调用vllm里custom all reduce

适配w8a8模型

See merge request OpenDAS/sglang!1

python/sglang/bench_serving.py

@@ -839,10 +839,12 @@ class BenchmarkMetrics:
     mean_ttft_ms: float
     median_ttft_ms: float
     std_ttft_ms: float
+    p95_ttft_ms: float
     p99_ttft_ms: float
     mean_tpot_ms: float
     median_tpot_ms: float
     std_tpot_ms: float
+    p95_tpot_ms: float
     p99_tpot_ms: float
     mean_itl_ms: float
     median_itl_ms: float
@@ -1665,10 +1667,12 @@ def calculate_metrics(
         * 1000,  # ttfts is empty if streaming is not supported by backend
         median_ttft_ms=np.median(ttfts or 0) * 1000,
         std_ttft_ms=np.std(ttfts or 0) * 1000,
+        p95_ttft_ms=np.percentile(ttfts or 0, 95) * 1000,
         p99_ttft_ms=np.percentile(ttfts or 0, 99) * 1000,
         mean_tpot_ms=np.mean(tpots or 0) * 1000,
         median_tpot_ms=np.median(tpots or 0) * 1000,
         std_tpot_ms=np.std(tpots or 0) * 1000,
+        p95_tpot_ms=np.percentile(tpots or 0, 95) * 1000,
         p99_tpot_ms=np.percentile(tpots or 0, 99) * 1000,
         mean_itl_ms=np.mean(itls or 0) * 1000,
         median_itl_ms=np.median(itls or 0) * 1000,
@@ -1974,6 +1978,12 @@ async def benchmark(
     print("{:<40} {:<10.2f}".format("Mean TTFT (ms):", metrics.mean_ttft_ms))
     print("{:<40} {:<10.2f}".format("Median TTFT (ms):", metrics.median_ttft_ms))
     print("{:<40} {:<10.2f}".format("P99 TTFT (ms):", metrics.p99_ttft_ms))
+    print("{:<40} {:<10.2f}".format("P95 TTFT (ms):", metrics.p95_ttft_ms))
+    print("{s:{c}^{n}}".format(s="Time per Output Token (excl. 1st token)", n=50, c="-"))
+    print("{:<40} {:<10.2f}".format("Mean TPOT (ms):", metrics.mean_tpot_ms))
+    print("{:<40} {:<10.2f}".format("Median TPOT (ms):", metrics.median_tpot_ms))
+    print("{:<40} {:<10.2f}".format("P99 TPOT (ms):", metrics.p99_tpot_ms))
+    print("{:<40} {:<10.2f}".format("P95 TPOT (ms):", metrics.p95_tpot_ms))
     print("{s:{c}^{n}}".format(s="Inter-Token Latency", n=50, c="-"))
     print("{:<40} {:<10.2f}".format("Mean ITL (ms):", metrics.mean_itl_ms))
     print("{:<40} {:<10.2f}".format("Median ITL (ms):", metrics.median_itl_ms))

python/sglang/srt/_custom_ops.py

@@ -22,9 +22,11 @@ use_vllm_custom_allreduce = get_bool_env_var(
 
 if not is_hpu():
     # ROCm does not use vllm custom allreduce
-    if use_vllm_custom_allreduce and not is_hip():
+    # if use_vllm_custom_allreduce and not is_hip():
+    if use_vllm_custom_allreduce:
         try:
             import vllm._C  # noqa: F401
+            print("[DEBUG] ✅ Using vLLM custom allreduce (vllm._C successfully imported)")
         except ImportError as e:
             logger.warning("Failed to import from vllm._C with %r", e)
     else:
@@ -34,9 +36,11 @@ if not is_hpu():
             logger.warning("Failed to import from custom_ar with %r", e)
 
