+
+
+
+A local, MoE-offload inference runtime with an OpenAI- and Anthropic-compatible
+HTTP API — Run DeepSeek-V4-Flash on your 5090 with 20+ TPS.
+
+
+
+## Quick start
+
+See [docs/install.md](https://github.com/FlashML-org/FreeToken/blob/main/docs/install.md) for requirements and installation.
+
+```bash
+ft serve --model ~/models/Qwen3.6-35B-A3B   # API server on http://127.0.0.1:1919
+ft launch claude                            # point an agent at it (codex / opencode / openclaw)
+ft shell                                    # or chat in the terminal
+```
+
+## Documentation
+
+- [Install](https://github.com/FlashML-org/FreeToken/blob/main/docs/install.md) — requirements and setup
+- [Supported models](https://github.com/FlashML-org/FreeToken/blob/main/docs/models.md) — model × quantization
+- [CLI reference](https://github.com/FlashML-org/FreeToken/blob/main/docs/cli.md) — `ft` commands and environment variables
+
+## Acknowledgment
+
+FreeToken was deeply inspired by [mini-sglang](https://github.com/sgl-project/mini-sglang), and
+learned the design and reused code from the following projects:
+[SGLang](https://github.com/sgl-project/sglang),
+[vLLM](https://github.com/vllm-project/vllm),
+[FlashInfer](https://github.com/flashinfer-ai/flashinfer),
+[flash-linear-attention](https://github.com/fla-org/flash-linear-attention),
+[LightLLM](https://github.com/ModelTC/lightllm) and [llama.cpp](https://github.com/ggml-org/llama.cpp).
+
+## License
+
+[Apache License 2.0](https://github.com/FlashML-org/FreeToken/blob/main/LICENSE).

freetoken-0.1.1/pyproject.toml

@@ -0,0 +1,110 @@
+[build-system]
+# torch must match the runtime range below: build isolation resolves this list on its
+# own, and a mismatch links the C++ extensions against the wrong libtorch.
+# setuptools floor: 77 is the first release that understands the PEP 639 `license`
+# SPDX string and `license-files` below.
+requires = ["setuptools>=77", "torch>=2.11,<2.12", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "freetoken"
+# Single source of truth: python/freetoken/version.py (read via the attr directive below).
+dynamic = ["version"]
+description = "FreeToken inference runtime"
+readme = "README.md"
+requires-python = ">=3.10"
+license = "Apache-2.0"
+license-files = ["LICENSE"]
+# Version ranges are floor=last-verified, ceiling=next-major (kept loose enough to
+# resolve, tight enough to bar a silent breaking bump). These ranges are the contract;
+# there is no lockfile.
+dependencies = [
+    "apache-tvm-ffi==0.1.13.post3", # pin it
+    "einops>=0.8,<1",
+    "fastapi>=0.115,<1",
+    # slot_cache: the device-side LRU admission kernel behind the MoE expert cache.
+    "flashlib==0.3.0", # pin it
+    "gguf>=0.19,<1",
+    "huggingface_hub>=1.5,<2",
+    "msgpack>=1.1,<2",
+    "modelscope>=1.37,<2",
+    # ceiling: numba (via flashlib) needs numpy<2.5
+    "numpy>=2.0,<2.5",
+    "openai>=2.0,<3",
+    "partial-json-parser>=0.2,<1",
+    "prompt_toolkit>=3.0,<4",
+    "pydantic>=2.9,<3",
+    "pyzmq>=27,<28",
+    "safetensors>=0.6,<1",
+    # floor+ceiling: sglang-kernel 0.4.5 links libtorch symbols only 2.11 has.
+    # PyPI's torch 2.11.0 wheel is itself the cu130 build, so plain pip resolves
+    # correctly from PyPI alone; uv additionally pins the index below.
+    "torch>=2.11,<2.12",
+    "tqdm>=4.66,<5",
+    "transformers>=5.5,<6",
+    "triton==3.6.0",
+    "uvicorn>=0.30,<1",
+]
+
+[project.urls]
+Homepage = "https://github.com/FlashML-org/FreeToken"
+Repository = "https://github.com/FlashML-org/FreeToken"
+Issues = "https://github.com/FlashML-org/FreeToken/issues"
+
+[project.scripts]
+ft = "freetoken.cli:main"
+
+[project.optional-dependencies]
+dev = ["pytest>=6.0"]
+# Native fused-kernel packages. Without them the runtime falls back to the pure-Triton
+# kernels in freetoken.kernel.triton (and the 'triton' attention backend). Install
+# `freetoken[accel]` for the full native fast path.
+fi = ["flashinfer-python[cu13]>=0.6,<0.7"]
+# renamed from sgl-kernel at 0.4; still imports as `sgl_kernel`, so never co-install both
+sgl = ["sglang-kernel==0.4.5"]
+accel = ["freetoken[fi,sgl]"]
+# NOTE: the Marlin W4A16 NVFP4 expert-GEMM path (sm_80-99) borrows vLLM's AOT wheel
+# (vllm>=0.14,<0.15), which pins transformers>=4.56,<5 and so is INCOMPATIBLE with the
+# core transformers>=5.5 requirement. It is therefore not a lockable extra and is left
+# out of the default resolution — install it separately (`pip install 'vllm>=0.14,<0.15'`)
+# in a dedicated environment when you specifically need the Marlin path.
+
+# uv-only provenance pins — pip ignores this section and resolves from PyPI, which
+# works: PyPI's torch 2.11.0 and sglang-kernel 0.4.5 are the same cu130 builds these
+# indexes serve. `explicit = true` scopes each index to the one package that needs it
+# (the torch index also mirrors stale copies of common deps, e.g. packaging<=24.1,
+# which would otherwise shadow PyPI under uv's first-index strategy).
+[tool.uv.sources]
+torch = { index = "pytorch-cu130" }
+sglang-kernel = { index = "sglang-cu130" }
+
+[[tool.uv.index]]
+name = "pytorch-cu130"
+url = "https://download.pytorch.org/whl/cu130"
+explicit = true
+
+[[tool.uv.index]]
+name = "sglang-cu130"
+url = "https://docs.sglang.io/whl/cu130"
+explicit = true
+
+[tool.setuptools]
+include-package-data = true
+package-dir = {"" = "python"}
+
+[tool.setuptools.dynamic]
+version = {attr = "freetoken.version.__version__"}
+
+[tool.setuptools.packages.find]
+where = ["python"]
+
+[tool.setuptools.package-data]
+"*" = ["csrc/**/*", "moe/configs/**/*.json"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+python_files = ["test_*.py"]
+markers = [
+    "slow: takes tens of seconds (big kernel sweeps, real-checkpoint reads); deselect with -m 'not slow'",
+    "needs_weights: needs a real local checkpoint, gated behind an env var (see tests/README.md)",
+]

