[Bugfix] vLLM should check Inductor config for compile cache enablement status #27637
Conversation
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
gmagogsfm
requested review from
ProExpertProg,
youkaichao and
zou3519
as code owners
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run You ask your reviewers to trigger select CI tests on top of Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add If you have any questions, please reach out to us on Slack at https://slack.vllm.ai. 🚀 |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Code Review
This pull request correctly addresses a bug where vLLM's compile cache status did not respect PyTorch's global setting (TORCHINDUCTOR_FORCE_DISABLE_CACHES). The changes ensure that vLLM checks the Inductor config for cache enablement, making the behavior consistent. However, the implementation introduces code duplication for the cache status check across three different locations. I've added a high-severity comment recommending a refactor to centralize this logic, which will improve maintainability and reduce the risk of future inconsistencies, especially with the planned update for PyTorch 2.9.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
gmagogsfm
force-pushed
the
disable_cache_error
branch
3 times, most recently
from
211f249 to
c0083f3
Compare
zou3519
reviewed
Oct 30, 2025
Comment on lines
167
to
169
| # TODO(gmagogsfm): Replace torch._inductor.config.force_disable_caches | ||
| # with torch.compiler.config.force_disable_caches when minimum PyTorch | ||
| # version reaches 2.9 |
Collaborator
There was a problem hiding this comment.
I don't see this API in 2.9
Collaborator
There was a problem hiding this comment.
If your proposal was to make this public -- The in vLLM code you have put this into is for InductorAdaptor, which we are trying to deprecate in favor of StandaloneInductorAdaptor, so I wouldn't worry about it too much. We should end up deleting this code anyways in 2.10+
Contributor
Author
There was a problem hiding this comment.
This force_disable_caches behavior actually affects more than just InductorAdaptor, it is also used in InductorStandaloneAdapter. In fact the new error I showed you earlier is from InductorStandaloneAdaptor.
I changed comment to mention 2.10. Also made a PR to expose this flag:
pytorch/pytorch#166699
Please take a look.
zou3519
reviewed
Oct 30, 2025
Collaborator
zou3519
left a comment
zou3519
left a comment
There was a problem hiding this comment.
code lgtm, I would just change the comment
gmagogsfm
force-pushed
the
disable_cache_error
branch
from
c0083f3 to
397534b
Compare
zou3519
approved these changes
Oct 31, 2025
ProExpertProg
requested changes
Oct 31, 2025
vllm/compilation/backends.py
Outdated
|
|
||
| # store the artifact in the cache | ||
| if not envs.VLLM_DISABLE_COMPILE_CACHE and handle is not None: | ||
| if is_compile_cache_enabled() and handle is not None: |
Collaborator
There was a problem hiding this comment.
The user might also pass force_disable_caches as part of CompilationConfig.inductor_compile_config, we should check that as well
gmagogsfm
force-pushed
the
disable_cache_error
branch
from
397534b to
60b709e
Compare
| not envs.VLLM_DISABLE_COMPILE_CACHE | ||
| and not torch._inductor.config.force_disable_caches | ||
| and not ( | ||
| get_current_vllm_config().compilation_config.inductor_compile_config.force_disable_caches |
Collaborator
There was a problem hiding this comment.
I don't think that vllm config is necessarily available at this time, why not just pass compilation config as a parameter?
Collaborator
There was a problem hiding this comment.
Also, inductor compile config is a dictionary, and the key might not be present:
Suggested change
| get_current_vllm_config().compilation_config.inductor_compile_config.force_disable_caches | |
| compilation_config.inductor_compile_config.get("force_disable_caches", False) |
Contributor
Author
There was a problem hiding this comment.
Fixed. Now is_compile_cache_enabled checks inductor configs explicitly passed in.
I noticed that vllm config is not present at this time during later testing. In general, when should I expect global vllm config to be available and when not? Is it mostly expected to be available during model init?
vllm/compilation/decorators.py
Outdated
| InliningInstructionTranslator, "inline_call_", patched_inline_call | ||
| ), | ||
| torch._dynamo.config.patch(**dynamo_config_patches), | ||
| set_current_vllm_config(self.vllm_config), |
Collaborator
There was a problem hiding this comment.
This is a huge change, we shouldn't be settling the vllm config globally here
Contributor
Author
There was a problem hiding this comment.
