sglang 0.4.6__py3-none-any.whl → 0.4.6.post1__py3-none-any.whl

sglang/test/send_one.py

@@ -6,11 +6,56 @@ python3 -m sglang.test.send_one
 """
 
 import argparse
+import dataclasses
 import json
 
 import requests
 
 
+@dataclasses.dataclass
+class BenchArgs:
+    host: str = "localhost"
+    port: int = 30000
+    batch_size: int = 1
+    temperature: float = 0.0
+    max_new_tokens: int = 512
+    frequency_penalty: float = 0.0
+    presence_penalty: float = 0.0
+    json: bool = False
+    return_logprob: bool = False
+    prompt: str = (
+        "Human: Give me a fully functional FastAPI server. Show the python code.\n\nAssistant:"
+    )
+    image: bool = False
+    stream: bool = False
+
+    @staticmethod
+    def add_cli_args(parser: argparse.ArgumentParser):
+        parser.add_argument("--host", type=str, default=BenchArgs.host)
+        parser.add_argument("--port", type=int, default=BenchArgs.port)
+        parser.add_argument("--batch-size", type=int, default=BenchArgs.batch_size)
+        parser.add_argument("--temperature", type=float, default=BenchArgs.temperature)
+        parser.add_argument(
+            "--max-new-tokens", type=int, default=BenchArgs.max_new_tokens
+        )
+        parser.add_argument(
+            "--frequency-penalty", type=float, default=BenchArgs.frequency_penalty
+        )
+        parser.add_argument(
+            "--presence-penalty", type=float, default=BenchArgs.presence_penalty
+        )
+        parser.add_argument("--json", action="store_true")
+        parser.add_argument("--return-logprob", action="store_true")
+        parser.add_argument("--prompt", type=str, default=BenchArgs.prompt)
+        parser.add_argument("--image", action="store_true")
+        parser.add_argument("--stream", action="store_true")
+
+    @classmethod
+    def from_cli_args(cls, args: argparse.Namespace):
+        attrs = [attr.name for attr in dataclasses.fields(cls)]
+        return cls(**{attr: getattr(args, attr) for attr in attrs})
+
+
 def send_one_prompt(args):
     if args.image:
         args.prompt = (
@@ -20,20 +65,42 @@ def send_one_prompt(args):
     else:
         image_data = None
 
-    response = requests.post(
-        "http://localhost:30000/generate",
-        json={
-            "text": args.prompt,
-            "image_data": image_data,
-            "sampling_params": {
-                "temperature": args.temperature,
-                "max_new_tokens": args.max_new_tokens,
-                "frequency_penalty": args.frequency_penalty,
-                "presence_penalty": args.presence_penalty,
-            },
-            "return_logprob": args.return_logprob,
-            "stream": args.stream,
+    prompt = args.prompt
+
+    if args.json:
+        prompt = (
+            "Human: What is the capital of France and how is that city like. "
+            "Give me 3 trivial information about that city. "
+            "Write in a format of json.\nAssistant:"
+        )
+        json_schema = "$$ANY$$"
+        json_schema = (
+            '{"type": "object", "properties": {"population": {"type": "integer"}}}'
+        )
+    else:
+        json_schema = None
+
+    if args.batch_size > 1:
+        prompt = [prompt] * args.batch_size
+
+    json_data = {
+        "text": prompt,
+        "image_data": image_data,
+        "sampling_params": {
+            "temperature": args.temperature,
+            "max_new_tokens": args.max_new_tokens,
+            "frequency_penalty": args.frequency_penalty,
+            "presence_penalty": args.presence_penalty,
+            "json_schema": json_schema,
+            "stop": ["Question", "Assistant:", "<|separator|>", "<|eos|>"],
         },
+        "return_logprob": args.return_logprob,
+        "stream": args.stream,
+    }
+
+    response = requests.post(
+        f"http://{args.host}:{args.port}/generate",
+        json=json_data,
         stream=args.stream,
     )
 
@@ -47,6 +114,9 @@ def send_one_prompt(args):
     else:
         ret = response.json()
 
+    if args.batch_size > 1:
+        ret = ret[0]
+
     latency = ret["meta_info"]["e2e_latency"]
 
     if "spec_verify_ct" in ret["meta_info"]:
@@ -68,21 +138,7 @@ def send_one_prompt(args):
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    parser.add_argument("--temperature", type=float, default=0.0)
-    parser.add_argument("--max-new-tokens", type=int, default=512)
-    parser.add_argument("--frequency-penalty", type=float, default=0.0)
-    parser.add_argument("--presence-penalty", type=float, default=0.0)
-    parser.add_argument("--return-logprob", action="store_true")
-    parser.add_argument(
-        "--prompt",
-        type=str,
-        default="Human: Give me a fully functional FastAPI server. Show the python code.\n\nAssistant:",
-    )
-    parser.add_argument(
-        "--image",
-        action="store_true",
-    )
-    parser.add_argument("--stream", action="store_true")
+    BenchArgs.add_cli_args(parser)
     args = parser.parse_args()
 
     send_one_prompt(args)

sglang/test/test_utils.py

@@ -732,6 +732,44 @@ def run_bench_one_batch(model, other_args):
     return output_throughput
 
