[Spec Decode] Remove Medusa and MLP speculator support by WoosukKwon · Pull Request #61701 · vllm-project/vllm

WoosukKwon · 2026-10-14T21:25:55Z

Overview

Removes Medusa and MLP speculator support and simplifies the code paths, configs, tests, docs, and CI rules that existed only for them. Net −858 lines, no new code.

Claims

method="medusa" and method="mlp_speculator" are removed. They are now rejected by SpeculativeConfig validation, which lists the valid methods.
The MLP speculator was already dead code: its registry entry was commented out ("temporarily disabled") until V1 support, which never landed. Medusa is not supported by Model Runner V2, and EAGLE/MTP/DFlash/DSpark cover its use case.
Other spec-decode methods are unchanged.

Validation

pre-commit (ruff, mypy, Buildkite/autolabel rule checks) passes.
pytest tests/models/test_registry.py tests/config/test_model_arch_config.py tests/config/test_speculative_draft_hf_overrides.py tests/v1/worker/test_gpu_warmup_blocks.py: 504 passed, 11 skipped, 2 failed. The 2 failures (test_registry_imports[Dots3Note*]) are unrelated: my local DeepGEMM build is older than the tag bump in [Build] Update deep_gemm git tag for locality_domain update #61491 (cannot import name 'get_block_m_for_nvfp4_mega_moe').
E2E smoke on the V1 model runner (VLLM_USE_V2_MODEL_RUNNER=0, Llama-3.1-8B-Instruct, GB200, 16 prompts, greedy, 128 tokens). Results are identical before and after:

Method	drafts	accepted	mean accept len
ngram (k=3)	151	192	2.27
eagle3 (k=3, yuhuili/EAGLE3-LLaMA3.1-Instruct-8B)	1480	568	1.38

No model eval is needed: no surviving method's output changes.

Details

Removed:

vllm/model_executor/models/{medusa,mlp_speculator}.py, vllm/transformers_utils/configs/{medusa,mlp_speculator}.py, vllm/v1/spec_decode/medusa.py, and their registry and config-map entries.
In SpeculativeConfig:
- the medusa/mlp_speculator method literals and their auto-detection;
- the old-format Medusa checkpoint workarounds (model_type hf_override injection and vocab-size patching);
- the mlp_speculator draft-TP=1 override, including the now-unused draft_hf_config parameter of _verify_and_get_draft_tp;
- the hf_config.num_lookahead_tokens sync (only Medusa/MLP speculator configs defined that attribute).
The MedusaProposer branches in the V1 gpu_model_runner.py.
The T/T_co generic on VllmModel/VllmModelForTextGeneration/VllmModelForPooling. It existed only for Medusa's list[torch.Tensor] hidden states; the default was already torch.Tensor, so typing for every other model is unchanged.
MedusaModelArchConfigConvertor, plus the check_head_size escape hatch in tests/config/test_model_arch_config.py, which only existed for Medusa's zero head size.
The Medusa e2e acceptance test, MLP docs page, MRV2-unsupported-method entries, Buildkite source deps, mergify/autolabel rules, and docs table rows.

Breaking change: users still passing method="medusa" or a Medusa/MLP-speculator draft checkpoint will get a config validation error.

Duplicate check: no open PR removes these. Draft #61357 migrates Medusa to the flat model layout; this PR would make it unnecessary.

Pull Request Checklist

I used vLLM's /pr-checklist skill. (Mandatory for agents, optional for humans).
AI assistance was used during the creation of this PR.
Design Fit: Minimizes impact on core components, reuses existing functionality, and justifies added complexity.
Testing and Validation: Validates the change and ensures any added tests are meaningful and reliable, with CI coverage or documented CI resource constraints and validation performed outside CI.
Code Quality and Style: Keeps code and comments clear and concise, and updates relevant documentation and examples.
Pull Request Contents: Includes a brief summary and relevant links, supports claims with evidence, explains root causes and implementation trade-offs, and follows the contributing guide.