Removed now that the inductor config is passed in explicitly.
gmagogsfm
force-pushed
the
disable_cache_error
branch
from
60b709e to
ef8a3fb
Compare
|
This pull request has merge conflicts that must be resolved before it can be |
Tags: Signed-off-by: Yanan Cao <[email protected]>
gmagogsfm
force-pushed
the
disable_cache_error
branch
from
ef8a3fb to
9f1dc0a
Compare
Contributor
Author
|
@ProExpertProg When you get a chance, could you take a look at the new revision? your suggested changes are addressed. Thanks |
ProExpertProg
approved these changes
Nov 5, 2025
1 task
zWaNg3
added a commit
to fangyuchu/vllm
that referenced
this pull request
Nov 7, 2025
* add fault_report_addr in FaultToleranceConfig * add handle fault&get_fault_info api Signed-off-by: w00689259 <[email protected]> * remove fault_report_address in CoreEngineActorManager __init__ Signed-off-by: a798347923 <[email protected]> * ruff format Signed-off-by: a798347923 <[email protected]> * add handle fault&get_fault_info api Signed-off-by: w00689259 <[email protected]> * fix one bug. Signed-off-by: fangyuchu <[email protected]> * add fault_report_port in FaultToleranceConfig Signed-off-by: a798347923 <[email protected]> * add zmq_addr concatenate with fault_report_addr and fault_report_port Signed-off-by: a798347923 <[email protected]> * fault reporter bug fix Signed-off-by: w00689259 <[email protected]> * fault reporter bug fix Signed-off-by: w00689259 <[email protected]> * fault reporter bug fix Signed-off-by: w00689259 <[email protected]> * fault reporter bug fix Signed-off-by: w00689259 <[email protected]> * fault reporter bug fix Signed-off-by: w00689259 <[email protected]> * fault reporter bug fix Signed-off-by: w00689259 <[email protected]> * fix some bug * fault reporter bug fix Signed-off-by: w00689259 <[email protected]> * fault reporter bug fix Signed-off-by: w00689259 <[email protected]> * remove fault_report_addr in FaultToleranceConfig Signed-off-by: a798347923 <[email protected]> * refactor: relocate method serialization functions to serial_util.py Signed-off-by: fangyuchu <[email protected]> * fix actor bug * fix actor bug * add engine_core_cmd_addr in FaultToleranceConfig Signed-off-by: a798347923 <[email protected]> * add and use _stop_worker_execution in EngineCoreGuard Signed-off-by: a798347923 <[email protected]> * add and use run in WorkerGuard Signed-off-by: a798347923 <[email protected]> * fix actor bug * fix bug * fix sentinel * fix bug vllm/v1/engine/core.py:847: error: Missing positional argument "tp_size" in call to "EngineCoreGuard" Signed-off-by: a798347923 <[email protected]> * fix bug error: Missing positional arguments "length", "byteorder" in call to "to_bytes" of "int" Signed-off-by: a798347923 <[email protected]> * fix bug in fault tolerance mode Signed-off-by: w00689259 <[email protected]> * fix bug in fault tolerance mode Signed-off-by: w00689259 <[email protected]> * change fault_report_port to internal_fault_report_port add external_fault_notify_port Signed-off-by: a798347923 <[email protected]> * change fault_report_port to internal_fault_report_port add external_fault_notify_port Signed-off-by: a798347923 <[email protected]> * add _recv_cmd func use deserialize_method_call and run_method in run func Signed-off-by: a798347923 <[email protected]> * Update core.py fix bug error: Need type annotation for "kwargs" (hint: "kwargs: dict[<type>, <type>] = ...") Signed-off-by: a798347923 <[email protected]> * add self.ctx.term() in shutdown() Signed-off-by: a798347923 <[email protected]> * changed import deserialize_method_call,serialize_method_call Signed-off-by: a798347923 <[email protected]> * changed init worker_guard in init_device Signed-off-by: a798347923 <[email protected]> * Update core.py add import serialize_method_call Signed-off-by: a798347923 <[email protected]> * Update gpu_worker.py changed init WorkerGuard in init_device Signed-off-by: a798347923 <[email protected]> * Update gpu_worker.py FIX BUG self.worker_guard: WorkerGuard|None = None Signed-off-by: a798347923 <[email protected]> * Update gpu_worker.py fix bug error: Argument 1 to "deserialize_method_call" has incompatible type "str | None"; expected "str" [arg-type] Signed-off-by: a798347923 <[email protected]> * Update gpu_worker.py