 
+def run_bench_offline_throughput(model, other_args):
+    command = [
+        "python3",
+        "-m",
+        "sglang.bench_offline_throughput",
+        "--num-prompts",
+        "1",
+        "--dataset-name",
+        "random",
+        "--random-input-len",
+        "256",
+        "--random-output-len",
+        "256",
+        "--model-path",
+        model,
+        *[str(x) for x in other_args],
+    ]
+
+    print(f"{command=}")
+    process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+
+    try:
+        stdout, stderr = process.communicate()
+        output = stdout.decode()
+        error = stderr.decode()
+        print(f"Output: {output}", flush=True)
+        print(f"Error: {error}", flush=True)
+
+        output_throughput = -1
+        for line in output.split("\n"):
+            if "Last generation throughput (tok/s):" in line:
+                output_throughput = float(line.split(":")[-1])
+    finally:
+        kill_process_tree(process.pid)
+
+    return output_throughput
+
+
 def lcs(X, Y):
     m = len(X)
     n = len(Y)

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.4.6"
+__version__ = "0.4.6.post1"

{sglang-0.4.6.dist-info → sglang-0.4.6.post1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sglang
-Version: 0.4.6
+Version: 0.4.6.post1
 Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License:                                  Apache License
                                    Version 2.0, January 2004
@@ -245,7 +245,7 @@ Requires-Dist: uvloop; extra == "runtime-common"
 Requires-Dist: xgrammar==0.1.17; extra == "runtime-common"
 Provides-Extra: srt
 Requires-Dist: sglang[runtime_common]; extra == "srt"
-Requires-Dist: sgl-kernel==0.0.9.post2; extra == "srt"
+Requires-Dist: sgl-kernel==0.1.0; extra == "srt"
 Requires-Dist: flashinfer_python==0.2.3; extra == "srt"
 Requires-Dist: torch==2.6.0; extra == "srt"
 Requires-Dist: torchvision==0.21.0; extra == "srt"

