diff --git a/docs/README.md b/docs/README.md
index 29afab90a..67c3ad194 100644
--- a/docs/README.md
+++ b/docs/README.md
@@ -1,4 +1,5 @@
 # SGLang Documentation
+This is the documentation repository for SGLang. It is auto-generated from https://github.com/sgl-project/sglang/tree/main/docs.
 
 ## Build the documentation website
 
diff --git a/python/sglang/bench_serving.py b/python/sglang/bench_serving.py
index a1e43e4cd..9662f61a8 100644
--- a/python/sglang/bench_serving.py
+++ b/python/sglang/bench_serving.py
@@ -407,7 +407,7 @@ async def async_request_profile(api_url: str) -> RequestFuncOutput:
 
 
 def get_model(pretrained_model_name_or_path: str) -> str:
-    if os.getenv("SGLANG_USE_MODELSCOPE", "False").lower() == "true":
+    if os.getenv("SGLANG_USE_MODELSCOPE", "false").lower() == "true":
         import huggingface_hub.constants
         from modelscope import snapshot_download
 
diff --git a/python/sglang/srt/layers/attention/flashinfer_backend.py b/python/sglang/srt/layers/attention/flashinfer_backend.py
index b72134a56..fb391e627 100644
--- a/python/sglang/srt/layers/attention/flashinfer_backend.py
+++ b/python/sglang/srt/layers/attention/flashinfer_backend.py
@@ -7,6 +7,7 @@ FlashInfer is faster and Triton is easier to customize.
 Each backend supports two operators: extend (i.e. prefill with cached prefix) and decode.
 """
 
+import os
 from enum import Enum, auto
 from typing import TYPE_CHECKING, List
 
@@ -45,13 +46,19 @@ class FlashInferAttnBackend(AttentionBackend):
         super().__init__()
 
         # Parse constants
-        if not _grouped_size_compiled_for_decode_kernels(
-            model_runner.model_config.num_attention_heads // model_runner.tp_size,
-            model_runner.model_config.get_num_kv_heads(model_runner.tp_size),
-        ):
-            self.decode_use_tensor_cores = True
+        if "SGLANG_FLASHINFER_USE_TENSOR_CORE" in os.environ:
+            self.decode_use_tensor_cores = (
+                os.environ["SGLANG_FLASHINFER_USE_TENSOR_CORE"].lower() == "true"
+            )
         else:
-            self.decode_use_tensor_cores = False
+            if not _grouped_size_compiled_for_decode_kernels(
+                model_runner.model_config.num_attention_heads // model_runner.tp_size,
+                model_runner.model_config.get_num_kv_heads(model_runner.tp_size),
+            ):
+                self.decode_use_tensor_cores = True
+            else:
+                self.decode_use_tensor_cores = False
+
         self.max_context_len = model_runner.model_config.context_len
 
         assert not (
diff --git a/python/sglang/srt/managers/scheduler.py b/python/sglang/srt/managers/scheduler.py
index 796164297..9ca9b7c64 100644
--- a/python/sglang/srt/managers/scheduler.py
+++ b/python/sglang/srt/managers/scheduler.py
@@ -81,7 +81,7 @@ from sglang.utils import get_exception_traceback
 logger = logging.getLogger(__name__)
 
 # Test retract decode
-test_retract = os.getenv("SGLANG_TEST_RETRACT", "false") == "true"
+test_retract = os.getenv("SGLANG_TEST_RETRACT", "false").lower() == "true"
 
 
 class Scheduler:
diff --git a/python/sglang/srt/utils.py b/python/sglang/srt/utils.py
index 8bb6a5830..e947d1a92 100644
--- a/python/sglang/srt/utils.py
+++ b/python/sglang/srt/utils.py
@@ -930,7 +930,7 @@ def get_nvgpu_memory_capacity():
 
 def crash_on_warnings():
     # Crash on warning if we are running CI tests
-    return os.getenv("SGLANG_IS_IN_CI", "false") == "true"
+    return os.getenv("SGLANG_IS_IN_CI", "false").lower() == "true"
 
 
 def get_device_name(device_id: int = 0) -> str:
diff --git a/python/sglang/test/test_utils.py b/python/sglang/test/test_utils.py
index d673d59ff..be1755bd3 100644
--- a/python/sglang/test/test_utils.py
+++ b/python/sglang/test/test_utils.py
@@ -44,7 +44,7 @@ DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_QUANT_TP1 = "hugging-quants/Meta-Llama-3.1-8
 
 def is_in_ci():
     """Return whether it is in CI runner."""
-    return os.getenv("SGLANG_IS_IN_CI", "false") == "true"
+    return os.getenv("SGLANG_IS_IN_CI", "false").lower() == "true"
 
 
 if is_in_ci():