freetoken-0.1.1/python/freetoken/__init__.py

@@ -0,0 +1,5 @@
+"""FreeToken inference runtime."""
+
+from freetoken.version import __version__
+
+__all__ = ["__version__"]

freetoken-0.1.1/python/freetoken/__main__.py

@@ -0,0 +1,5 @@
+from .server import launch_server
+
+assert __name__ == "__main__"
+
+launch_server(prog="python -m freetoken")

freetoken-0.1.1/python/freetoken/attention/__init__.py

@@ -0,0 +1,183 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import TYPE_CHECKING, Protocol
+
+from freetoken.utils import Registry, init_logger
+
+from .base import AttentionSpec, AttnType, BaseAttnBackend, BaseAttnMetadata, HybridBackend
+
+if TYPE_CHECKING:
+    from freetoken.models import ModelConfig
+
+logger = init_logger(__name__)
+
+
+class BackendCreator(Protocol):
+    def __call__(self, config: ModelConfig) -> BaseAttnBackend: ...
+
+
+@dataclass(frozen=True)
+class BackendInfo:
+    """Declarative capability matrix entry for one backend. The engine's config-time
+    validation interprets the requirement flags through its own (monkeypatch-able)
+    availability probes; this stays pure data so registration never imports kernels."""
+
+    supported_types: frozenset[AttnType]
+    requires_flashinfer: bool = False
+    requires_sgl_kernel: bool = False
+    requires_sm100: bool = False
+    # Allowed page sizes (None -> any). Config-time resolution coerces to the last
+    # entry when the resolved page_size is not in the list.
+    page_sizes: tuple[int, ...] | None = None
+    # Whether forward() honors a per-call AttentionSpec (window/sm_scale/sinks).
+    # Non-consumers raise on a non-None spec instead of silently dropping it.
+    consumes_attn_spec: bool = False
+    # Whether this backend coexists with hybrid-linear (GDN/mamba) models. The
+    # linear layers bypass the backend entirely, but a backend whose metadata or
+    # graph machinery assumes layer 0 is an attention layer can opt out here.
+    hybrid_linear_ok: bool = True
+
+
+SUPPORTED_ATTENTION_BACKENDS = Registry[BackendCreator]("Attention Backend")
+
+
+@SUPPORTED_ATTENTION_BACKENDS.register(
+    "trtllm",
+    BackendInfo(
+        supported_types=frozenset({AttnType.FULL}),
+        requires_flashinfer=True,
+        requires_sm100=True,
+        page_sizes=(16, 32, 64),
+    ),
+)
+def create_trtllm_backend(config: ModelConfig):
+    from .trtllm import TensorRTLLMBackend
+
+    return TensorRTLLMBackend(config)
+
+
+@SUPPORTED_ATTENTION_BACKENDS.register(
+    "fi",
+    BackendInfo(
+        supported_types=frozenset({AttnType.FULL}),
+        requires_flashinfer=True,
+    ),
+)
+def create_fi_backend(config: ModelConfig):
+    from .fi import FlashInferBackend
+
+    return FlashInferBackend(config)
+
+
+@SUPPORTED_ATTENTION_BACKENDS.register(
+    "fa",
+    BackendInfo(
+        supported_types=frozenset({AttnType.FULL}),
+        requires_sgl_kernel=True,
+    ),
+)
+def create_fa_backend(config: ModelConfig):
+    from .fa import FlashAttentionBackend
+
+    return FlashAttentionBackend(config)
+
+
+@SUPPORTED_ATTENTION_BACKENDS.register(