{sglang-0.4.6.dist-info → sglang-0.4.6.post1.dist-info}/RECORD

@@ -10,7 +10,7 @@ sglang/global_config.py,sha256=xzLdk8W53fneFblNh8iIjGF9C3-7mnzR1-LleD9Btxg,1495
 sglang/launch_server.py,sha256=mDXfwha8LHpWQJekcCosR98QhCQsbmilsBlI5jAIgg0,420
 sglang/llama3_eval.py,sha256=gWSboDchIGybIce88bJlrCG0yiLZ513mw4gcutJlzGM,10017
 sglang/utils.py,sha256=GIcgiRHkZ-gyPxXOdn1qFF41jkg4-YdDxbPc4mzO-qk,16159
-sglang/version.py,sha256=bbBpXE_PBbo_SaI807mDML0QJywD0_ufCDPgAMlDHaE,22
+sglang/version.py,sha256=mb7cZWFtBTYPgotnX_1oAZadFITLHrAXwTSs2Eb1dvU,28
 sglang/lang/chat_template.py,sha256=MwNL5dNTe8g_l2ljZubnrazEgT2xEv-9O2D0Ezwxy4I,19658
 sglang/lang/choices.py,sha256=-W1DVw9N9ZliVpvmWrzIXG4cswAah8eMQrHWzkS3D8o,6234
 sglang/lang/compiler.py,sha256=MAuzoOOpb98njJ7Io2SDmFkhTroDYiq0te0ZpfHkMY4,7597
@@ -34,10 +34,10 @@ sglang/srt/hf_transformers_utils.py,sha256=N2f-gA8yUq-UP_TJT276gNbDNzmddWsmWnq3p
 sglang/srt/mm_utils.py,sha256=1ScBunw_x4W8ebM_AcJ62-1T2mfT8NlMJqdAhkF1lb0,12367
 sglang/srt/model_parallel.py,sha256=eLXZhvJ4wG6dh0FontNCIdVZvHYdWgaeY-5cu7TD9tE,6078
 sglang/srt/patch_torch.py,sha256=OUPCGQSQz3MVZB1zZ_Eq8lXiw0uIKJ_HWjqQolI8FsM,3088
-sglang/srt/reasoning_parser.py,sha256=JnaEVW0KG1yJpn9uxmrjwErb9imzni05QDFjExryoqM,5584
-sglang/srt/server_args.py,sha256=ojKNFeiZ1W7nQScImdlG4xuHVR1yQdugwrlHj30WVK8,54804
+sglang/srt/reasoning_parser.py,sha256=8AMk3XI8mfvz4AUuRHf_pNYpM_Mr64uT9EZ3o90cqQ8,6341
+sglang/srt/server_args.py,sha256=jRHEskSyfmHbCnyqRzp3deI5HizDenDiyLjF65ZUqvg,55149
 sglang/srt/torch_memory_saver_adapter.py,sha256=HYlgYJ2sgmjs2RSjU2KbCaXijRg3mTDZ0ZcCB5Bt6Ps,2211
-sglang/srt/utils.py,sha256=FsLjbxNf1K1hmQTTpQ7PyZxze0m1nfyKfNhngfC8atU,62944
+sglang/srt/utils.py,sha256=u_YB-FXi3AY6mhRmk8wdPcKKAo1sZY0bMUgnjq8BtJI,63033
 sglang/srt/warmup.py,sha256=FmJiYfjRr3X_eAe7ojQaPoN17LvHpjDmRWRnO-k86AQ,1469
 sglang/srt/configs/__init__.py,sha256=vulncVn70WqIT6s0HaB8p_Q6FjOiaLwNZWpoJS9FIuQ,399
 sglang/srt/configs/chatglm.py,sha256=j-b0YkdYUmQm2y1kNmMJtKeACxWKmBbvNNkDWbs6kbI,2907
@@ -64,12 +64,14 @@ sglang/srt/constrained/outlines_jump_forward.py,sha256=Gyubp-FVetxd6wP4FA_kD6cCX
 sglang/srt/constrained/reasoner_grammar_backend.py,sha256=XFxdZqvPofmtCeIMqR10NOyph06HwbdXfiVI8rIoV5s,3646
 sglang/srt/constrained/xgrammar_backend.py,sha256=oc3BTTe8mB5Szv_O-5nZzWbKEKFb22oUniqTUZhewLQ,7409
 sglang/srt/constrained/triton_ops/bitmask_ops.py,sha256=WjTen9iuuFWLzkE1mAHQZB9_7aIy5QH8Wjf-lB-Fams,4614
-sglang/srt/disaggregation/decode.py,sha256=G9iYiCGEyI-bFFuLQhgsEN75bzhYtb--QCvlzbbSm1o,24884
+sglang/srt/disaggregation/decode.py,sha256=nSHCBfEtD3a6c2a7XPAcCh4c0jw3BLG9EL-L3LlW_V0,25139
 sglang/srt/disaggregation/mini_lb.py,sha256=zyJo20GI6G1ZIoBVY3ltcr0dDcH5qOJrtMfiGKGnBLI,10959
-sglang/srt/disaggregation/prefill.py,sha256=VE0t6SasG_4QNeEpBkNbwicEpR-SyEsyYSMPROERH2k,15100
-sglang/srt/disaggregation/utils.py,sha256=6ackrcitNF94ur9bRvPerF29E7g8uU_0a4-nSdw1EMs,3078
+sglang/srt/disaggregation/prefill.py,sha256=4wLYQtPMbKWMQvF3mGnvr8ygd9xRLO9zTwLKeM5BIf8,15424
