yunshu 0.1.1__tar.gz

yunshu-0.1.1/.gitignore

@@ -0,0 +1,73 @@
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+.eggs/
+dist/
+build/
+*.egg
+.venv/
+
+# uv
+.python-version
+
+# Node / Next.js
+node_modules/
+webui/.next/
+webui/out/
+
+# Metal
+metal/build/
+
+# IDE
+.idea/
+.vscode/
+*.swp
+*.swo
+.DS_Store
+
+# Testing
+.coverage
+htmlcov/
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+
+# Benchmarks (local results only)
+bench/**/results/
+bench/**/*.json
+bench/**/*.csv
+
+# Environment
+.env
+.envrc
+
+# Logs
+*.log
+
+# Models (too large for git) — root-anchored, NO trailing slash so it also matches the
+# `models` symlink to an external drive (a trailing slash would match dirs only).
+/models
+
+# Internal / development
+.claude/
+reference/
+# Local archive of the pre-refocus tree (never published)
+/_archive_yunshu/
+PROGRESS_GUIDE.md
+CLAUDE.md
+reports/
+
+.build/
+testresource/
+# Auxiliary tool venvs (oMLX comparison env, etc.) — large, never committed
+.venvs/
+
+# The broad `reports/` rule above also catches docs/reports/, but that dir IS the
+# deliverable (unified report + trend charts) — track it, except the unbounded
+# local version-history snapshots.
+!docs/reports/
+!docs/reports/**
+docs/reports/history/
+docs/research/
+scripts/research/local.env

yunshu-0.1.1/CHANGELOG.md

@@ -0,0 +1,210 @@
+# Changelog
+
+All notable changes to Yunshu are documented here. The format is based on
+[Keep a Changelog](https://keepachangelog.com/). Versions go up in small steps: a patch
+(0.1.1, 0.1.2, …) for fixes and improvements, a minor (0.2.0) only for a real capability jump.
+Release steps: [RELEASING.md](RELEASING.md).
+
+## [Unreleased]
+
+## [0.1.1] - 2026-09-29
+
+Yunshu is now positioned as a local LLM / VLM inference engine (decode speed, TTFT, prefix reuse,
+API completeness), with Qwen3.8-27B as the first fully tuned model. Speech-to-speech and the other
+modalities remain supported capabilities; the README, CLAUDE.md and AGENTS.md were restructured to
+match.
+
+### Added
+
+- First-run commands: `yunshu doctor` checks the Mac (Apple Silicon / Rosetta, macOS, Python, MLX
+  and Metal, memory, settings, models directory, the model and whether it fits, port) and prints a
+  fix for each problem; `yunshu pull <org/name>` downloads into the models directory, refuses to
+  download a model that is already on disk (models directory or Hugging Face cache) and resumes an
+  interrupted download; `yunshu model list` also lists the Hugging Face cache; `yunshu --version`.
+- Models from the Hugging Face cache are usable in place: `yunshu serve -m org/name` (or a name under
+  the models directory) serves the local copy instead of downloading, and multi-model mode lists
+  cached models next to the models directory (`YUNSHU_HF_CACHE_MODELS`, on by default).
+- `yunshu config set KEY VALUE` / `unset` / `path`: settings saved in `~/.yunshu/config.toml`, read
+  by every command and the service (e.g. `yunshu config set models_dir /Volumes/Models`).
+- `yunshu service install|uninstall|start|stop|restart|status|logs`: a per-user launchd agent
+  that starts Yunshu at login, restarts it after a crash, and logs to `~/Library/Logs/Yunshu`.
+- Docs: [connecting clients](docs/guides/CLIENTS.md), [service](docs/guides/SERVICE.md),
+  [troubleshooting](docs/guides/TROUBLESHOOTING.md), [benchmark sources](docs/BENCHMARKS.md),
+  [RELEASING.md](RELEASING.md); a tag-triggered release workflow (PyPI trusted publishing behind
+  an approval step) and a Homebrew formula for the `yuhuanstudio/tap` tap (`packaging/homebrew/`).
+- One settings registry for every `YUNSHU_*` setting: a TOML config file (`--config`),
+  `yunshu serve --set KEY=VALUE`, `yunshu config` (effective value and source of each), generated
+  [docs/CONFIGURATION.md](docs/CONFIGURATION.md); a bad value stops startup, a misspelled name
+  warns.
+- Qwen3.5-family VLM runner (Qwen3.5 / 3.6 / 3.8) on `mlx-vlm`'s generator: prefix cache with
+  exact hybrid checkpoints (text + image-pixel keys, optional SSD tier), MTP or DFlash speculative
+  decode, streaming reasoning split, tools, JSON schema, stop sequences, logprobs and cancel.