 
-if not is_hip() and not is_npu():
+# if not is_hip() and not is_npu():
+if not is_npu():
     if use_vllm_custom_allreduce:
         custom_op = torch.ops._C_custom_ar
+        print("[DEBUG] ✅ custom_op = torch.ops._C_custom_ar (vLLM path active)")
     else:
         custom_op = sgl_kernel.allreduce
 

python/sglang/srt/configs/model_config.py

@@ -614,6 +614,8 @@ class ModelConfig:
             "petit_nvfp4",
             "quark",
             "mxfp4",
+            "slimquant_w4a8_marlin",
+            "w8a8_int8",
         ]
         optimized_quantization_methods = [
             "fp8",
@@ -633,6 +635,7 @@ class ModelConfig:
             "qoq",
             "w4afp8",
             "petit_nvfp4",
+            "slimquant_w4a8_marlin",
         ]
         compatible_quantization_methods = {
             "modelopt_fp4": ["modelopt"],

python/sglang/srt/distributed/device_communicators/custom_all_reduce.py

@@ -27,7 +27,8 @@ _is_hip = is_hip()
 
 
 try:
-    if ops.use_vllm_custom_allreduce and not _is_hip:
+    # if ops.use_vllm_custom_allreduce and not _is_hip:
+    if ops.use_vllm_custom_allreduce:
         # Use vLLM custom allreduce
         ops.meta_size()
     else:

python/sglang/srt/layers/layernorm.py

@@ -169,6 +169,14 @@ class RMSNorm(CustomOp):
             try:
                 output = torch.empty_like(x)
                 residual_out = torch.empty_like(x)
+                fused_add_rms_norm(
+                    x,
+                    residual,
+                    self.weight.data,
+                    self.variance_epsilon,
+                )
+                return x, residual
+            except TypeError:
                 fused_add_rms_norm(
                     output,
                     x,
@@ -178,14 +186,7 @@ class RMSNorm(CustomOp):
                     self.variance_epsilon,
                 )
                 return output, residual_out
-            except TypeError:
+                
-                fused_add_rms_norm(
-                    x,
-                    residual,
-                    self.weight.data,
-                    self.variance_epsilon,
-                )
-                return x, residual
 
         out = torch.empty_like(x)
         rms_norm(out, x, self.weight.data, self.variance_epsilon)

python/sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_kernels.py

@@ -14,9 +14,10 @@ from sglang.srt.layers.quantization.fp8_kernel import (
 )
 from sglang.srt.layers.quantization.int8_kernel import (
     per_token_group_quant_int8,
-    per_token_quant_int8,
+    # per_token_quant_int8,
     sglang_per_token_group_quant_int8,
 )
+from lmslim.layers.gemm.int8_utils import per_token_quant_int8
 from sglang.srt.utils import (
     cpu_has_amx_support,
     get_bool_env_var,

python/sglang/srt/layers/quantization/__init__.py

@@ -57,6 +57,7 @@ from sglang.srt.layers.quantization.qoq import QoQConfig
 from sglang.srt.layers.quantization.w4afp8 import W4AFp8Config
 from sglang.srt.layers.quantization.w8a8_fp8 import W8A8Fp8Config
 from sglang.srt.layers.quantization.w8a8_int8 import W8A8Int8Config
+from sglang.srt.layers.quantization.slimquant_w4a8_marlin import SlimQuantW4A8Int8MarlinConfig
 from sglang.srt.utils import is_cuda, is_hip, mxfp_supported
 
 _is_mxfp_supported = mxfp_supported()
@@ -83,6 +84,7 @@ BASE_QUANTIZATION_METHODS: Dict[str, Type[QuantizationConfig]] = {
     "w4afp8": W4AFp8Config,
     "petit_nvfp4": PetitNvFp4Config,
     "fbgemm_fp8": FBGEMMFp8Config,
+    "slimquant_w4a8_marlin":SlimQuantW4A8Int8MarlinConfig,
 }
 
 