🤖 Generated with Claude Code

Medusa and MLP speculator are superseded by EAGLE/MTP/DFlash/DSpark. The MLP speculator was already unregistered ("temporarily disabled") and Medusa is unsupported by Model Runner V2. Removes the models, HF configs, V1 proposer, spec method literals, auto-detection, Medusa checkpoint hacks, mlp_speculator draft-TP override, the num_lookahead_tokens sync, the Medusa-only generic on the VllmModel protocols, tests, docs, and CI/mergify references. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>

claude

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.

_{Tip: disable this comment in your organization's Code Review settings.}

mergify · 2026-10-14T21:26:37Z

Documentation preview: https://vllm--61701.org.readthedocs.build/en/61701/

WoosukKwon · 2026-10-14T21:27:42Z

/ci run

github-actions · 2026-10-14T21:28:13Z

✅ Triggered Buildkite CI #94942 for commit f4fa28231af9.

WoosukKwon · 2026-10-14T22:20:14Z

/ci run

github-actions · 2026-10-14T22:22:04Z

✅ Triggered Buildkite CI #94958 for commit da3b7e75cc40.

…lative-decoding README method list One conflict, docs/features/speculative_decoding/README.md: upstream vllm-project#61701 removed the MLP speculator line next to the Uno entry. Keep Uno, drop the MLP line. Signed-off-by: MJ <121681560+mjbrn@users.noreply.github.com>

Upstream vllm-project#61701 removed the Medusa and MLP speculator branches from the method lists and the speculator paths. Uno's factory branch is the trailing `elif` of init_speculator, and no test built a Uno config through it, so a broken import or an earlier branch claiming method='uno' would surface only at engine start on a GPU. The new test builds a real Uno SpeculativeConfig and checks init_speculator hands it to UnoSpeculator. Assisted-by: Claude <noreply@anthropic.com> Signed-off-by: MJ <121681560+mjbrn@users.noreply.github.com>

…61701) Signed-off-by: Woosuk Kwon <woosuk@inferact.ai> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

