From 8673e85e6c312f4f7c618a5f05c0a9eef4dd6886 Mon Sep 17 00:00:00 2001
From: Lianmin Zheng <lianminzheng@gmail.com>
Date: Sun, 5 Jul 2026 12:45:24 -0700
Subject: [PATCH] Remove `# fmt: off` from environ.py Envs class (#30153)

---
 python/sglang/srt/environ.py | 20 +++++++++++---------
 1 file changed, 11 insertions(+), 9 deletions(-)

diff --git a/python/sglang/srt/environ.py b/python/sglang/srt/environ.py
index 6cf331007..72bae8358 100644
--- a/python/sglang/srt/environ.py
+++ b/python/sglang/srt/environ.py
@@ -216,7 +216,6 @@ class ToolStrictLevel(IntEnum):
 
 
 class Envs:
-    # fmt: off
 
     # Model & File Download
     SGLANG_USE_MODELSCOPE = EnvBool(False)
@@ -320,7 +319,7 @@ class Envs:
     # Scheduler: memory leak test
     SGLANG_TEST_RETRACT = EnvBool(False)
     SGLANG_TEST_RETRACT_INTERVAL = EnvInt(3)
-    SGLANG_TEST_RETRACT_NO_PREFILL_BS = EnvInt(2 ** 31)
+    SGLANG_TEST_RETRACT_NO_PREFILL_BS = EnvInt(2**31)
     # Scheduler: force lazy extra_buffer prealloc to fail at decode boundaries
     SGLANG_TEST_MAMBA_LAZY_ALLOC_FAIL = EnvBool(False)
     # KL tests: skip the cache-hit count assertion (e.g. when alloc failure reduces hits)
@@ -362,7 +361,8 @@ class Envs:
     SGLANG_DISAGGREGATION_FORCE_QUERY_PREFILL_DP_RANK = EnvBool(False)
 
     # Scheduler: others:
-    SGLANG_EMPTY_CACHE_INTERVAL = EnvFloat(-1)  # in seconds. Set if you observe high memory accumulation over a long serving period.
+    # in seconds. Set if you observe high memory accumulation over a long serving period.
+    SGLANG_EMPTY_CACHE_INTERVAL = EnvFloat(-1)
     SGLANG_DISABLE_CONSECUTIVE_PREFILL_OVERLAP = EnvBool(False)
     # Force-enable the WAR (write-after-read) barrier for the overlap scheduler
     # even when is_cuda() is False (e.g. AMD/ROCm). On CUDA the barrier is
@@ -566,7 +566,7 @@ class Envs:
     # fine-grained opt switch reads False, keeping non-experimental paths byte-identical.
     SGLANG_EXPERIMENTAL_LORA_OPTI = EnvBool(False)
     # Quantize x to int8 in the dispatch operator
-    DEEP_NORMAL_MODE_USE_INT8_QUANT = EnvBool(False) # This argument is deprecated
+    DEEP_NORMAL_MODE_USE_INT8_QUANT = EnvBool(False)  # This argument is deprecated
     SGLANG_NPU_FUSED_MOE_MODE = EnvInt(1)
 
     # MTHREADS & MUSA
@@ -638,7 +638,7 @@ class Envs:
     SGLANG_MAX_KV_CHUNK_CAPACITY = EnvInt(128 * 1024)
 
     # DeepEP
-    SGLANG_DEEPEP_BF16_DISPATCH = EnvBool(False) # This argument is deprecated
+    SGLANG_DEEPEP_BF16_DISPATCH = EnvBool(False)  # This argument is deprecated
     SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK = EnvInt(128)
     SGLANG_DEEPEP_LL_COMBINE_SEND_NUM_SMS = EnvInt(32)
     SGLANG_BLACKWELL_OVERLAP_SHARED_EXPERTS_OUTSIDE_SBO = EnvBool(False)
@@ -651,7 +651,9 @@ class Envs:
     SGLANG_NIXL_EP_NUM_MAX_DISPATCH_TOKENS_PER_RANK = EnvInt(128)
 
     # DSA Backend (canonical names; fall back to SGLANG_NSA_* with deprecation warning)
-    SGLANG_DSA_FUSE_TOPK = EnvBoolWithAlias(True, deprecated_name="SGLANG_NSA_FUSE_TOPK")
+    SGLANG_DSA_FUSE_TOPK = EnvBoolWithAlias(
+        True, deprecated_name="SGLANG_NSA_FUSE_TOPK"
+    )
     SGLANG_DSA_TOPK_FLASHINFER_DETERMINISTIC = EnvBool(False)
     SGLANG_DSA_TOPK_FLASHINFER_TIE_BREAK = EnvStr(None)
     SGLANG_DSA_ENABLE_MTP_PRECOMPUTE_METADATA = EnvBoolWithAlias(
@@ -745,7 +747,6 @@ class Envs:
     # preserve the user's original tokens to avoid retokenization drift.
     SGLANG_MM_AVOID_RETOKENIZE = EnvBool(True)
 
-
     # VLM Item CUDA IPC Transport
     SGLANG_USE_CUDA_IPC_TRANSPORT = EnvBool(False)
     SGLANG_USE_IPC_POOL_HANDLE_CACHE = EnvBool(False)
@@ -784,7 +785,9 @@ class Envs:
     SGLANG_NGRAM_FORCE_GREEDY_VERIFY = EnvBool(False)
 
     # Warmup
-    SGLANG_WARMUP_TIMEOUT = EnvFloat(-1) # in seconds. If a warmup forward batch takes longer than this, the server will crash to prevent hanging. Recommend to increase warmup timeout to 1800 to accommodate some kernel JIT precache e.g. deep gemm
+    # in seconds. If a warmup forward batch takes longer than this, the server will crash to prevent hanging.
+    # Recommend to increase warmup timeout to 1800 to accommodate some kernel JIT precache e.g. deep gemm
+    SGLANG_WARMUP_TIMEOUT = EnvFloat(-1)
 
     # HTTP Server
     SGLANG_TIMEOUT_KEEP_ALIVE = EnvInt(5)
@@ -958,7 +961,6 @@ class Envs:
 
     # Aiter
     SGLANG_USE_AITER_FP8_PER_TOKEN = EnvBool(False)
-    # fmt: on
 
     # EPD
     SGLANG_ENCODER_RECV_TIMEOUT = EnvFloat(180.0)