python/sglang/srt/layers/quantization/slimquant_w4a8_marlin.py

@@ -1,6 +1,6 @@
 from typing import Any, Callable, Dict, List, Optional
-from sglang.srt.layers.moe.token_dispatcher.base import CombineInput
+# from sglang.srt.layers.moe.token_dispatcher.base import CombineInput
-from sglang.srt.layers.moe.token_dispatcher.standard import StandardCombineInput, StandardDispatchOutput
+
 import torch
 from sglang.srt import _custom_ops as ops
 from sglang.srt.utils import set_weight_attrs
@@ -218,8 +218,9 @@ class SlimQuantW4A8Int8MarlinMoEMethod:
     def apply(
         self,
         layer: torch.nn.Module,
-        dispatch_output: StandardDispatchOutput,
+        dispatch_output,
-    ) -> CombineInput:
+    ) :
+        from sglang.srt.layers.moe.token_dispatcher.standard import StandardCombineInput
         x = dispatch_output.hidden_states
         topk_output = dispatch_output.topk_output
         from sglang.srt.layers.moe.topk import apply_topk_weights_cpu
@@ -241,7 +242,7 @@ class SlimQuantW4A8Int8MarlinMoEMethod:
             use_int4_w4a8=True,
             per_channel_quant=True,
             activation=layer.moe_runner_config.activation,
-            expert_map=layer.expert_map_gpu,
+            # expert_map=layer.expert_map_gpu,
             apply_router_weight_on_input=self.moe_runner_config.apply_router_weight_on_input,
             global_num_experts=layer.moe_runner_config.num_experts,
             w1_scale=(layer.w13_weight_scale),

python/sglang/srt/layers/quantization/w4a8_utils.py

@@ -0,0 +1,92 @@
+import torch
+import numpy as np
+
+try:
+    from lightop import awq_marlin_repack_w4a8
+    use_lightop = False
+except Exception:
+    use_lightop = False
+
+def unpack_int8_to_int4(tensor_int8: torch.Tensor) -> torch.Tensor:
+    """
+    将[N, K//2]大小的torch.int8 Tensor，转换为[N, K]大小的torch.int32 Tensor。
+    每个int8包含两个int4，分别提取到int32的低4位，其余位为0。
+
+    Args:
+        tensor_int8 (torch.Tensor): 输入张量，形状为[N, K//2]，类型为torch.int8。
+
+    Returns:
+        torch.Tensor: 输出张量，形状为[N, K]，类型为torch.int32。
+    """
+    if tensor_int8.dtype != torch.int8:
+        raise ValueError("Input tensor must be of type torch.int8")
+
+    N, K_half = tensor_int8.shape
+    tensor_uint8 = tensor_int8.to(torch.uint8)
+    high4 = tensor_uint8 & 0x0F
+    low4 = (tensor_uint8 >> 4) & 0x0F
+    unpacked = torch.empty((N, K_half * 2), dtype=torch.int32, device=tensor_int8.device)
+    unpacked[:, 0::2] = low4.to(torch.int32)
+    unpacked[:, 1::2] = high4.to(torch.int32)
+
+    return unpacked
+
+def get_weight_perms(interleave: bool=True):
+    perm = []
+    for i in range(64):
+
+        for col in range(4):
+            cur_col = (i % 16) * 4 + col
+            for row in range(8):
+                cur_row = (i // 16) * 8 + row
+                cur_idx =  cur_row * 64 + cur_col
+                perm.append(cur_idx)
+
+    perm = np.array(perm)
+    if interleave:
+        interleave = np.array([4, 0, 5, 1, 6, 2, 7, 3])
+        perm = perm.reshape((-1, 8))[:, interleave].ravel()
+
+    perm = torch.from_numpy(perm)
+
+    return perm
+
+def marlin_weights(q_w,weight_perm,k_tile=32,n_tile=64,pack_factor=8):
+    size_k, size_n = q_w.shape
+    q_w = q_w.reshape((size_k // k_tile, k_tile, size_n // n_tile, n_tile))
+    q_w = q_w.permute((0, 2, 1, 3))
+    q_w = q_w.reshape((size_k // k_tile, size_n * k_tile))
+    q_w = q_w.reshape((-1, weight_perm.numel()))[:, weight_perm].reshape(q_w.shape)
+
+    orig_device = q_w.device
+    q_w = q_w.contiguous().to(torch.int32)
+    M, N = q_w.shape
+    assert N % pack_factor == 0, f"size_n ({N}) must be divisible by pack_factor ({pack_factor})"
+    q_packed = torch.zeros((M, N // pack_factor), dtype=torch.int32, device=orig_device)
+    for i in range(pack_factor):
+        q_packed += q_w[:, i::pack_factor] << (4 * i)
+
+    return q_packed
+
+def w4a8_2_marlin_weight(w4a8_w):
+    full_w4a8_w = unpack_int8_to_int4(w4a8_w)
+    full_w4a8_w = full_w4a8_w.T
+    weight_perm = get_weight_perms()
+    marlin_q_w = marlin_weights(full_w4a8_w, weight_perm, k_tile=32, n_tile=64, pack_factor=8)
+    return marlin_q_w
+
+def w4a8_weight_repack_impl(input):
+    if use_lightop:
+        size_batch = input.shape[0]
+        size_n = input.shape[1]
+        size_k = input.shape[2] * 2
+        output = torch.zeros((size_batch, size_k // 32, size_n * 4), device=input.device, dtype=torch.int32)
+        awq_marlin_repack_w4a8(input, output, size_batch, size_k, size_n)
+    else:
+        w_marlin_list = []
+        for e in range(input.shape[0]):
+            w_marlin_in = w4a8_2_marlin_weight(input[e])
+            w_marlin_list.append(w_marlin_in)
+        output = torch.stack(w_marlin_list, dim=0)
+
+    return output