+  On Qwen3.8-27B (M5 Max) warm chat TTFT went from 2.5 s to 0.2 s, a repeated 8K prompt from 9 s to
+  0.1 s, and decode from 30 to 57 tok/s.
+- Exact verify kernels (GatedDeltaNet prework/replay, split SDPA, 5-bit streamed matmul) and
+  batch-invariant decode kernels, partly vendored from oMLX (Apache-2.0).
+- Per-row-length (ragged) KV cache for models with Qwen3.5-family attention, now the default: rows
+  in the shared batch no longer read or copy the longest row's padding, and the single-request
+  speculative lane runs decode and verify attention on one per-row kernel. Qwen3.8-27B (M5 Max):
+  MMLU-Pro 300 at 8 in flight 249/300 in 28.3 min at 139 tok/s, peak 33.7 GiB (padded cache:
+  250/300, 46.5 min, 88 tok/s, 45 GiB); single-request decode at 131K context 24.5 → 46.6 tok/s,
+  at 32K 39.6 → 59.1; short contexts unchanged within noise; spec on == off at 1K / 32K / 131K.
+- `YUNSHU_KV_PRECISION=bf16|int8` (default bf16): int8 K/V with one fp16 scale per 32 dims for the
+  shared batch — about half the KV memory and bandwidth, lossy (MMLU-Pro 251/300, peak 28.1 GiB).
+- Experimental own round driver for the Qwen3.5 family (`YUNSHU_ROUND_DRIVER`, off by default):
+  multi-row MTP drafting with a cost-aware draft depth (TensorFold's allocation rule, MIT) and
+  multi-prompt prefill; being measured.
+- `YUNSHU_LOG_LEVEL`; `scripts/research/` benchmark matrix, realistic soak and MMLU-Pro soak.
+
+- Automatic speculative path: a DFlash2 drafter that matches the served model (Qwen3.8-27B:
+  `incoai/Qwen3.8-27B-DFlash2`, in the models directory or the Hugging Face cache) is used without
+  a flag, with the cost-aware chain depth and 8-bit drafter weights. Qwen3.8-27B server, novel_en
+  1K/8K/32K/131K 57.1 / 48.2 / 46.1 / 32.9 tok/s (MTP 47.5 at 1K), code corpus 82.0 / 89.1 / 70.5 /
+  79.9, 34/34 checks, TTFT about 207 s at 131K; lossless (spec on == off). `YUNSHU_VLM_DRAFT=mtp`
+  forces the MTP head, `=off` disables drafting, a path picks a drafter; `yunshu doctor` prints the
+  path that will be used.
+- `/health/ready` returns a `reason` when the model failed to load.
+- `scripts/dev/robustness.py`: black-box probe (load failure, disconnect during prefill and
+  streaming, limits, mixed concurrent features, 500-request memory, SIGTERM / SIGINT with requests
+  in flight).
+
+### Fixed
+
+- A non-streaming request whose client disconnected kept the GPU busy to the end of its prefill or
+  `max_tokens`: Starlette's `is_disconnected()` never fires behind `BaseHTTPMiddleware`. A
+  pure-ASGI watcher now records the disconnect and the request's row is dropped at the next chunk
+  (prefill of a 60K-token prompt: next request waited 4.0 s before, 0.8 s after, on a 0.8B model).
+- JSON-schema, regex and grammar constraints no longer allow special tokens (`<|im_end|>`) inside
+  a string: a model that chose one ended the response mid-string (Qwen3.5-0.8B stopped at
+  `{"name": "Alice`).
+- Ctrl-C / SIGTERM with streams in flight waited for them forever (a 27B stream takes minutes);
+  connections are now cut after `YUNSHU_DRAIN_TIMEOUT` (default 30 s) and the server exits.
+
+### Changed
+
+- A model that fails to load (missing path, truncated weights, out of memory) no longer stops
+  `yunshu serve`: the server stays up, `/health/ready` answers 503 with the reason and chat requests
+  get a 503. `yunshu serve` still prints the problem and its fix first.
+- Rate limiting is off by default (`YUNSHU_RATE_LIMIT_RPM=0`; the old 120 per minute rejected a
+  local client's 500 sequential requests with 429). Set it above 0 to enable it.
+- `yunshu serve` binds `127.0.0.1` by default (was `0.0.0.0`) and warns when it serves the network
+  without `--auth-token`. It checks the model before loading and stops with a fix when the path
+  does not exist, a download is incomplete, or the weights exceed memory.