+sglang/srt/disaggregation/utils.py,sha256=7gO734GOr4u03qwOf2UvFsfj4n-I37iyzQh7lFKbJRk,3501
 sglang/srt/disaggregation/base/__init__.py,sha256=KR8xXoRCDAy2U623mfP6ujXu42m1_F9EiudjrKu2I_A,130
 sglang/srt/disaggregation/base/conn.py,sha256=gpf32bhYXWm_iaYB6WcrDaJ-UoL1ZzPI_xpi5pMhRQo,2443
+sglang/srt/disaggregation/fake/__init__.py,sha256=zmfeKYXjonRhfFOck1c_mP7Q4cW5G0f1RsTwRivKu0s,47
+sglang/srt/disaggregation/fake/conn.py,sha256=DKEVBgmzV3CNzZ0-r7rFV4orue2iP_7apEtgn-fcTEA,2552
 sglang/srt/disaggregation/mooncake/__init__.py,sha256=1vacEHmWjf7zgbMPzsXKB08FqNKNCquJdUiDlO41BOk,122
 sglang/srt/disaggregation/mooncake/conn.py,sha256=DQ_PTxq_nZHFZ4LxHDhCIvQFPA1xUbaw1Sleyqhkq6U,28224
 sglang/srt/disaggregation/mooncake/transfer_engine.py,sha256=MxDAB9ZetRF1pFS2LP3FVHPtQ1HjIt_SK3UMaYHZ94o,2604
@@ -88,8 +90,8 @@ sglang/srt/distributed/device_communicators/pynccl_wrapper.py,sha256=LblisImY9d6
 sglang/srt/distributed/device_communicators/shm_broadcast.py,sha256=bbruDIM1GgKIdB6gi71_I0mpB179I-qyvwKuSj1Kaic,20816
 sglang/srt/distributed/device_communicators/xpu_communicator.py,sha256=ajW6132BvA6jkeipEIgN27TFycI0U06Ih2Z8WNjlA4s,1593
 sglang/srt/entrypoints/EngineBase.py,sha256=xoyvp6XAeDLY2_Q2Ng33H-fRhrXHv2ldJJKd-HuDhqE,1870
-sglang/srt/entrypoints/engine.py,sha256=-qHfYFDremLj8GhwQa7GuWHQKop7Tyq2HvmBXULtCos,22282
-sglang/srt/entrypoints/http_server.py,sha256=4fbn2KfUs2fjSvDulnhGfcEGA5a9ahUEre59TazwVA4,28508
+sglang/srt/entrypoints/engine.py,sha256=e5BPBIewPTQpYlk1c2eC4C_xyQgJ3mgNEm2Sg3GfV2s,22518
+sglang/srt/entrypoints/http_server.py,sha256=vvyvCosUp5aTFlD8k4IyZDzj2yXQIsndhPkTl4u1nGc,29573
 sglang/srt/entrypoints/http_server_engine.py,sha256=ihA6y3GXRs28Y9U3SgdQcJQjnw_SVIby7QrVgiafX04,4846
 sglang/srt/entrypoints/verl_engine.py,sha256=XLYdwTwhH0jTjw8xczgZXWfBXMRb_ur2bg4TN0dTwfI,6975
 sglang/srt/layers/activation.py,sha256=oSkdo8B8najXFcVay3Y__CEvgXh87lAIhG0CMp2Ugqs,5954
@@ -106,6 +108,7 @@ sglang/srt/layers/sampler.py,sha256=PNgMXm2vsNsE6Rt89R5GLDC3lDxdIujoWli8F3vldng,
 sglang/srt/layers/torchao_utils.py,sha256=Ws24FdRBSkTpyeyA6bQrdDm-W5wfDxKvSIPUSahyMfA,4063
 sglang/srt/layers/vocab_parallel_embedding.py,sha256=QUxd4sELx6p3dHvEKmccPZ-phdd_9EjNdwjH3SJ9zxI,22238
 sglang/srt/layers/attention/base_attn_backend.py,sha256=lGujcYJ_CxHJy0Q9L6Phn3ds-nBGMy0OGj3R54R65iQ,3334
+sglang/srt/layers/attention/cutlass_mla_backend.py,sha256=kqtTVCIgDhcW5y9iWP8xcwGPuev-V5ipAUG-Ae3ot7g,9883
 sglang/srt/layers/attention/double_sparsity_backend.py,sha256=2ZRL_gYz14idoVqQzeQ6N77nXer0f_8_TUYw40XUUz0,9161
 sglang/srt/layers/attention/flashattention_backend.py,sha256=ysJt9pJ8pg_kVxvVUTvUL22-O7ABHCenLGGcqCotD6A,83206
 sglang/srt/layers/attention/flashinfer_backend.py,sha256=YtMTgMhxxNrAbSoWTPJczgY4SR3WjnAPXPoJ2d5PUZY,46394
@@ -113,7 +116,7 @@ sglang/srt/layers/attention/flashinfer_mla_backend.py,sha256=pnVhvVEK87iFW8gUb1G
 sglang/srt/layers/attention/flashmla_backend.py,sha256=IyE4w7GcNOxjjy3mQeuAMjPtBNvI-6JkoxvBlFxFvec,10270
 sglang/srt/layers/attention/torch_native_backend.py,sha256=K5hUqBgakk2COSQqsaxWs0yEVOHS-7BlOygZTOeI8kE,9444
