Import flash_mla from sgl-kernel (#12135)

.github/workflows/pr-test.yml

@@ -747,7 +747,7 @@ jobs:
 
       - name: Install dependencies
         run: |
-          CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} RUN_DEEPSEEK_V32=1 bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} bash scripts/ci/ci_install_dependency.sh
 
       - name: Run test
         timeout-minutes: 20

docker/Dockerfile

@@ -8,8 +8,6 @@ ARG GRACE_BLACKWELL=0
 
 ARG GRACE_BLACKWELL_DEEPEP_BRANCH=gb200_blog_part_2
 ARG DEEPEP_COMMIT=9af0e0d0e74f3577af1979c9b9e1ac2cad0104ee
-ARG FLASHMLA_COMMIT=1408756a88e52a25196b759eaf8db89d2b51b5a1
-
 ARG TRITON_LANG_COMMIT=4caa0328bf8df64896dd5f6fb9df41b0eb2e750a
 
 ARG SGL_KERNEL_VERSION=0.3.16.post4
@@ -179,17 +177,6 @@ RUN cd /sgl-workspace/DeepEP && \
     fi && \
     NVSHMEM_DIR=${NVSHMEM_DIR} TORCH_CUDA_ARCH_LIST="${CHOSEN_TORCH_CUDA_ARCH_LIST}" pip install --no-build-isolation .
 
-# Install flashmla
-RUN if [ "$CUDA_VERSION" != "13.0.1" ]; then \
-    git clone https://github.com/deepseek-ai/FlashMLA.git flash-mla && \
-      cd flash-mla && \
-      git checkout ${FLASHMLA_COMMIT} && \
-      git submodule update --init --recursive && \
-      if [ "$CUDA_VERSION" = "12.6.1" ]; then \
-        export FLASH_MLA_DISABLE_SM100=1; \
-      fi && \
-      pip install --no-build-isolation -v . ; \
-    fi
 
 # In order to use flashinfer_cutedsl without IMA for WideEP configs we must install
 # latest flashinfer_cutedsl. Once 0.4.3 is officially released, remove this