python/sglang/srt/layers/quantization/w8a8_int8.py

@@ -22,7 +22,8 @@ from sglang.srt.layers.quantization.base_config import (
     QuantizeMethodBase,
 )
 from sglang.srt.layers.quantization.compressed_tensors.utils import should_ignore_layer
-from sglang.srt.layers.quantization.int8_kernel import per_token_quant_int8
+# from sglang.srt.layers.quantization.int8_kernel import per_token_quant_int8
+from lmslim.layers.gemm.int8_utils import per_token_quant_int8
 from sglang.srt.layers.quantization.unquant import UnquantizedLinearMethod
 from sglang.srt.utils import (
     apply_module_patch,
@@ -39,6 +40,8 @@ if TYPE_CHECKING:
         CombineInput,
         StandardDispatchOutput,
     )
+from lmslim import quant_ops
+
 
 _is_cuda = is_cuda()
 _is_cpu_amx_available = cpu_has_amx_support()
@@ -405,7 +408,7 @@ class W8A8Int8LinearMethod(LinearMethodBase):
         x_scale_2d = x_scale.view(-1, x_scale.shape[-1])
         output_shape = [*x_q.shape[:-1], layer.weight.shape[1]]
 
-        output = int8_scaled_mm(
+        output = quant_ops.triton_scaled_mm(
             x_q_2d,
             layer.weight,
             x_scale_2d,

python/sglang/srt/managers/schedule_batch.py

@@ -1618,7 +1618,7 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
         self.seq_lens_cpu = self.seq_lens_cpu[keep_indices]
         self.orig_seq_lens = self.orig_seq_lens[keep_indices_device]
         self.out_cache_loc = None
-        self.seq_lens_sum = self.seq_lens.sum().item()
+        self.seq_lens_sum = self.seq_lens.sum()
         self.output_ids = self.output_ids[keep_indices_device]
         self.return_logprob = any(req.return_logprob for req in self.reqs)
         if self.return_logprob:

python/sglang/srt/model_executor/model_runner.py

@@ -203,7 +203,7 @@ _is_xpu_xmx_available = xpu_has_xmx_support()
 SGLANG_CI_SMALL_KV_SIZE = os.getenv("SGLANG_CI_SMALL_KV_SIZE", None)
 
 # Detect stragger ranks in model loading
-UNBALANCED_MODEL_LOADING_TIMEOUT_S = 300
+UNBALANCED_MODEL_LOADING_TIMEOUT_S = 3600
 
 # the ratio of mamba cache pool size to max_running_requests, it will be safe when it is larger than 2 (yizhang2077)
 MAMBA_CACHE_SIZE_MAX_RUNNING_REQUESTS_RATIO = 3