 sglang/srt/layers/attention/triton_backend.py,sha256=oEEiUwHbm4rNw5ExbQ2c3n0TwAgkk77yuLFenj9bHOo,26902
-sglang/srt/layers/attention/utils.py,sha256=J9mA-cbZT3uTlaKXo0HEAaeMei_TS2o4McTna9LVDCE,2750
+sglang/srt/layers/attention/utils.py,sha256=rxB4sbNIHDTges78bDbnpd_hUgtyb3e16wUwgI4WmoU,2751
 sglang/srt/layers/attention/vision.py,sha256=CtFU1wyz5191LcuyDzGJ01mB-mM-upPj2pXg4DO6wh4,11985
 sglang/srt/layers/attention/triton_ops/decode_attention.py,sha256=DPu_aCPgwPqKWZPEQmp_xA7MPbpV2ip-MEICCB470Ao,19120
 sglang/srt/layers/attention/triton_ops/double_sparsity_attention.py,sha256=vsDZZ5QGb8-KBzADgKshnVQbsW8zRJF1h5hgdPGW5lU,31124
@@ -128,8 +131,8 @@ sglang/srt/layers/moe/ep_moe/kernels.py,sha256=ijqRzS-tb0LGnDU5hW-g0JH104ppADrWa
 sglang/srt/layers/moe/ep_moe/layer.py,sha256=SZ0shPwgDp7xj-TCv9bfg5O7f2AXjF6xmBP5xkZ0Ips,36440
 sglang/srt/layers/moe/ep_moe/token_dispatcher.py,sha256=zQV7Qr-Zrcr3D3efVvZepRQM02bj5djHPsijPssavk8,20430
 sglang/srt/layers/moe/fused_moe_triton/__init__.py,sha256=h9yMFAL_bagUf-qBED8gSWdCOb7d8IdA-pE-L_nIg8E,842
-sglang/srt/layers/moe/fused_moe_triton/fused_moe.py,sha256=bgKelgR3IPgdK4ihasWL4EaFi_z4mVnOKDeMLAio8hc,55951
-sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=XofcceHnkmqhVczEIVqGSbsxps5LiYTGcaCk1Say9YM,24951
+sglang/srt/layers/moe/fused_moe_triton/fused_moe.py,sha256=13ygSeBoRkiqsERSHOIbIxLplVsSl-SUT6JxYPB-ViM,55968
+sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=BMOV76fabrZcoyDmRpRbH11Jc0ogWH2k2QAQwvZIpgI,25084
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json",sha256=iNGsE2ZeVnQEnN4A8UJ9Jv0d3hbRF2MJ9oBgjup5Szk,2737
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=JJN0hryyLr5Zv3dSS7C8cPFhAwTT6XxUVnBGMZvV6JA,2752
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json",sha256=ouRyZ5PEMPP2njPftCNhs-1g1y6wueWLmhI7G1SjV1k,4131
@@ -141,7 +144,19 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=XofcceHnkmqhVczEIVqGSbsxp
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=yf33YmWlVSjjyg0Q4OMAWvc9gjRxvttMrQBUEOfPl4I,4153
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json",sha256=ZWMClYN1moVRUP2f0hYac38di_pUgZggyl9d2D5rnoc,4136
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=C65Q2Mv1LxFQ_qDnv11IZ9nwl7sGZo72nWDflMttu4g,4147
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=192,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=I3k416HbXU_rYb8scD8gAI4fuBlElHl06PM347Qa11w,3253
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=192,device_name=NVIDIA_H20.json",sha256=RgV8C4F1LO09h01YsgF_eqX6GNoBtC7ulPfJRUUbg_g,3241
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=192,device_name=NVIDIA_H200.json",sha256=nsNEuDNks0tVLfQfIm7xxFwEeptTfQcoa9fJy0NS8xQ,3247
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=qbqjisJ4oKmcYzumHPRk5UyOzsdi8J6xas82UWHMeAI,3263
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=384,device_name=NVIDIA_H20.json",sha256=vS2DRIDOqWyiBvbG6H746ownfkD1F8Aj2YZ0ET9xll8,3232
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=1n5XyZZ5sKAi-Z1duWOhLUfr6gkvnOpvxfbqIT6iU_4,3265