docs/basic_usage/deepseek_v32.md

@@ -27,13 +27,7 @@ docker pull lmsysorg/sglang:dsv32-a3
 git clone https://github.com/sgl-project/sglang
 cd sglang
 pip3 install pip --upgrade
-pip3 install -e "python[all]"
+pip3 install -e "python"
-
-# Install flash_mla
-git clone https://github.com/deepseek-ai/FlashMLA.git flash-mla
-cd flash-mla
-git submodule update --init --recursive
-pip install -v .
 ```
 ## Launch DeepSeek V3.2 with SGLang
 

python/sglang/srt/layers/attention/flashmla_backend.py

@@ -9,7 +9,7 @@ from typing import TYPE_CHECKING, Callable, Optional, Tuple, Union
 
 import torch
 import triton
-from flash_mla import flash_mla_with_kvcache, get_mla_metadata
+from sgl_kernel.flash_mla import flash_mla_with_kvcache, get_mla_metadata
 
 from sglang.srt.layers.attention.flashinfer_mla_backend import FlashInferMLAAttnBackend
 from sglang.srt.layers.attention.utils import create_flashmla_kv_indices_triton

python/sglang/srt/layers/attention/nsa_backend.py

@@ -1098,7 +1098,7 @@ class NativeSparseAttnBackend(AttentionBackend):
         page_table_1: torch.Tensor,
         sm_scale: float,
     ) -> torch.Tensor:
-        from flash_mla import flash_mla_sparse_fwd
+        from sgl_kernel.flash_mla import flash_mla_sparse_fwd
 
         o, _, _ = flash_mla_sparse_fwd(
             q=q_all,
@@ -1119,7 +1119,7 @@ class NativeSparseAttnBackend(AttentionBackend):
         metadata: NSAMetadata,
         page_table_1,
     ) -> torch.Tensor:
-        from flash_mla import flash_mla_with_kvcache
+        from sgl_kernel.flash_mla import flash_mla_with_kvcache
 
         cache_seqlens = metadata.nsa_cache_seqlens_int32
 
@@ -1261,7 +1261,7 @@ class NativeSparseAttnBackend(AttentionBackend):
         )
 
     def _compute_flashmla_metadata(self, cache_seqlens: torch.Tensor, seq_len_q: int):
-        from flash_mla import get_mla_metadata
+        from sgl_kernel.flash_mla import get_mla_metadata
 
         flashmla_metadata, num_splits = get_mla_metadata(
             cache_seqlens=cache_seqlens,

scripts/ci/ci_install_dependency.sh

@@ -3,7 +3,6 @@
 set -euxo pipefail
 
 IS_BLACKWELL=${IS_BLACKWELL:-0}
-RUN_DEEPSEEK_V32=${RUN_DEEPSEEK_V32:-0}
 CU_VERSION="cu129"
 
 if [ "$CU_VERSION" = "cu130" ]; then
@@ -113,22 +112,6 @@ if [ "$IS_BLACKWELL" != "1" ]; then
     $PIP_CMD install xformers --index-url https://download.pytorch.org/whl/${CU_VERSION} --no-deps $PIP_INSTALL_SUFFIX
 fi
 
-# Install dependencies for deepseek-v3.2
-if [ "$RUN_DEEPSEEK_V32" = "1" ]; then
-    # Install flashmla
-    FLASHMLA_COMMIT="1408756a88e52a25196b759eaf8db89d2b51b5a1"
-    FLASH_MLA_DISABLE_SM100="0"
-    if [ "$IS_BLACKWELL" != "1" ]; then
-        FLASH_MLA_DISABLE_SM100="1"
-    fi
-    git clone https://github.com/deepseek-ai/FlashMLA.git flash-mla
-    cd flash-mla
-    git checkout ${FLASHMLA_COMMIT}
-    git submodule update --init --recursive
-    FLASH_MLA_DISABLE_SM100=${FLASH_MLA_DISABLE_SM100} $PIP_CMD install -v . $PIP_INSTALL_SUFFIX --no-build-isolation
-    cd ..
-fi
-
 # Show current packages
 $PIP_CMD list
 python3 -c "import torch; print(torch.version.cuda)"

test/srt/run_suite.py

@@ -82,7 +82,7 @@ suites = {
         TestFile("test_ebnf_constrained.py", 108),
         TestFile("test_eval_fp8_accuracy.py", 303),
         TestFile("test_fa3.py", 376),
-        # TestFile("test_flashmla.py", 352),
+        TestFile("test_flashmla.py", 352),
         TestFile("rotary_embedding/test_mrope.py", 300),
         TestFile("test_function_call_parser.py", 10),
         TestFile("test_fused_moe.py", 30),

test/srt/test_flashmla.py

@@ -11,6 +11,7 @@ import torch
 
 from sglang.srt.utils import kill_process_tree
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
+from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_TEST_MLA,
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -18,7 +19,7 @@ from sglang.test.test_utils import (
     CustomTestCase,
     is_in_ci,
     popen_launch_server,
-    run_bench_one_batch,
+    write_github_step_summary,
 )
 
 
@@ -31,7 +32,6 @@ class TestFlashMLAAttnBackend(unittest.TestCase):
         if torch.cuda.is_available() and torch.version.cuda:
             other_args.extend(
                 [
-                    "--enable-torch-compile",
                     "--cuda-graph-max-bs",
                     "2",
                     "--attention-backend",
@@ -65,24 +65,6 @@ class TestFlashMLAAttnBackend(unittest.TestCase):
         self.assertGreater(metrics["accuracy"], 0.60)
 
 
-class TestFlashMLAAttnLatency(unittest.TestCase):
-    def test_latency(self):
-        _, output_throughput, _ = run_bench_one_batch(
-            DEFAULT_MODEL_NAME_FOR_TEST_MLA,
-            [
-                "--attention-backend",
-                "flashmla",
-                "--enable-torch-compile",
-                "--cuda-graph-max-bs",
-                "16",
-                "--trust-remote-code",
-            ],
-        )
-
-        if is_in_ci():
-            self.assertGreater(output_throughput, 100)
-
-
 class TestFlashMLAMTP(CustomTestCase):
     @classmethod
     def setUpClass(cls):