ruff format Signed-off-by: a798347923 <[email protected]> * Update core.py ruff-format Signed-off-by: a798347923 <[email protected]> * actively send exception information Signed-off-by: w00689259 <[email protected]> * actively send exception information Signed-off-by: w00689259 <[email protected]> * actively send exception information Signed-off-by: w00689259 <[email protected]> * change engine_core_cmd_addr(str) to engine_core_cmd_addrs(list[str]) in EngineZmqAddresses Signed-off-by: a798347923 <[email protected]> * change engine_core_cmd_addr(str) to engine_core_cmd_addrs(list[str]) in EngineZmqAddresses Signed-off-by: a798347923 <[email protected]> * Update utils.py delete engine_core_cmd_addr in EngineZmqAddresses Signed-off-by: a798347923 <[email protected]> * Remove redundant configuration: fault-pub-port Signed-off-by: fangyuchu <[email protected]> * Send pause instructions after receiving fault info in ClientGuard Signed-off-by: fangyuchu <[email protected]> * change engine_core_guard_identities from dict[int, bytes] to list[bytes] Signed-off-by: a798347923 <[email protected]> * fix bug "only the worker guard of engine core 0 can receive messages sent from engine core guard Signed-off-by: a798347923 <[email protected]> * change local_rank to rank_in_group in WorkerGuard Signed-off-by: a798347923 <[email protected]> * changed del self.client_cmd_registry[int(unhealthy_engine.engine_id)] Signed-off-by: a798347923 <[email protected]> * add gloo communication timeout * fix some bug * add stateless_process_group gloo_comm_timeout * reconstruct fault receiver&fault handler Signed-off-by: w00689259 <[email protected]> * fix some bug * reconstruct fault receiver&fault handler Signed-off-by: w00689259 <[email protected]> * reconstruct fault receiver&fault handler Signed-off-by: w00689259 <[email protected]> * fix return format Signed-off-by: w00689259 <[email protected]> * fix return format Signed-off-by: w00689259 <[email protected]> * fix return format Signed-off-by: w00689259 <[email protected]> * add abort request * fix some bug * fix some bug * fix some bug * add dt for client guard Signed-off-by: w00689259 <[email protected]> * add dt for client guard Signed-off-by: w00689259 <[email protected]> * add dt for client guard Signed-off-by: w00689259 <[email protected]> * Implementation of two types of pause: a soft one by using flag signals and a hard one by aborting nccl communicators. Signed-off-by: fangyuchu <[email protected]> * Refine certain log forms and fix a minor bug in pause function. Signed-off-by: fangyuchu <[email protected]> * Refactor and abstract the recv_msg logic in CG,ECG,WG. Signed-off-by: fangyuchu <[email protected]> * [Frontend] Align finish_reason when tool is called with OpenAI (vllm-project#25054) Signed-off-by: Sungyoon Jeong <[email protected]> Co-authored-by: Chauncey <[email protected]> * [Hybrid] Pass kernel block size to builders (vllm-project#27753) Signed-off-by: Thomas Parnell <[email protected]> * [Bugfix] Padded Eagle Specdec with Chunked Prefill (vllm-project#26263) Signed-off-by: Rémi Delacourt <[email protected]> Signed-off-by: Rémi Delacourt <[email protected]> Signed-off-by: remi <[email protected]> Co-authored-by: Benjamin Chislett <[email protected]> * [XPU]Refine Dockerfile.xpu, avoid oneccl dependency issue (vllm-project#27964) Signed-off-by: Kunshang Ji <[email protected]> * Add and check method uuid when sending commands and receiving results. Signed-off-by: fangyuchu <[email protected]> * Add ORCA endpoint load metrics support (vllm-project#24905) Signed-off-by: Misha Efimov <[email protected]> * [CI/Build] Remove the flaky gpt-oss lora test (vllm-project#27966) Signed-off-by: Jee Jee Li <[email protected]> * Abstract the logic of sending instructions and waiting responses from FaultHandler Signed-off-by: fangyuchu <[email protected]> * [Model] Add PaddleOCR-VL Model Support (vllm-project#27758) Signed-off-by: zhangyue <[email protected]> Signed-off-by: Roger Wang <[email protected]> Signed-off-by: Isotr0py <[email protected]> Signed-off-by: zhangyue66 <[email protected]> Co-authored-by: Roger Wang <[email protected]> Co-authored-by: Isotr0py <[email protected]> * Add options in EngineCoreGuard to recv execution results from WorkerGuard