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=384,device_name=NVIDIA_H200.json",sha256=xqhl748it8GV2KXX0XixitE_ywnsKksqK8AGL7tAgT8,3254
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=512,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=FsWbV4Q6AzAtgegVuENBDz2ZcSJsqNiwUIVfQbpP7hQ,3244
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=IuvyC8TNhCVAmUZfLSoETsyCKsmejKXrs_0zuwFLPAU,3265
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=768,device_name=NVIDIA_H20.json",sha256=10Ntu2aVD5vGLonx-jW0qNw-tgZWdZmzMGx7utDVeng,3237
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=768,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=pdQ1RvXvdWDn8Y8-8MAX3vn-T-wbtkZvHV9GZZvNjnc,3266
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=768,device_name=NVIDIA_H200.json",sha256=JraM-Nvbg5V_TJkSl6UPFYZN1zHHoIbr2pAcksenoTY,3248
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=96,device_name=NVIDIA_H20.json",sha256=JtcHRlPz8xQEAqJ9EWI63oYvdmjQFG6VTHqtt85VOSA,3221
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=144,N=512,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=pk6VZChh2Y0CsJSzjtUhOnlta1QLTUEWy33aKQU47XY,3244
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=Gmk24hc5lVIfQtqSa5wLOcWKedMN8aZUe93DBh6J1AY,3249
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=1024,device_name=NVIDIA_H200.json",sha256=uY_XMPomaXMXxIkTR4ctU_Ybri_jMv2VvCcV-f6O_bw,3255
@@ -270,9 +285,9 @@ sglang/srt/layers/quantization/awq.py,sha256=KemDG55U3B6YZVjMV71awVAIj0islFvtxcU
 sglang/srt/layers/quantization/base_config.py,sha256=jWk_egQrVNMYmQgbTI9vkcgzScLFjB5_sywFlAfE5J0,4776
 sglang/srt/layers/quantization/blockwise_int8.py,sha256=cu9-JiCZDfMfvB97Kv_-eEG87VX5bRFIllFkzpO_xIg,15122
 sglang/srt/layers/quantization/deep_gemm.py,sha256=UFzsd0iiqVTBo0Ow_6ylVVFK8B9EUWTNQQYGvsNfm2s,13129
-sglang/srt/layers/quantization/fp8.py,sha256=Niu89OfZd4pIpkwZ1zd2Hrlffx0c5L5zkix6Lzi0Vys,40970
+sglang/srt/layers/quantization/fp8.py,sha256=da-6ji_HBISKwIgMMX-JGlDKMLi-qL9j2XLer5cFAsU,40945
 sglang/srt/layers/quantization/fp8_kernel.py,sha256=C2_hOLRO27-Yvjy-Nm2niehD2gWSMuP6TnNX07ESqh4,32018
-sglang/srt/layers/quantization/fp8_utils.py,sha256=1AmKsAFlpPcwbqpFR6P2CaCOvpp8dYY7lzXLoKjVikE,18631
+sglang/srt/layers/quantization/fp8_utils.py,sha256=vqH-bMb2DD0A7Y7hZjN-TGTg5h6aJ-cLW9H2adyZzqk,18651
 sglang/srt/layers/quantization/gptq.py,sha256=gyGMOPXHzozK7pPWSjKgLdFX9h7MCEww7n8FqEVEVac,15364
 sglang/srt/layers/quantization/int8_kernel.py,sha256=CR-VuTTR4GYluOZTpS5mmEz3hYrsY4GOX-G-h3XAYKc,12163
 sglang/srt/layers/quantization/int8_utils.py,sha256=YK9CS-lb_n91kNCTKK5o5apYF31V2giDg5G5VKrpcUA,2356
@@ -464,14 +479,14 @@ sglang/srt/managers/expert_distribution.py,sha256=r3o5RGI0gnV7xb60AApqKYa0oiSB37
 sglang/srt/managers/io_struct.py,sha256=9mdBGOkblguT1x6Ds9wL3j0MWAQiUQVdVRL4a7IUnA4,31631
 sglang/srt/managers/mm_utils.py,sha256=JTu5B7jZWTtZi8LCpVa6ITvSToxcuf5PDbb3FJC9M6o,18089
 sglang/srt/managers/multimodal_processor.py,sha256=XlRYvNhF6XOssreRX9DZPhLSpps_VE62gSKw3EGdNPo,2088
