From 681b9e6425fb8a8433eb8a599e0e126cbd76e8db Mon Sep 17 00:00:00 2001 From: Qiaolin Yu Date: Fri, 21 Nov 2025 10:37:41 -0800 Subject: [PATCH] Revert "enable csgmv automatically on cuda" (#13707) --- docs/references/environment_variables.md | 1 - python/sglang/srt/utils/common.py | 6 ------ 2 files changed, 7 deletions(-) diff --git a/docs/references/environment_variables.md b/docs/references/environment_variables.md index 6aea6d8b9..fda66255c 100644 --- a/docs/references/environment_variables.md +++ b/docs/references/environment_variables.md @@ -113,7 +113,6 @@ SGLang supports various environment variables that can be used to configure its | --- | --- | --- | | `SGLANG_WAIT_WEIGHTS_READY_TIMEOUT` | Timeout period for waiting on weights | `120` | | `SGLANG_DISABLE_OUTLINES_DISK_CACHE` | Disable Outlines disk cache | `true` | -| `SGLANG_USE_CUSTOM_TRITON_KERNEL_CACHE` | Use SGLang's custom Triton kernel cache implementation for lower overheads (automatically enabled on CUDA) | `false` | ## Function Calling / Tool Use diff --git a/python/sglang/srt/utils/common.py b/python/sglang/srt/utils/common.py index 05274e558..0ac752b93 100644 --- a/python/sglang/srt/utils/common.py +++ b/python/sglang/srt/utils/common.py @@ -3618,12 +3618,6 @@ def cached_triton_kernel(key_fn=None): """ def decorator(fn): - # Auto-enable the custom kernel cache for CUDA, where it is - # known to be compatible. - if is_cuda() and not envs.SGLANG_USE_CUSTOM_TRITON_KERNEL_CACHE.is_set(): - logger.debug("Detected platform CUDA, using custom triton kernel cache.") - return CachedKernel(fn, key_fn) - if envs.SGLANG_USE_CUSTOM_TRITON_KERNEL_CACHE.get(): logger.debug( f"{envs.SGLANG_USE_CUSTOM_TRITON_KERNEL_CACHE.name} = True. Using custom triton kernel cache."