Signed-off-by: fangyuchu <[email protected]> * Early exit for MoE LoRA kernels (vllm-project#27131) Signed-off-by: gnovack <[email protected]> Co-authored-by: Jee Jee Li <[email protected]> * [Bugfix] Skip gs:// model paths for speculator detection (vllm-project#27846) Signed-off-by: Peter Schuurman <[email protected]> * [BUG] Make 'binary' default option for saving torch compile artifacts when using standalone_compile (vllm-project#27616) Signed-off-by: ahao-anyscale <[email protected]> * [CI/Testing] Add basic single node dual batch overlap test (vllm-project#27235) Signed-off-by: Lucas Wilkinson <[email protected]> * [Spec Decode] Integrate Suffix Decoding from Arctic Inference (vllm-project#25784) Co-authored-by: Aurick Qiao <[email protected]> * [Feature][Benchmarks] Support `inf` burstiness (vllm-project#26941) Signed-off-by: Sophie du Couédic <[email protected]> * [Bugfix][Qwen][Multimodal] Move Qwen2_5_vl sdpa to custom op and reenable compile (vllm-project#27764) Signed-off-by: Lucas Kabela <[email protected]> * [Bugfix] change FlashMLA reorder_batch_threshold (vllm-project#27777) Signed-off-by: Matthew Bonanni <[email protected]> * [Docs] add runai_streamer_sharded to LoadConfig (vllm-project#27937) Signed-off-by: Andy Xie <[email protected]> * Add TP parameter to attention tests (vllm-project#27683) Signed-off-by: Matthew Bonanni <[email protected]> * [Bugfix][plugin] fla crash on plugin (vllm-project#27322) * [Bugfix] Fix MoE Routing Simulation (vllm-project#28002) Signed-off-by: Tyler Michael Smith <[email protected]> * Remove the tpu docker image nightly build. (vllm-project#27997) Signed-off-by: Qiliang Cui <[email protected]> * [Bugfix][ROCm] Fix ViT rotary embeddings for torch.compile compatibility on ROCm (vllm-project#27748) Signed-off-by: vllmellm <[email protected]> * [LoRA] Lora shrink swizzle (vllm-project#27694) Signed-off-by: li2haipeng <[email protected]> Signed-off-by: Haipeng Li <[email protected]> Co-authored-by: Jee Jee Li <[email protected]> * [Refactor] Lazy import tool_parser (vllm-project#27974) Signed-off-by: chaunceyjiang <[email protected]> * [NIXL][XPU] Pin NIXL version to 0.7.0 (vllm-project#27849) Signed-off-by: zhenwei-intel <[email protected]> * [Metrics] Enable sleep state metric outside of dev mode (vllm-project#27867) Signed-off-by: Mark McLoughlin <[email protected]> * [Bug] Batch invariant: Fix flash attn MLA `RuntimeError: scheduler_metadata must have shape (metadata_size)` (vllm-project#27884) * [CPU]Improve dynamic 4bit moe performance (vllm-project#27240) Signed-off-by: Zhang Xiangze <[email protected]> * [CI/Build] Update LM Eval Version in AMD CI (vllm-project#27944) Signed-off-by: zhewenli <[email protected]> * [KV Connector] Make KVCacheConfig an explicit constructor argument (vllm-project#27887) Signed-off-by: Mark McLoughlin <[email protected]> * [Model] fix ernie45 reasoning_parser (vllm-project#27973) Signed-off-by: wangyafeng <[email protected]> * [CI/Build] Fix OpenAI API correctness on AMD CI (vllm-project#28022) Signed-off-by: zhewenli <[email protected]> * [BugFix][Performance] Restore flashinfer autotuning for all scenarios (vllm-project#27904) * Support worker reinitialization after hard pause; add task queue in FaultHandler to ensure sequential task execution Signed-off-by: fangyuchu <[email protected]> * resolve conflicts Signed-off-by: w00689259 <[email protected]> * resolve conflicts Signed-off-by: w00689259 <[email protected]> * resolve conflicts Signed-off-by: w00689259 <[email protected]> * Load tuned fused_moe_lora shrink and expand kernel configs separately (vllm-project#27435) Signed-off-by: Yu Gong <[email protected]> Co-authored-by: Jee Jee Li <[email protected]> * resolve conflicts Signed-off-by: w00689259 <[email protected]> * resolve conflicts Signed-off-by: w00689259 <[email protected]> * resolve conflicts Signed-off-by: w00689259 <[email protected]> * Support using Int4PreshuffledTensor after loading (vllm-project#26066) Signed-off-by: Jerry Zhang <[email protected]> * [Core] Enable StatLogger in LLMEngine (vllm-project#28020) Signed-off-by: Zhuohan Li <[email protected]> * [Model][Bugfix] fix pipeline parallelism support for NemotronH (vllm-project#27968) Signed-off-by: Tomer Asida <[email protected]> * [Model] add