-sglang/srt/managers/schedule_batch.py,sha256=NXeAXZVvXcB8ApByiDLNaFww3aNQl_rZDxQKujhD5uo,63553
+sglang/srt/managers/schedule_batch.py,sha256=zUQGVjLbi9TK5tfyzHNMSAnPeNeFi9GFI2AC8Fr2pbo,63824
 sglang/srt/managers/schedule_policy.py,sha256=E1qVq2G3jptKdX9nlqfayeRBUll9xB6bK8nBf3EW32E,19469
-sglang/srt/managers/scheduler.py,sha256=3EJ__bAck1xF8GWR6hFlt0HU7VwHMO7uM2eXKHOf3KI,80627
-sglang/srt/managers/scheduler_output_processor_mixin.py,sha256=u2sj6MViFTov0lVZSysZ-wph2pEqRCtCjwA1UdttZ7I,26338
+sglang/srt/managers/scheduler.py,sha256=7o03npmnu775d1DRDAkTJjl8OuJlE_xuM3BQji6BYLI,80808
+sglang/srt/managers/scheduler_output_processor_mixin.py,sha256=GxdkTR24_P_2C3ib0dc7Xqklrz8SiHtUTlM0c7AlKlk,26754
 sglang/srt/managers/session_controller.py,sha256=o-ifit0n4_xHLNmyD0Ams8FxGRgxFybX-Vz1hwgr3UQ,5755
 sglang/srt/managers/tokenizer_manager.py,sha256=4l4PAvfQrJqlYADQbl7cgpLhBBY52pzI5AzRYIzAjLs,50693
-sglang/srt/managers/tp_worker.py,sha256=OV0Zc-oK-HXSWO6x26ItDEf-qB2IqyJq2z0TFuBZz0s,8970
-sglang/srt/managers/tp_worker_overlap_thread.py,sha256=3_ZJ8Rq7v2ZDaRNTRu5Dy8AbqiAlJQp3IAKnn_WAwd8,9127
+sglang/srt/managers/tp_worker.py,sha256=LhbhCovDvab6Cx4faR88s4q_3D-Di9s5sKCndsDxF9E,8966
+sglang/srt/managers/tp_worker_overlap_thread.py,sha256=lkG_yN6_UEv5mhmZ7cKP7_A5sIVMQw1GPwkqM91EWSE,9304
 sglang/srt/managers/utils.py,sha256=5i75uLlQOF_5CaT02CrWtwozMTtwTg2_nLP8Dtr-JZQ,1536
 sglang/srt/managers/multimodal_processors/base_processor.py,sha256=ata9H6Ry4QfqBoA_g0auG0sMnKfGrlZn74lM77ihtiA,10172
 sglang/srt/managers/multimodal_processors/clip.py,sha256=lRc2mcuDbAhZVf-0EfkO81pqDiol9zLvTpDqtPIBQ2k,1525
@@ -494,7 +509,7 @@ sglang/srt/metrics/collector.py,sha256=zHg4twFQJvuK1mSme3-EYQa9PJryfp_u7a4RxQ5Rc
 sglang/srt/metrics/func_timer.py,sha256=VFyNRrbnKVCwnQsrlLin1lITJfjQpf9m8sGPqL5LIsQ,3438
 sglang/srt/model_executor/cuda_graph_runner.py,sha256=iFryO9dglpnFCoNWxZqKdUhQycT8In29C0kIba3G1Dw,23687
 sglang/srt/model_executor/forward_batch_info.py,sha256=T9B5vWaJwlKUH0fQTPe3XdbkTYEUI6iKxBxUHs-cAMM,26632
-sglang/srt/model_executor/model_runner.py,sha256=Eei9JGCX-SQaTF1T4iBS0zg4UPdyjQ67UkKwgbK_200,46136
+sglang/srt/model_executor/model_runner.py,sha256=O4vKZ4c-u69ZeKPBjAfiunvtnQHskZVmbUUK4fKFb5E,46417
 sglang/srt/model_loader/__init__.py,sha256=zGZkOBz1zx-pkaIy47BasL3fjDlAcxAXUTjInOhXHAE,919
 sglang/srt/model_loader/loader.py,sha256=YYmtvkQw0B1qgPw0_gN-K4yy7CEYbTSR__0Dl1Fnm6k,55342
 sglang/srt/model_loader/utils.py,sha256=0NaMR67fESFopaklmsleiL27XH1QUrjZW246MUu1EJ0,1369
@@ -561,8 +576,8 @@ sglang/srt/models/torch_native_llama.py,sha256=5tfFSMAXB3ScToqTALtCXa8Oo-qPCJh-K
 sglang/srt/models/xverse.py,sha256=DsNVI9JpzN4jj0Ry6aTrj7r-xq5YLOoDX2kH4YLJA-I,14035
 sglang/srt/models/xverse_moe.py,sha256=7KCM2-j12towDMNvXkuuYiBOmNauH6NG4Ip40x0khqA,16782
 sglang/srt/models/yivl.py,sha256=oToK7-u5IGO7xwpJIQ7VtudlK6-zPqJX4bt6_wv0SH8,4850
-sglang/srt/openai_api/adapter.py,sha256=MMHLsz6Thepwtr4rtOlYwOlaKk212vgpC0p5nmqGt5U,75550