+    "triton",
+    BackendInfo(
+        supported_types=frozenset({AttnType.FULL, AttnType.SWA}),
+        consumes_attn_spec=True,
+    ),
+)
+def create_triton_backend(config: ModelConfig):
+    from .triton import TritonAttentionBackend
+
+    return TritonAttentionBackend(config)
+
+
+@SUPPORTED_ATTENTION_BACKENDS.register(
+    "dsv4_sparse",
+    BackendInfo(supported_types=frozenset({AttnType.DSV4})),
+)
+def create_dsv4_sparse_backend(config: ModelConfig):
+    from .dsv4_sparse import DSV4SparseAttnBackend
+
+    return DSV4SparseAttnBackend(config)
+
+
+@SUPPORTED_ATTENTION_BACKENDS.register(
+    "dsa",
+    BackendInfo(supported_types=frozenset({AttnType.MLA, AttnType.DSA})),
+)
+def create_dsa_backend(config: ModelConfig):
+    from .dsa import DSAAttnBackend
+
+    return DSAAttnBackend(config)
+
+
+@SUPPORTED_ATTENTION_BACKENDS.register(
+    "m3_sparse",
+    BackendInfo(
+        supported_types=frozenset({AttnType.BSA}),
+        # One KV page == one 128-token sparse block: the top-k block ids ARE page
+        # indices and the block-base-row addressing needs page-aligned 128-row runs.
+        # Config-time resolution coerces any other page size here.
+        page_sizes=(128,),
+    ),
+)
+def create_m3_sparse_backend(config: ModelConfig):
+    from .m3_sparse import M3SparseAttnBackend
+
+    return M3SparseAttnBackend(config)
+
+
+def attention_backend_info(name: str) -> BackendInfo:
+    return SUPPORTED_ATTENTION_BACKENDS.info(name)
+
+
+def validate_attn_backend(backend: str, allow_auto: bool = True):
+    if backend != "auto":
+        parts = backend.split(",")
+        if len(parts) > 2:
+            from argparse import ArgumentTypeError
+
+            raise ArgumentTypeError(
+                f"At most two comma-separated attention backends are allowed "
+                f"(prefill,decode), got {backend!r}"
+            )
+        SUPPORTED_ATTENTION_BACKENDS.assert_supported(parts)
+    else:
+        assert allow_auto, "auto is not allowed here"
+    return backend
+
+
+def create_attention_backend(
+    backend: str,
+    config: ModelConfig,
+) -> BaseAttnBackend:
+    validate_attn_backend(backend, allow_auto=False)
+    if "," in backend:
+        p_backend, d_backend = backend.split(",", 1)
+        if p_backend != d_backend:
+            logger.info(f"Using hybrid attention backend: prefill={p_backend}, decode={d_backend}")
+            p_backend = create_attention_backend(p_backend, config)
+            d_backend = create_attention_backend(d_backend, config)
+            return HybridBackend(p_backend, d_backend)
+        backend = p_backend  # both are the same, fall through to single backend
+        logger.warning(f"P/D attention backends are the same: {backend}, using single backend.")
+
+    return SUPPORTED_ATTENTION_BACKENDS[backend](config)
+
+
+__all__ = [
+    "AttnType",
+    "BackendInfo",
+    "BaseAttnMetadata",
+    "BaseAttnBackend",
+    "AttentionSpec",
+    "attention_backend_info",
+    "create_attention_backend",
+    "SUPPORTED_ATTENTION_BACKENDS",
+    "validate_attn_backend",
+]