### What this PR does / why we need it? Port `vllm-ascend` to vLLM main `c0b6f74` (bumped from `a98247a`) while keeping the `v0.31.0` release lane green. The upstream window moved the KV-cache planning helpers into a new module, made EAGLE draft groups explicit, typed DeepSeek-V4 configs through upstream Transformers, removed Medusa/MLP speculator support, and changed several spec-decode/KV-cache/EPLB signatures. Every gate below is either a `vllm_version_is("0.31.0")` split or an attribute/argument probe where two main dev builds share one version string. | Files | Upstream vLLM change (why it breaks vllm-ascend) | vllm-ascend adaptation | |-------|--------------------------------------------------|------------------------| | `.github/vllm-main-verified.commit` | — (main-lane pin only) | Verified main commit bumped `a98247a` -> `c0b6f74` | | `tests/e2e/pull_request/four_card/test_minimax_m3.py` | The main-lane pin re-JITs the static kernels, so the first generate on a cold cache shifts the near-uniform dummy logprobs (~3e-3) enough to flip a greedy token | On main run a throwaway `generate_greedy_logprobs` before the baseline/replay pair so both measure the warmed steady state and share the cached prefix; v0.31.0 keeps the single-shot comparison | | `tests/ut/core/test_circular_buffer.py` | [vllm#61126](vllm-project/vllm#61126) (cascade-attention removal) moved the prefix-block query up to `KVCacheManager` and dropped it from single-type managers | On main assert only the manager's own `get_num_skipped_tokens(65536)`; keep the combined `get_num_common_prefix_blocks(...)` assert on v0.31.0 | | `tests/ut/core/test_kv_cache_interface.py` | [vllm#61183](vllm-project/vllm#61183) moved `is_kv_cache_spec_uniform` into `vllm.v1.core.kv_cache_planning` | Version-aware import (`kv_cache_utils` on v0.31.0 / `kv_cache_planning` on main) | | `tests/ut/distributed/ascend_store/_mock_deps.py` | [vllm#61183](vllm-project/vllm#61183) added the `vllm.v1.core.kv_cache_planning` module | Add it to the mocked-vLLM module list | | `tests/ut/distributed/ascend_store/test_request_lifecycle.py` | [vllm#58668](vllm-project/vllm#58668) trims `num_computed_tokens` to the tokens actually kept when a request stops mid token-list | On main subtract `(accepted + 1) - <newly appended output tokens>` from the expected computed count; keep the old expectation on v0.31.0 | | `tests/ut/distributed/kv_transfer/kv_p2p/test_mooncake_connector.py` | [vllm#61183](vllm-project/vllm#61183) moved `get_kv_cache_config_from_groups` / `is_kv_cache_spec_uniform` | Version-aware import | | `tests/ut/models/test_deepseek_v41_cache.py` | [vllm#61183](vllm-project/vllm#61183) relocated the planning helpers; [vllm#53558](vllm-project/vllm#53558) gave `_max_memory_usage_bytes_from_groups` a leading `builder` argument | Import from the live planning module and probe the upstream signature to call the 2-arg (v0.31.0) vs 3-arg (main) form, feeding an Ascend `_get_pool_bytes_per_block` builder | | `tests/ut/ops/test_gdn_attn_builder.py` | [vllm#58737](vllm-project/vllm#58737) derives `num_reqs` from the block table and asserts its row count equals `seq_lens`; Ascend's `unpadded()` keeps the graph-padded rows | New UT: a padded 4-row block table with 2 real requests must still yield a 2-row `cache_indices` | | `tests/ut/patch/platform/test_packed_kv_cache_compat.py` | [vllm#61183](vllm-project/vllm#61183) relocated the patched module; [vllm#61455](vllm-project/vllm#61455) removed `_annotate_eagle_groups` from main | Patch `cache_patch._kv_cache_planning`; new UT asserts the packed hook tolerates an upstream module without the EAGLE-annotation API | | `tests/ut/patch/platform/test_patch_eplb.py` | [vllm#57549](vllm-project/vllm#57549) emits the async EPLB cycle-committed log from the async worker wrapper on main | Expect the workspace-move log only on v0.31.0 | | `tests/ut/patch/platform/test_patch_kv_cache_coordinator.py` | [vllm#61183](vllm-project/vllm#61183) relocated the patch entry point | Assert against `kcu_patch._kv_cache_planning` instead of `vllm.v1.core.kv_cache_utils` | | `tests/ut/patch/platform/test_patch_pp_mtp.py` | [vllm#56322](vllm-project/vllm#56322) makes `ModelConfig.verify_with_parallel_config` read `parallel_config.enable_hc_sp` | Add `enable_hc_sp=False` to the constructed parallel config (ignored on v0.31.0) | | `tests/ut/patch/platform/test_prefix_cache_cp_patches.py` | [vllm#60043](vllm-project/vllm#60043) removed the `mamba_cache_mode="all"` leftover and requires `"align"` with prefix caching; [vllm#53558](vllm-project/vllm#53558) added the `builder` argument | Version-select