+- The default models directory is `~/.yunshu/models` (was the source tree's `models/`, which
+  pointed inside the installed package for a wheel install). `YUNSHU_MODELS_DIR` still overrides.
+- `/version`, the OpenAPI schema and the MCP `serverInfo` report the installed package version
+  (they said `0.1.0-dev`).
+- With `--auth-token`, the token is also accepted as `x-api-key` (what Anthropic SDKs send).
+- The wheel and sdist include `THIRD_PARTY_NOTICES.md`; the sdist no longer carries benchmark data.
+- Default decode for MTP models is lossless batch-invariant decode at MTP block 6: decode and verify
+  matmuls of a drafting request share one row-invariant kernel (M5 tensor-unit packed where
+  available), so greedy output with speculation on equals speculation off. Qwen3.8-27B (M5 Max,
+  in-process) code / prose / JSON-like: 88.6 / 59.9 / 67.3 tok/s, vs 57–67 / 50–53 / 58–62 for the
+  previous exact kernels at block 3; server matrix decode 57 → 80 tok/s, 33/33 checks. Requests that
+  cannot draft (sampling, logits processors, logprobs) keep the stock kernels.
+- Every `mlx-vlm` model is served by the VLM batch runner: concurrent requests share one
+  continuous batch with per-row sampling settings, and a request that is alone uses speculative
+  decoding (Qwen3.5 family). The older per-request VLM loop was deleted; how it worked is recorded
+  in `docs/archive/legacy_vlm_loop/`.
+- 5/6/8-bit projections run on TensorFold's integer-code tensor-unit matmul when packed (M5).
+- Lossy optimizations are options with a lossless default. Text engine: KV cache quantization
+  (`YUNSHU_KV_QUANT_BITS`, was `auto`) and the 4-bit warm prefix tier (`YUNSHU_PREFIX_HOT_LIMIT`)
+  are off by default, and the SSD prefix cache stores KV and recurrent state bit-exact
+  (`YUNSHU_SSD_CACHE_PRECISION=native`; `int8` stays available and old int8 files still load).
+- `reasoning_effort` (top-level or in `chat_template_kwargs`) is passed to chat templates that
+  support it (Qwen3.8: low / medium / xhigh) instead of being mapped to a thinking-token cap.
+- Dependencies: MLX 0.32.2, transformers 5.17, upstream `mlx-vlm` 0.7.3 (the fork is gone).
+- JSON-schema constraint: cached vocab split (in-string step 144 ms → 3 ms); structural whitespace
+  limited to spaces and newlines.
+- Research scripts and parity tests take model paths from arguments or the environment
+  (`M`/`D`/`TF` for the validation scripts, optionally from a gitignored
+  `scripts/research/local.env`; `YUNSHU_PARITY_MODEL` for the parity tests, which skip when unset).
+- A model served from the Hugging Face cache is listed under its repo id (`org/name`) in
+  `/v1/models`, not the snapshot path.
+
+### Fixed
+
+- Streaming thinking on Qwen3.8 was sent as content; it is now sent as reasoning.
+- Gateway startup failure no longer crashes in shutdown on an unbound engine.
+- VLM responses under concurrent load are returned as soon as each generation finishes; before, a
+  post-request cache clear queued behind every other request held all responses until the last
+  one finished. The non-streaming VLM path also no longer applies a 120 s default timeout that
+  counted time spent waiting in the queue (a client-set `timeout` still applies).
+- Tool calls from non-Qwen families (Gemma-4, GLM-4.7, Mistral, pythonic, DeepSeek, Kimi,
+  MiniMax, …) were left in `content` with `tool_calls` empty. The tool-call format now comes from
+  the model's chat template (upstream mlx-vlm/mlx-lm registry) instead of guessing from the
+  request's model name, and every route (chat, Anthropic, Responses, Realtime; streaming and not)
+  parses with that format's upstream parser. Arguments are typed with the request's tool schemas.
+  Gemma-4 E4B capability checks: 7/34 → 33/34.
+- DFlash2 speculative decode: the drafter's hidden-state capture covered the whole prompt instead
+  of the drafter's 2047-position window (about 48 TFLOP extra at 131K), the draft block was capped
+  at 4 (the drafter is trained for 8), and speculation on did not equal speculation off at longer
+  contexts; it now uses the per-row prefill path and batch-invariant kernels like MTP.
+- The SSD prefix cache's native-precision bf16 writer reinterpreted values instead of bits (it was
+  never reached while every write went through int8).