optimal triton fused moe configs for NemotronH MoE (vllm-project#27967) Signed-off-by: Tomer Asida <[email protected]> * [Kernels] Isolate modular kernel code from FusedMoEMethodBase subclasses. (vllm-project#27123) * [BugFix] Fix incorrect preallocated sampled_token_ids tensor size (vllm-project#28025) Signed-off-by: Nick Hill <[email protected]> * [Perf] SM100 - add swap AB optimization to CUTLASS FP8 GEMM (vllm-project#27284) Signed-off-by: Faqin Zhong <[email protected]> Co-authored-by: Faqin Zhong <[email protected]> Co-authored-by: Michael Goin <[email protected]> * [PERF] Decouple projections from GDN custom op (vllm-project#27512) Signed-off-by: Vadim Gimpelson <[email protected]> * [model] Add support for openPangu_Ultra_MoE (vllm-project#27521) Signed-off-by: yuantao <[email protected]> Signed-off-by: yt0428 <[email protected]> Co-authored-by: Jee Jee Li <[email protected]> * [PerfFix] Avoid separate thread for MP executor shm spin (vllm-project#28012) Signed-off-by: Nick Hill <[email protected]> * [AsyncScheduling] Don't schedule past request max_tokens (vllm-project#27922) Signed-off-by: Nick Hill <[email protected]> * Remove deprecated `--rope-scaling` and `--rope-theta` (vllm-project#28006) Signed-off-by: Harry Mellor <[email protected]> * [ROCm][Perf] New design on ROCm AITER MHA backend Implementation (vllm-project#25763) Signed-off-by: ganyi <[email protected]> * Added disable rule to track files under benchmarks/lib (vllm-project#28048) Signed-off-by: Nadav Kluger <[email protected]> * [Multimodal] Make MediaConnector extensible. (vllm-project#27759) Signed-off-by: Chenheli Hua <[email protected]> * [ROCm] gemm_a16w16 upstreaming (vllm-project#26969) Signed-off-by: Aleksandr Malyshev <[email protected]> Co-authored-by: Aleksandr Malyshev <[email protected]> * Revert "[PERF] Decouple projections from GDN custom op" (vllm-project#28080) Signed-off-by: Vadim Gimpelson <[email protected]> * add engine core ut Signed-off-by: w00689259 <[email protected]> * add engine core ut Signed-off-by: w00689259 <[email protected]> * [Qwen3-Next] MOE configs for A100-SXM4-80GB TP4 TP8 (vllm-project#27740) * [XPU] Add gpt-oss model support for Intel GPU (vllm-project#27786) Signed-off-by: Kunshang Ji <[email protected]> * [CI/Build] Enable some fixed tests in AMD CI (vllm-project#28078) Signed-off-by: zhewenli <[email protected]> * [V0 deprecation] Remove VLLM_USE_V1 usage in most modules (vllm-project#27955) Signed-off-by: wangxiyuan <[email protected]> * [Bugfix] Fix encoder-only model support for transformers backend (vllm-project#28021) Signed-off-by: Isotr0py <[email protected]> Signed-off-by: Harry Mellor <[email protected]> Co-authored-by: Harry Mellor <[email protected]> * [BugFix] Fix DCP Assert (AssertionError: DCP not support reorder_batch_threshold > 1 now.) (vllm-project#28100) Signed-off-by: Lucas Wilkinson <[email protected]> * [Model, Core] Support Granite Speech & LoRA for STT (vllm-project#24455) * [Refactor] Lazy-loaded reasoning_parser (vllm-project#28092) Signed-off-by: chaunceyjiang <[email protected]> * [Refactor] to simplify and extract the shared logic between chat completion and responses (vllm-project#27961) Signed-off-by: chaunceyjiang <[email protected]> * [bugfix] fix wrong `dcp_local_seq_lens` calc (vllm-project#27518) Signed-off-by: Qiu <[email protected]> * [Hybrid allocator + kv connector] revert connector test changes related to hybrid allocator (vllm-project#28011) Signed-off-by: KuntaiDu <[email protected]> * [Misc] fix import error for DeepSeekR1ReasoningParser (vllm-project#28114) Signed-off-by: chaunceyjiang <[email protected]> * Fix excessive logging noise by reducing the log level of the MinimaxM2ToolParser import success message (vllm-project#27635) Signed-off-by: minatoaquaMK2 <[email protected]> * Bugfix: Cutlass FP8 FusedMoE bad scaling factors (vllm-project#27255) Signed-off-by: Amir Klein <[email protected]> Co-authored-by: Tyler Michael Smith <[email protected]> Co-authored-by: Michael Goin <[email protected]> * [Graph Partition][Cache] Use inductor partition ops config (vllm-project#27702) Signed-off-by: Boyuan Feng <[email protected]> * [XPU] Enable custom routing functions in IPEX for Llama4 (vllm-project#28004) Signed-off-by: frost-intel <[email protected]> * add kimi reasoning parser (vllm-project#28128) Signed-off-by: wangzhengtao <[email protected]> Co-authored-by: wangzhengtao <[email protected]> * [DCP] check return_lse for all layers in dcp (vllm-project#27929) Signed-off-by: Chen Zhang <[email protected]> * [BugFix] Support EP/DP + EPLB with MTP (vllm-project#25311) Signed-off-by: ilmarkov <[email protected]> Signed-off-by: Sage Moore <[email protected]> Co-authored-by: Sage Moore <[email protected]> Co-authored-by: Tyler Michael Smith <[email protected]> Co-authored-by: Lucas Wilkinson <[email protected]> * Enabling cooperative multi-gpu tests on multi-gpu nodes (vllm-project#27986) Signed-off-by: Alexei V. Ivanov <[email protected]> * [ROCm][MLA] Support block-size > 1 for AITER MLA backend (vllm-project#27224) Signed-off-by: ganyi <[email protected]> Co-authored-by: wuhuikx <[email protected]> * [Bugfix] Validate custom logits processor xargs for online serving (vllm-project#27560) Signed-off-by: Isotr0py <[email protected]> * [misc] add vLLM Beijing Meetup (vllm-project#28127) Signed-off-by: Jiaju Zhang <[email protected]> * [Kernel] Fuse computation of g and beta for Gated Delta Net (vllm-project#28095) Signed-off-by: zjy0516 <[email protected]> * [Core] add support for reasoning parser plugins (vllm-project#28075) Signed-off-by: walter beller-morales <[email protected]> * [Bugfix] vLLM should check Inductor config for compile cache enablement status (vllm-project#27637) Signed-off-by: Yanan Cao <[email protected]> * [FlashInfer] Avoid FlashInfer block_size 16 + head_size 256 on blackwell (vllm-project#27994) Signed-off-by: Chen Zhang <[email protected]> * [CI]: Add LMCacheConnector Unit Tests (vllm-project#27852) Signed-off-by: Samuel Shen <[email protected]> Co-authored-by: Samuel Shen <[email protected]> Co-authored-by: Yihua Cheng <[email protected]> * [Feature] Extend batch invariant torch.compile to B200 (vllm-project#27856) Signed-off-by: PaulZhang12 <[email protected]> * [Bugfix] Fix Qwen3-Reranker-8B load (vllm-project#28117) Signed-off-by: wang.yuqi <[email protected]> * [Docs] Clean up README_TUNING.md (vllm-project#28088) Signed-off-by: windsonsea <[email protected]> * [Hardware][IBM Z] Optimize s390x Dockerfile (vllm-project#28023) Signed-off-by: Rehan Khan <[email protected]> * [Chore] Remove Nemotron-Nano-VL config copy (vllm-project#28126) Signed-off-by: Isotr0py <[email protected]> * [Docs] Add guide to debugging vLLM-torch.compile integration (vllm-project#28094) Signed-off-by: Richard Zou <[email protected]> * [Feature]: Add corrupted request metric to V1 metrics system. (vllm-project#27306) Signed-off-by: atalhens <[email protected]> * [CI/Build] Update checking logic in cutlass_group_gemm_supported (vllm-project#27948) Signed-off-by: zhewenli <[email protected]> * [CI/Build] Fix `test_defaults_with_usage_context` in AMD CI (vllm-project#27926) Signed-off-by: zhewenli <[email protected]> * [Core][Hybrid allocator + connector 2/n] Unify `remove_skipped_blocks` by `get_last_useful_token` (vllm-project#25431) Signed-off-by: KuntaiDu <[email protected]> * [Debugging] Add annotation for easier trace analysis (vllm-project#22496) * [PERF] Decouple projections from GDN custom op. Attempt 2 (vllm-project#28083) Signed-off-by: Vadim Gimpelson <[email protected]> * [Bug] Fix cpu disable shared_experts `VLLM_DISABLE_SHARED_EXPERTS_STREAM` (vllm-project#28157) Signed-off-by: yewentao256 <[email protected]> * [Bug] Fix env string `"0"` same to `True` (vllm-project#28159) Signed-off-by: yewentao256 <[email protected]> * Ensure WorkerGuard command execution returns result; fix missing set_device when TP>1 Signed-off-by: fangyuchu <[email protected]> * [Feature] Enable TP + EP `shared_experts` overlap with router, 3.7% E2E performance improvement (vllm-project#28164) Signed-off-by: yewentao256 <[email protected]> * [CI Failure] `nm-testing/Qwen2-0.5B-Instruct-FP8-SkipQKV` was removed from HF. Skip it in tests (vllm-project#28170) Signed-off-by: Vadim Gimpelson <[email protected]> * [Misc] Remove the duplicate code (vllm-project#28111) Signed-off-by: chaunceyjiang <[email protected]> * rename& format logger Signed-off-by: w00689259 <[email protected]> * rename& format logger Signed-off-by: w00689259 <[email