-sglang/srt/openai_api/protocol.py,sha256=1IVfYt0LUNLqFz0DNZyqxU02IRYVHcL7jnpMG9bNq34,13532
+sglang/srt/openai_api/adapter.py,sha256=7WMplmT0SWJXo5F8s1s3Q9_6WV_cTscMS1Bodbl9Xes,76746
+sglang/srt/openai_api/protocol.py,sha256=8Iu4t9JlH99QggKl55PYQWTW81u5mpOj0aA-bs44A_c,13621
 sglang/srt/platforms/interface.py,sha256=hym3iooBB4C8if5hDZezgVN6h4NIOu7sg2ZUBIV6XmM,11246
 sglang/srt/sampling/custom_logit_processor.py,sha256=tDvoLgLqn-sy1qcY6vSrpbnHCeqbdk0uhMOO-uy4p4E,1099
 sglang/srt/sampling/sampling_batch_info.py,sha256=4LCowU2bk0TOSfIGpEy90N1SpTsiOKK8Rx1ZYcklUFQ,11988
@@ -582,7 +597,7 @@ sglang/test/few_shot_gsm8k.py,sha256=7VLbWl4nCQs1wjtW4q-46jf9jUCycSs5Iw8v7sUSzBw
 sglang/test/few_shot_gsm8k_engine.py,sha256=QQbrwOX6-cJDD3RZC_e7zPnt6aSo8JdF8X_lRHSjdDM,3886
 sglang/test/run_eval.py,sha256=9yO0hXZOcn4abEOs96T-XPguDEklK16Ltco0pGF3zCg,4020
 sglang/test/runners.py,sha256=vSOl38rVDR3l2ezVCs672vE-LcOA2rJHjlkhLgEjcz8,30260
-sglang/test/send_one.py,sha256=6FhbJ3c8RpXxvFTELRXaF97GpT7zXXsCDYZh1DqG22E,2550
+sglang/test/send_one.py,sha256=_l72sRfuXRUldyD3PD63hg_WxNvvhW5unNnbe4XuAwk,4380
 sglang/test/simple_eval_common.py,sha256=joqrGysuLnJFtzDRIgFkMsRyKUSyjVPFWp0_PHAL3Ik,12378
 sglang/test/simple_eval_gpqa.py,sha256=8Xt9Bw05c7SZTYrCZgB68OZUqUbLo69ywiyx0bTvSUk,3220
 sglang/test/simple_eval_humaneval.py,sha256=zmV3xWYc2OrpiT9Dy55RTKZL5DEROD1cJ0NA_-cU5zI,5685
@@ -596,13 +611,13 @@ sglang/test/test_custom_ops.py,sha256=2bSo9P5_rJZYFq8Y8IKRimDfFyZZGJluhL7Ngny0Pf
 sglang/test/test_dynamic_grad_mode.py,sha256=L76yUCuk_ymNpXD2CmO8r2GiGjIvD_gtTsuFDs2NolI,1638
 sglang/test/test_layernorm.py,sha256=2GMWqqNDuGvSMSsEBF5eDCzwVSYA9E6hGhRo6s4ecKg,3764
 sglang/test/test_programs.py,sha256=VZ3vXtUDBnXz0M7gFdDH8hXg9Wa0j_qI8CVqjEgRN_E,18877
-sglang/test/test_utils.py,sha256=OPTl_PKVlHitrRJW5DFClpQA6xN6FvxatiOpgngCuJU,31153
+sglang/test/test_utils.py,sha256=1U4Jtx_oz_UtS3SSJdqGuh3ujnj2g8pZjN5MYsbsBwI,32164
 sglang/test/attention/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/test/attention/test_flashattn_backend.py,sha256=_rTG849FwQdVTyGKkqhczaOqngBmRWXFmkl5NnuK1GM,13914
 sglang/test/attention/test_flashattn_mla_backend.py,sha256=g4O50WblTpM7_Gq2b76k0i25_z01BOUBQ4i6PmyxpO4,10774
 sglang/test/attention/test_prefix_chunk_info.py,sha256=er0i3KGHMkw-4UZB1GCFd4oYwRcXfU5wpO1ORqpNGGA,7626
-sglang-0.4.6.dist-info/licenses/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
-sglang-0.4.6.dist-info/METADATA,sha256=ITL_ELbps-k7hM4JnrS_YQCgwSO1glI5-dE9kX0Hft0,25361
-sglang-0.4.6.dist-info/WHEEL,sha256=ck4Vq1_RXyvS4Jt6SI0Vz6fyVs4GWg7AINwpsaGEgPE,91
-sglang-0.4.6.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
-sglang-0.4.6.dist-info/RECORD,,
+sglang-0.4.6.post1.dist-info/licenses/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
+sglang-0.4.6.post1.dist-info/METADATA,sha256=UTh1TF2jiAdQunwLv7_bmww5_18c4uD7FCaeO-Z3gAs,25361
+sglang-0.4.6.post1.dist-info/WHEEL,sha256=ck4Vq1_RXyvS4Jt6SI0Vz6fyVs4GWg7AINwpsaGEgPE,91
+sglang-0.4.6.post1.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.4.6.post1.dist-info/RECORD,,