the mamba cache mode; probe the upstream admission signature for the `builder` argument | | `tests/ut/spec_decode/test_mla_dspark.py` | [vllm#60692](vllm-project/vllm#60692) renamed the capture hook `precompute_context_kv(num_reqs)` -> `prepare_forward(num_reqs, attn_metadata)` | Version-split the expected capture args/kwargs | | `tests/ut/test_compressed_prefix_cache.py` | [vllm#61183](vllm-project/vllm#61183) moved `is_kv_cache_spec_uniform` | Version-aware import | | `tests/ut/worker/test_attn_utils_v2.py` | [vllm#61455](vllm-project/vllm#61455) annotates EAGLE groups from `spec.is_draft`; [vllm#60766](vllm-project/vllm#60766) resolves pin memory lazily via `is_pin_memory_available()` | New UT that draft-layer specs get tagged; on main monkeypatch `is_pin_memory_available` to `False` | | `tests/ut/worker/test_model_runner_v2.py` | [vllm#56716](vllm-project/vllm#56716) suppresses EPLB even for `skip_eplb` dummy runs; [vllm#60946](vllm-project/vllm#60946) renamed the PCP staging hook | Add a `runner.eplb` stub; call the version-appropriate staging method | | `tests/ut/worker/test_mtp_pcp_speculator_v2.py` | [vllm#60885](vllm-project/vllm#60885) rejects a mismatched explicit float16/bfloat16 KV-cache dtype | Build the model config with `bfloat16` instead of `float32` | | `tests/ut/worker/test_pcp_manager_v2.py` | [vllm#60946](vllm-project/vllm#60946) renamed `prepare_inputs_to_capture` -> `prepare_inputs_for_capture_or_profile` | Patch and call the live method name per lane | | `tests/ut/worker/test_v2_dummy_padding.py` | [vllm#61767](vllm-project/vllm#61767) collects DP votes through `get_dp_group().cpu_all_reduce_()` instead of `dist.all_reduce(..., group=cpu_group)` | Version-split the mocked collective | | `tests/ut/worker/v2/test_eplb_controller.py` | [vllm#57549](vllm-project/vllm#57549) gives `add_model` an optional `model_name` and names the draft entry | Expect the 3-arg call (`model_name="draft (draft)"`) on main, 2-arg on v0.31.0 | | `vllm_ascend/_310p/worker_310p.py` | [vllm#61183](vllm-project/vllm#61183) moved `get_kv_cache_groups` | Version-aware import | | `vllm_ascend/distributed/ec_transfer/ec_memcache_connector.py` | [vllm#57904](vllm-project/vllm#57904) dropped `_get_encoder_cache_hidden_dim` from `cpu.common` (encoder widths are measured at startup) | On main size the receive buffer as `out_hidden_size * (1 + len(deepstack_visual_indexes))` (Qwen3-VL DeepStack) else `get_inputs_embeds_size()`; keep the helper on v0.31.0 | | `vllm_ascend/models/deepseek_v4/compressor.py` | [vllm#61392](vllm-project/vllm#61392) switched `DeepseekV4Config` to the upstream Transformers class | Import from `transformers` on main, from `vllm.transformers_utils.configs.deepseek_v4` on v0.31.0 | | `vllm_ascend/models/deepseek_v4/indexer.py` | Same [vllm#61392](vllm-project/vllm#61392) | Same version-aware import | | `vllm_ascend/models/deepseek_v4/mm_preprocess.py` | Same [vllm#61392](vllm-project/vllm#61392) | Same version-aware import | | `vllm_ascend/models/deepseek_v4/model.py` | [vllm#61392](vllm-project/vllm#61392) moves the config to Transformers, folds `compress_ratios`/`num_hash_layers` into `layer_types`/`mlp_layer_types`, and keys YaRN under the compressor (`compress`) group | Restore the legacy fields on main from the new lists, read the `compress`-scoped rope params with safe defaults, and keep the legacy path on v0.31.0 | | `vllm_ascend/models/deepseek_v41/cache_config.py` | [vllm#61183](vllm-project/vllm#61183) moved `may_override_num_blocks` | Version-aware import | | `vllm_ascend/models/glm5next/cache_config.py` | [vllm#61183](vllm-project/vllm#61183) moved `create_kv_cache_group_specs` / `may_override_num_blocks` | Version-aware import | | `vllm_ascend/ops/gdn_attn_builder.py` | [vllm#58737](vllm-project/vllm#58737) derives `num_reqs` from the block table and asserts row parity with `seq_lens` | On main trim the padded block table to `seq_lens.shape[0]` before `mamba_get_block_table_tensor` when the sizes differ | | `vllm_ascend/patch/__init__.py` | [vllm#61183](vllm-project/vllm#61183) moved `get_kv_cache_config_from_groups` to `kv_cache_planning` | Update the patch-registry doc | | `vllm_ascend/patch/platform/__init__.py` | [vllm#61183](vllm-project/vllm#61183) moved the KV-cache planning/patcher module | Update the patch-registry doc | | `vllm_ascend/patch/platform/patch_eplb.py` | [vllm#57549](vllm-project/vllm#57549) moves the async EPLB cycle-committed log onto the async worker path on main | Report completion from the wrapped async worker (rank 0) on main; keep the workspace-move log on v0.31.0 | | `vllm_ascend/patch/platform/patch_kv_cache_utils.py` | [vllm#61183](vllm-project/vllm#61183) moved the planning helpers; [vllm#53558](vllm-project/vllm#53558) added the `builder` argument; [vllm#61455](vllm-project/vllm#61455) dropped `_annotate_eagle_groups` | Resolve the live planning module by version, patch its symbols, gate the EAGLE annotation on attribute presence, and dispatch the admission helper by argument count | | `vllm_ascend/quantization/methods/kv_cache/turboquant/cache.py` | [vllm#61183](vllm-project/vllm#61183) moved `may_override_num_blocks` | Version-aware import | | `vllm_ascend/spec_decode/__init__.py` | [vllm#61701](vllm-project/vllm#61701) removed Medusa/MLP speculator support | Import `AscendMedusaProposer` only on v0.31.0; raise a clear error for `method="medusa"` on main | | `vllm_ascend/spec_decode/medusa_proposer.py` | [vllm#61701](vllm-project/vllm#61701) removed `vllm.v1.spec_decode.medusa` | Add `# type: ignore[import-not-found]`; the module only loads on the lane that still ships it | | `vllm_ascend/worker/v1/model_runner.py` | Same [vllm#61701](vllm-project/vllm#61701) | Gate the Medusa import and the `isinstance(self.drafter, AscendMedusaProposer)` branch on v0.31.0 | | `vllm_ascend/worker/v2/attn_utils.py` | [vllm#61455](vllm-project/vllm#61455) annotates EAGLE groups from `spec.is_draft` | On main tag the spec with `is_draft` from `attn_module.is_draft_layer` | | `vllm_ascend/worker/v2/pcp_manager.py` | [vllm#60946](vllm-project/vllm#60946) added the `moe_gathers_tokens` argument and renamed `prepare_inputs_to_capture` -> `prepare_inputs_for_capture_or_profile` | Forward the new argument; version-split the staging hook | | `vllm_ascend/worker/v2/sample/apply_top_k_top_p.py` | [vllm#61996](vllm-project/vllm#61996) added `has_topp_only_rows` to the split top-p pipeline | Accept and ignore the extra argument on main (the PyTorch fallback handles top-p-only rows) | | `vllm_ascend/worker/v2/spec_decode/dflash/aclgraph.py` | [vllm#60692](vllm-project/vllm#60692) renamed `precompute_context_kv` -> `prepare_forward` | Pass the capture hook the installed vLLM expects | | `vllm_ascend/worker/v2/spec_decode/lmhead_tp_utils.py` | [vllm#59554](vllm-project/vllm#59554) added `num_steps` to `Speculator.sample_draft` | Version-split the override signature and forward `num_steps` on main | | `vllm_ascend/worker/v2/spec_decode/rejection_sampler_utils.py` | [vllm#61451](vllm-project/vllm#61451) renamed the logits-stats kernel argument `num_speculative_steps` -> `num_logits` and moved bonus-token detection into `_is_bonus_logit` | Pass `num_logits` on main; keep `num_speculative_steps` on v0.31.0 | | `vllm_ascend/worker/worker.py` | [vllm#61183](vllm-project/vllm#61183) moved `get_kv_cache_groups`/`unify_hybrid_kv_cache_specs`; [vllm#57516](vllm-project/vllm#57516) reclaims cached memory via `torch.accelerator.empty_cache()` inside `load_model` | Version-aware imports; on main neutralize `torch.npu.empty_cache` while loading weights under the sleep-mode pluggable memory pool | ### Does this PR introduce _any_ user-facing change? No. This is an internal main2main compatibility update; on the `v0.31.0` release lane behavior is unchanged, and on main the only behavioral note is that Medusa speculative decoding now raises a clear error (upstream removed it in vllm#61701). ### How was this patch tested? - Unit tests: the changed UT files above (KV-cache planning/coordinator, EPLB, PCP/spec-decode, DeepSeek-V4/V4.1 cache, mamba align, `worker/worker.py`). - E2E: `tests/e2e/pull_request/four_card/test_minimax_m3.py`. - Run the main lane against the pinned `c0b6f74` and the `v0.31.0` lane via `vllm_version_is`. --- ### Checklist #### Testing - [ ] **Boundary / edge-case coverage**: I reviewed whether this change involves boundary conditions (e.g. empty input, single/zero element, min/max values, off-by-one, sequence-length limits, integer overflow). Where applicable, I added unit tests covering these edge cases in `tests/ut/` or `tests/e2e/`. - [ ] **Regression test**: If this PR fixes a bug, I added a regression test that would fail without the fix. - [ ] **Error / failure path coverage**: I covered abnormal paths (exceptions, timeouts, out-of-range, division by zero) in addition to the happy path. - [ ] **Meaningful & reliable tests**: New tests assert real behavior and are not written merely to satisfy CI. - [ ] **Performance impact**: If this PR touches a performance-sensitive path, I attached benchmark results and added performance regression tests. - vLLM main: vllm-project/vllm@a98247a Signed-off-by: zhangxinyuehfad <starmoon_zhang@163.com>