protected]> * feat(nccl): enable non-blocking NCCL communicators to support ncclCommAbort Signed-off-by: fangyuchu <[email protected]> --------- Signed-off-by: w00689259 <[email protected]> Signed-off-by: a798347923 <[email protected]> Signed-off-by: fangyuchu <[email protected]> Signed-off-by: zWaNg3 <[email protected]> Signed-off-by: a798347923 <[email protected]> Signed-off-by: Sungyoon Jeong <[email protected]> Signed-off-by: Thomas Parnell <[email protected]> Signed-off-by: Rémi Delacourt <[email protected]> Signed-off-by: Rémi Delacourt <[email protected]> Signed-off-by: remi <[email protected]> Signed-off-by: Kunshang Ji <[email protected]> Signed-off-by: Misha Efimov <[email protected]> Signed-off-by: Jee Jee Li <[email protected]> Signed-off-by: zhangyue <[email protected]> Signed-off-by: Roger Wang <[email protected]> Signed-off-by: Isotr0py <[email protected]> Signed-off-by: zhangyue66 <[email protected]> Signed-off-by: gnovack <[email protected]> Signed-off-by: Peter Schuurman <[email protected]> Signed-off-by: ahao-anyscale <[email protected]> Signed-off-by: Lucas Wilkinson <[email protected]> Signed-off-by: Sophie du Couédic <[email protected]> Signed-off-by: Lucas Kabela <[email protected]> Signed-off-by: Matthew Bonanni <[email protected]> Signed-off-by: Andy Xie <[email protected]> Signed-off-by: Tyler Michael Smith <[email protected]> Signed-off-by: Qiliang Cui <[email protected]> Signed-off-by: vllmellm <[email protected]> Signed-off-by: li2haipeng <[email protected]> Signed-off-by: Haipeng Li <[email protected]> Signed-off-by: chaunceyjiang <[email protected]> Signed-off-by: zhenwei-intel <[email protected]> Signed-off-by: Mark McLoughlin <[email protected]> Signed-off-by: Zhang Xiangze <[email protected]> Signed-off-by: zhewenli <[email protected]> Signed-off-by: wangyafeng <[email protected]> Signed-off-by: Yu Gong <[email protected]> Signed-off-by: Jerry Zhang <[email protected]> Signed-off-by: Zhuohan Li <[email protected]> Signed-off-by: Tomer Asida <[email protected]> Signed-off-by: Nick Hill <[email protected]> Signed-off-by: Faqin Zhong <[email protected]> Signed-off-by: Vadim Gimpelson <[email protected]> Signed-off-by: yuantao <[email protected]> Signed-off-by: yt0428 <[email protected]> Signed-off-by: Harry Mellor <[email protected]> Signed-off-by: ganyi <[email protected]> Signed-off-by: Nadav Kluger <[email protected]> Signed-off-by: Chenheli Hua <[email protected]> Signed-off-by: Aleksandr Malyshev <[email protected]> Signed-off-by: wangxiyuan <[email protected]> Signed-off-by: Qiu <[email protected]> Signed-off-by: KuntaiDu <[email protected]> Signed-off-by: minatoaquaMK2 <[email protected]> Signed-off-by: Amir Klein <[email protected]> Signed-off-by: Boyuan Feng <[email protected]> Signed-off-by: frost-intel <[email protected]> Signed-off-by: wangzhengtao <[email protected]> Signed-off-by: Chen Zhang <[email protected]> Signed-off-by: ilmarkov <[email protected]> Signed-off-by: Sage Moore <[email protected]> Signed-off-by: Alexei V. Ivanov <[email protected]> Signed-off-by: Jiaju Zhang <[email protected]> Signed-off-by: zjy0516 <[email protected]> Signed-off-by: walter beller-morales <[email protected]> Signed-off-by: Yanan Cao <[email protected]> Signed-off-by: Samuel Shen <[email protected]> Signed-off-by: PaulZhang12 <[email protected]> Signed-off-by: wang.yuqi <[email protected]> Signed-off-by: windsonsea <[email protected]> Signed-off-by: Rehan Khan <[email protected]> Signed-off-by: Richard Zou <[email protected]> Signed-off-by: atalhens <[email protected]> Signed-off-by: yewentao256 <[email protected]> Co-authored-by: fangyuchu <[email protected]> Co-authored-by: a798347923 <[email protected]> Co-authored-by: w00689259 <[email protected]> Co-authored-by: fangyuchu <[email protected]> Co-authored-by: TianZhuo <[email protected]> Co-authored-by: a798347923 <[email protected]> Co-authored-by: Sungyoon Jeong <[email protected]> Co-authored-by: Chauncey <[email protected]> Co-authored-by: Thomas Parnell <[email protected]> Co-authored-by: Rémi Delacourt <[email protected]> Co-authored-by: Benjamin Chislett <[email protected]> Co-authored-by: Kunshang Ji <[email protected]> Co-authored-by: Misha Efimov <[email protected]> Co-authored-by: Jee Jee Li <[email protected]> Co-authored-by: zhang-prog <[email protected]> Co-authored-by: Roger Wang <[email protected]> Co-authored-by: Isotr0py <[email protected]> Co-authored-by: gnovack <[email protected]> Co-authored-by: pwschuurman <[email protected]> Co-authored-by: ahao-anyscale <[email protected]> Co-authored-by: Lucas Wilkinson <[email protected]> Co-authored-by: Aurick Qiao <[email protected]> Co-authored-by: Aurick Qiao <[email protected]> Co-authored-by: Sophie du Couédic <[email protected]> Co-authored-by: Lucas Kabela <[email protected]> Co-authored-by: Matthew Bonanni <[email protected]> Co-authored-by: Ning Xie <[email protected]> Co-authored-by: Hank_ <[email protected]> Co-authored-by: Tyler Michael Smith <[email protected]> Co-authored-by: QiliangCui <[email protected]> Co-authored-by: vllmellm <[email protected]> Co-authored-by: li2haipeng <[email protected]> Co-authored-by: liuzhenwei <[email protected]> Co-authored-by: Mark McLoughlin <[email protected]> Co-authored-by: Wentao Ye <[email protected]> Co-authored-by: xiangze-arm <[email protected]> Co-authored-by: Zhewen Li <[email protected]> Co-authored-by: CSWYF3634076 <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: yugong333 <[email protected]> Co-authored-by: Jerry Zhang <[email protected]> Co-authored-by: Zhuohan Li <[email protected]> Co-authored-by: tomeras91 <[email protected]> Co-authored-by: bnellnm <[email protected]> Co-authored-by: Nick Hill <[email protected]> Co-authored-by: lyrisz <[email protected]> Co-authored-by: Faqin Zhong <[email protected]> Co-authored-by: Michael Goin <[email protected]> Co-authored-by: Vadim Gimpelson <[email protected]> Co-authored-by: yt0428 <[email protected]> Co-authored-by: Harry Mellor <[email protected]> Co-authored-by: Pleaplusone <[email protected]> Co-authored-by: nadavkluger <[email protected]> Co-authored-by: Chenheli Hua <[email protected]> Co-authored-by: Aleksandr Malyshev <[email protected]> Co-authored-by: Aleksandr Malyshev <[email protected]> Co-authored-by: tou <[email protected]> Co-authored-by: wangxiyuan <[email protected]> Co-authored-by: Alex Brooks <[email protected]> Co-authored-by: Qiu <[email protected]> Co-authored-by: Kuntai Du <[email protected]> Co-authored-by: Eric Yue <[email protected]> Co-authored-by: amirkl94 <[email protected]> Co-authored-by: Boyuan Feng <[email protected]> Co-authored-by: Frost Mitchell <[email protected]> Co-authored-by: bigmoyan <[email protected]> Co-authored-by: wangzhengtao <[email protected]> Co-authored-by: Chen Zhang <[email protected]> Co-authored-by: Ilya Markov <[email protected]> Co-authored-by: Sage Moore <[email protected]> Co-authored-by: Alexei-V-Ivanov-AMD <[email protected]> Co-authored-by: wuhuikx <[email protected]> Co-authored-by: Jiaju Zhang <[email protected]> Co-authored-by: Jiangyun Zhu <[email protected]> Co-authored-by: Walter Beller-Morales <[email protected]> Co-authored-by: gmagogsfm <[email protected]> Co-authored-by: Samuel Shen <[email protected]> Co-authored-by: Samuel Shen <[email protected]> Co-authored-by: Yihua Cheng <[email protected]> Co-authored-by: Paul Zhang <[email protected]> Co-authored-by: wang.yuqi <[email protected]> Co-authored-by: Michael Yao <[email protected]> Co-authored-by: R3hankhan <[email protected]> Co-authored-by: Richard Zou <[email protected]> Co-authored-by: Snehlata <[email protected]> Co-authored-by: Dayeol Lee <[email protected]>
ZhengHongming888
pushed a commit
to ZhengHongming888/vllm
that referenced
this pull request
Nov 8, 2025
…nt status (vllm-project#27637) Signed-off-by: Yanan Cao <[email protected]>
xuebwang-amd
pushed a commit
to xuebwang-amd/vllm
that referenced
this pull request
Nov 13, 2025
…nt status (vllm-project#27637) Signed-off-by: Yanan Cao <[email protected]> Signed-off-by: xuebwang-amd <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
Summary:
vLLM should not assume compile cache is enabled just because VLLM_DISABLE_COMPILE_CACHE=0. Users may use TORCHINDUCTOR_FORCE_DISABLE_CACHES=1 to disable compile cache at PyTorch level, effectively making it impossible for vLLM compile cache to function.
This PR adds necessary checks to ensure vLLM considers Torch Inductor compile cache enablement status.
Signed off by: Yanan Cao [email protected]