python/sglang/srt/profile/prof.py

@@ -0,0 +1,58 @@
+from ctypes import *
+import os
+import time
+import threading
+
+class Prof:
+    def __init__(self):
+        self.use_roctx = os.getenv('SGLANG_HIP_PROF') is not None
+        if self.use_roctx:
+            self.lib = cdll.LoadLibrary("libroctracer64.so")
+            self.lib.roctxRangePushA.argtypes = [c_char_p]
+            self.lib.roctxRangePushA.restype = c_int
+            self.lib.roctxRangePop.restype = c_int
+        self.tm = time.perf_counter()
+        self.push_depth = {}
+
+    def StartTracer(self):
+        if self.use_roctx:
+            if self.lib is None:
+                self.lib = cdll.LoadLibrary("libroctracer64.so")
+            self.lib.roctracer_start()
+            self.roc_tracer_flag = True
+
+    def StopTracer(self):
+        if self.use_roctx:
+            if self.lib is None:
+                self.lib = cdll.LoadLibrary("libroctracer64.so")
+            self.lib.roctracer_stop()
+            self.roc_tracer_flag = False
+
+    def thread_depth_add(self, num):
+        current_thread = threading.current_thread()
+        thread_id = current_thread.ident
+        if thread_id not in self.push_depth.keys():
+            self.push_depth[thread_id] = 0
+        if num < 0 and self.push_depth[thread_id] == 0:
+            return False
+        self.push_depth[thread_id] += num
+        return True
+
+    def ProfRangePush(self, message):
+        if profile.use_roctx and self.roc_tracer_flag:
+            profile.lib.roctxRangePushA(message.encode('utf-8'))
+            profile.lib.roctxRangePushA(message.encode('utf-8'))
+            self.thread_depth_add(1)
+
+    def ProfRangePop(self):
+        if profile.use_roctx and self.roc_tracer_flag:
+            if not self.thread_depth_add(-1):
+                return
+            profile.lib.roctxRangePop()
+
+    def ProfRangeAutoPush(self, message):
+        self.ProfRangePop()
+        self.ProfRangePush(message)
+
+
+profile = Prof()

python/sglang/srt/server_args.py

@@ -93,6 +93,7 @@ QUANTIZATION_CHOICES = [
     "w4afp8",
     "mxfp4",
     "compressed-tensors",  # for Ktransformers
+    "slimquant_w4a8_marlin",
 ]
 
 ATTENTION_BACKEND_CHOICES = [

sgl-kernel/csrc/gemm/qserve_w4a8_per_chn_gemm.cu

@@ -25,7 +25,7 @@
 #define INTRIN_M 16
 #define INTRIN_N 16
 #define INTRIN_K 32
-#define WARP_SIZE 32
+#define WARP_SIZE 64
 #define SMEM_PAD_A 0
 #define SMEM_PAD_B 0
 #define PACK_SIZE 16

sgl-kernel/csrc/gemm/qserve_w4a8_per_group_gemm.cu

@@ -25,7 +25,7 @@
 #define INTRIN_M 16
 #define INTRIN_N 16
 #define INTRIN_K 32
-#define WARP_SIZE 32
+#define WARP_SIZE 64
 #define SMEM_PAD_A 0
 #define SMEM_PAD_B 0
 #define PACK_SIZE 16

sgl-kernel/csrc/kvcacheio/transfer.cu

@@ -5,7 +5,7 @@
 #include <cstdint>
 
 #ifndef USE_ROCM
-#define WARP_SIZE 32
+#define WARP_SIZE 64
 #include "pytorch_extension_utils.h"
 #else
 #include "pytorch_extension_utils_rocm.h"

sgl-kernel/csrc/quantization/gguf/ggml-common.h

@@ -3,7 +3,7 @@
 // copied from https://github.com/ggerganov/llama.cpp/blob/b2899/ggml-common.h
 #define QK_K 256
 #define K_QUANTS_PER_ITERATION 2
-#define WARP_SIZE_GGUF 32
+#define WARP_SIZE_GGUF 64
 #define K_SCALE_SIZE 12
 #define CUDA_DEQUANTIZE_BLOCK_SIZE 256
 #define CUDA_QUANTIZE_BLOCK_SIZE 256

sgl-kernel/include/utils.h

@@ -340,7 +340,7 @@ inline bool getEnvEnablePDL() {
 #define CEILDIV(x, y) (((x) + (y) - 1) / (y))
 
 #ifndef USE_ROCM
-#define WARP_SIZE 32
+#define WARP_SIZE 64
 #else
 #if defined(__GFX9__) || !defined(__HIP_DEVICE_COMPILE__)
 #define WARP_SIZE 64