… meta test Upstream dropped its draft_hf_config parameter with the Medusa and MLP speculator removal (vllm-project#61701); the GLM wrapper meta cases passed it and failed with a TypeError before building anything.

WoosukKwon requested review from AndreasKaratzas, DarkLight1337, Harry-Chen, MatthewBonanni, ProExpertProg, benchislett, hmellor, houseroad, khluu, mgoin, njhill, robertgshaw2-redhat, tlrmchlsmth, yewentao256, youkaichao and ywang96 as code owners October 14, 2026 21:25

claude Bot reviewed Oct 14, 2026

View reviewed changes

mergify Bot added documentation Improvements or additions to documentation ci/build speculative-decoding labels Oct 14, 2026

Merge branch 'main' into woosuk/remove-medusa-mlp-speculator

f4fa282

Merge branch 'main' into woosuk/remove-medusa-mlp-speculator

da3b7e7

njhill approved these changes Oct 14, 2026

View reviewed changes

WoosukKwon enabled auto-merge (squash) October 14, 2026 22:35

github-actions Bot added the ready ONLY add when PR is ready to merge/full CI is needed label Oct 14, 2026

WoosukKwon disabled auto-merge October 15, 2026 00:29

WoosukKwon merged commit 8530aed into main Oct 15, 2026
265 of 267 checks passed

WoosukKwon deleted the woosuk/remove-medusa-mlp-speculator branch October 15, 2026 00:29

pst2154 mentioned this pull request Oct 15, 2026

[Model] Add NemotronLabsDiffusion support with tri-mode inference #58387

Open

5 of 6 tasks

hmellor approved these changes Oct 15, 2026

View reviewed changes

kunling-anyscale mentioned this pull request Oct 15, 2026

Fix/spec draft inherit model weights #48023

Open

3 of 4 tasks

This was referenced Oct 16, 2026

[Model][Spec Decode] Migrate Medusa to the Flat Model layout #61357

Closed

[Model][Spec Decode] Migrate HY V3 to the Flat Model layout #61987

Open

[Roadmap] vLLM Roadmap Q4 2026 #61024

Open

zhangxinyuehfad mentioned this pull request Oct 21, 2026

[Misc] vllm main2main 1017 vllm-project/vllm-ascend#19302

Merged

5 tasks

Navigation Menu

Repository navigation

[Spec Decode] Remove Medusa and MLP speculator support - #61701

[Spec Decode] Remove Medusa and MLP speculator support#61701
WoosukKwon merged 3 commits into
mainfrom
woosuk/remove-medusa-mlp-speculator

WoosukKwon commented Oct 14, 2026 •

edited by github-actions Bot

Loading

Uh oh!

claude Bot left a comment

Uh oh!

mergify Bot commented Oct 14, 2026

Uh oh!

WoosukKwon commented Oct 14, 2026

Uh oh!

github-actions Bot commented Oct 14, 2026

Uh oh!

WoosukKwon commented Oct 14, 2026

Uh oh!

github-actions Bot commented Oct 14, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

WoosukKwon commented Oct 14, 2026 • edited by github-actions Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Overview

Claims

Validation

Details

Uh oh!

claude Bot left a comment

Choose a reason for hiding this comment

Claude Code Review

Uh oh!

mergify Bot commented Oct 14, 2026

Uh oh!

WoosukKwon commented Oct 14, 2026

Uh oh!

github-actions Bot commented Oct 14, 2026

Uh oh!

WoosukKwon commented Oct 14, 2026

Uh oh!

github-actions Bot commented Oct 14, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

WoosukKwon commented Oct 14, 2026 •

edited by github-actions Bot

Loading