+
+### Removed
+
+- The deprecated `Engine` class, multi-node / multi-tenant leftovers (distributed diffusion,
+  connection pool, data-parallel endpoint, tenant auth naming), and unused modules
+  (`vlm_async_engine`, `wan_vae`, `lid`, `optimizations`, `vision_encoding`).
+- The older VLM MTP / APC side paths, superseded by the runner.
+- The runner's upstream quantized-KV option (slower than bf16 at every measured length) and other
+  measured-out flags (`YUNSHU_PACKED_5BIT`, `YUNSHU_VLM_KV_BITS`, `YUNSHU_VLM_INVARIANT`,
+  `YUNSHU_MTP_FAST_VERIFY`, `YUNSHU_RAGGED_KV` — ragged is now always on for qwen3_5 attention).
+- Yunshu's own tool-call parsers (`tool_call_parser.py`, `tool_call_parsers.py`, about 4,500 lines
+  with their tests), superseded by the upstream parser registry; Yunshu keeps only the formats
+  upstream lacks (its prompt-injected JSON format, DeepSeek, whole-message JSON).
+
+## [0.1.0] - 2026-07-01
+
+First public release — the full capability surface below, each endpoint smoke-verified
+against a real model on Apple Silicon.
+
+### Added
+
+- **Native streaming speech-to-speech** (the flagship): Qwen3-Omni Thinker→Talker
+  on Apple MLX via `mlx-vlm` — `POST /v1/omni/speech/stream` (SSE) and the
+  OpenAI-Realtime `WS /v1/realtime` socket. Speech-in (raw audio, no ASR) and
+  speech-out, ~1.2 s first-audio text-in / ~1.4 s speech-in (warm). Multi-turn
+  conversation context preserved (bounded to protect TTFT). Tool-calling works on the
+  voice path too — a tool turn emits `function_call` items and the spoken JSON is
+  suppressed (the voice doesn't read the call aloud).
+- **OpenAI-compatible API**: chat/completions (tool-calling, JSON-schema/grammar,
+  streaming, logprobs), completions, responses, embeddings, audio
+  (transcriptions/translations/speech), images (generation + edits/variations/
+  inpaint/controlnet), tokenizer utilities, batch inference.
+- **Tool-calling**: tools are rendered via the model's own chat template when it supports
+  them natively (better adherence), and a forced `tool_choice` (required / named) is
+  structurally enforced with an assistant prefill — consistently across
+  `/chat/completions`, `/v1/responses`, and Anthropic `/v1/messages`.
+- **Trained image ControlNet**: the `/images/controlnet` + `/depth-guided` routes run the
+  real Z-Image Fun-Controlnet-Union when weights are present (canny/depth preprocessing),
+  and inline `<lora:name:weight>` applies on every image route.
+- **Anthropic-compatible** `/v1/messages` (+ `count_tokens`).
+- **Multimodal retrieval**: `Qwen3-VL-Embedding` (text / image / cross-modal in one
+  shared space) on `/v1/embeddings`, and `Qwen3-VL-Reranker` true cross-encoder on
+  `/v1/rerank` (bi-encoder cosine fallback otherwise).
+- **Vision/VLM, OCR, ASR, TTS, image & video generation, MCP** server/client.
+- **Single-node decoding/optimization**: single-request fast path (`mlx-lm
+  generate_step`), KV prefix + prompt caching, lossless n-gram speculative decode
+  on greedy (default-on), per-request top-nσ sampler, constrained decoding
+  (JSON-schema/regex/grammar), jump-forward, GPU sampler, in-memory MXFP4/NVFP4
+  weight quant, automatic KV quant. The situational ones are opt-in (see
+  [docs/CONFIGURATION.md](docs/CONFIGURATION.md)).
+- Single-model mode serves the loaded model under any requested name
+  (Ollama/LM-Studio behavior); multi-model auto-discovery via `YUNSHU_MODELS_DIR`.
+- Docs: [API reference](docs/API.md), [configuration reference](docs/CONFIGURATION.md),
+  runnable [examples/](examples/).
+
+### Changed
+
+- **Refocused** from an over-scoped "distributed inference platform" to an honest
+  single-node omni engine. Removed: multi-node mesh / distributed paths
+  (sharded-load, disaggregated prefill/decode), the multi-tenant control plane,
+  and tiered-KV offload. All single-node decoding/optimization tech is kept.
+
+### Notes
+
+- macOS + Apple Silicon only (wraps MLX; no custom Metal kernels). Python 3.13+, `uv`.
+- Apache-2.0 licensed.

yunshu-0.1.1/LICENSE

@@ -0,0 +1,202 @@
+
+                                 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 2024-2026 Yuhuan
+
+   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.