sglang 0.5.3rc2__py3-none-any.whl → 0.5.4__py3-none-any.whl

sglang/test/test_utils.py

@@ -16,11 +16,10 @@ import unittest
 from concurrent.futures import ThreadPoolExecutor
 from dataclasses import dataclass
 from datetime import datetime
-from functools import partial
+from functools import partial, wraps
 from pathlib import Path
 from types import SimpleNamespace
 from typing import Any, Awaitable, Callable, List, Optional, Tuple
-from urllib.parse import quote
 
 import aiohttp
 import numpy as np
@@ -76,6 +75,11 @@ DEFAULT_MODEL_NAME_FOR_TEST_FP8_WITH_MOE = "gaunernst/DeepSeek-V2-Lite-Chat-FP8"
 DEFAULT_MODEL_NAME_FOR_TEST_W8A8 = "RedHatAI/Llama-3.2-3B-quantized.w8a8"
 DEFAULT_MODEL_NAME_FOR_TEST_W8A8_WITH_MOE = "nytopop/Qwen3-30B-A3B.w8a8"
 
+# INT4 models
+DEFAULT_MODEL_NAME_FOR_TEST_AWQ_INT4 = (
+    "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4"
+)
+
 # EAGLE
 DEFAULT_EAGLE_TARGET_MODEL_FOR_TEST = "meta-llama/Llama-2-7b-chat-hf"
 DEFAULT_EAGLE_DRAFT_MODEL_FOR_TEST = "lmsys/sglang-EAGLE-llama2-chat-7B"
@@ -122,7 +126,12 @@ def is_in_ci():
 
 def is_in_amd_ci():
     """Return whether it is in an AMD CI runner."""
-    return get_bool_env_var("SGLANG_AMD_CI")
+    return get_bool_env_var("SGLANG_IS_IN_CI_AMD")
+
+
+def is_blackwell_system():
+    """Return whether it is running on a Blackwell (B200) system."""
+    return get_bool_env_var("IS_BLACKWELL")
 
 
 def _use_cached_default_models(model_repo: str):
@@ -136,17 +145,20 @@ def _use_cached_default_models(model_repo: str):
 
 if is_in_ci():
     DEFAULT_PORT_FOR_SRT_TEST_RUNNER = (
-        5000 + int(os.environ.get("CUDA_VISIBLE_DEVICES", "0")[0]) * 100
+        10000 + int(os.environ.get("CUDA_VISIBLE_DEVICES", "0")[0]) * 1000
     )
 else:
     DEFAULT_PORT_FOR_SRT_TEST_RUNNER = (
-        7000 + int(os.environ.get("CUDA_VISIBLE_DEVICES", "0")[0]) * 100
+        20000 + int(os.environ.get("CUDA_VISIBLE_DEVICES", "0")[0]) * 1000
     )
 DEFAULT_URL_FOR_TEST = f"http://127.0.0.1:{DEFAULT_PORT_FOR_SRT_TEST_RUNNER + 1000}"
 
 if is_in_amd_ci():
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH = 3000
 
+if is_blackwell_system():
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH = 3000
+
 
 def call_generate_lightllm(prompt, temperature, max_tokens, stop=None, url=None):
     assert url is not None
@@ -397,8 +409,6 @@ def _get_call_generate(args: argparse.Namespace):
         return partial(call_generate_vllm, url=f"{args.host}:{args.port}/generate")
     elif args.backend == "srt-raw":
         return partial(call_generate_srt_raw, url=f"{args.host}:{args.port}/generate")
-    elif args.backend == "gserver":
-        return partial(call_generate_gserver, url=f"{args.host}:{args.port}")
     elif args.backend == "outlines":
         return partial(call_generate_outlines, url=f"{args.host}:{args.port}/generate")
     elif args.backend == "guidance":
@@ -504,11 +514,12 @@ def popen_launch_server(
     base_url: str,
     timeout: float,
     api_key: Optional[str] = None,
-    other_args: list[str] = [],
+    other_args: Optional[list[str]] = None,
     env: Optional[dict] = None,
     return_stdout_stderr: Optional[tuple] = None,
     device: str = "auto",
     pd_separated: bool = False,
+    num_replicas: Optional[int] = None,
 ):
     """Launch a server process with automatic device detection.
 
@@ -516,17 +527,19 @@ def popen_launch_server(
         device: Device type ("auto", "cuda", "rocm" or "cpu").
                 If "auto", will detect available platforms automatically.
     """
+    other_args = other_args or []
+
     # Auto-detect device if needed
     if device == "auto":
         device = auto_config_device()
-        print(f"Auto-configed device: {device}", flush=True)
         other_args = list(other_args)
         other_args += ["--device", str(device)]
 
     _, host, port = base_url.split(":")
     host = host[2:]
 
-    if pd_separated:
+    use_mixed_pd_engine = not pd_separated and num_replicas is not None
+    if pd_separated or use_mixed_pd_engine:
         command = "sglang.launch_pd_server"
     else:
         command = "sglang.launch_server"
@@ -540,7 +553,7 @@ def popen_launch_server(
         *[str(x) for x in other_args],
     ]
 
-    if pd_separated:
+    if pd_separated or use_mixed_pd_engine:
         command.extend(
             [
                 "--lb-host",
@@ -559,6 +572,15 @@ def popen_launch_server(
             ]
         )
 
+    if use_mixed_pd_engine:
+        command.extend(
+            [
+                "--mixed",
+                "--num-replicas",
+                str(num_replicas),
+            ]
+        )
+
     if api_key:
         command += ["--api-key", api_key]
 
@@ -597,7 +619,6 @@ def popen_launch_server(
     start_time = time.perf_counter()
     with requests.Session() as session:
         while time.perf_counter() - start_time < timeout:
-
             return_code = process.poll()
             if return_code is not None:
                 # Server failed to start (non-zero exit code) or crashed
@@ -1149,7 +1170,7 @@ def run_bench_offline_throughput(model, other_args):
         *[str(x) for x in other_args],
     ]
 
-    print(f"{command=}")
+    print(f"command={' '.join(command)}")
     process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
 
     try:
@@ -1608,6 +1629,9 @@ class CustomTestCase(unittest.TestCase):
             max_retry=max_retry,
         )
 
+    def setUp(self):
+        print(f"[Test Method] {self._testMethodName}", flush=True)
+
 
 def dump_bench_raw_result(
     path: str,
@@ -1641,15 +1665,26 @@ def _ensure_remove_suffix(text: str, suffix: str):
     return text.removesuffix(suffix)
 
 
-class ModelDeploySetup:
-    def __init__(self, model_path: str, extra_args: List[str] = []):
+class ModelLaunchSettings:
+    def __init__(
+        self,
+        model_path: str,
+        tp_size: int = 1,
+        extra_args: Optional[List[str]] = None,
+        env: Optional[dict] = None,
+    ):
         self.model_path = model_path
-        if "--enable-multimodal" not in extra_args:
-            extra_args.append("--enable-multimodal")
-        if "--trust-remote-code" not in extra_args:
-            extra_args.append("--trust-remote-code")
+        self.tp_size = tp_size
+        self.extra_args = list(extra_args) if extra_args else []
+        self.env = env
+
+        if self.tp_size > 1 and "--tp" not in self.extra_args:
+            self.extra_args.extend(["--tp", str(self.tp_size)])
 
-        self.extra_args = extra_args
+        fixed_args = ["--enable-multimodal", "--trust-remote-code"]
+        for fixed_arg in fixed_args:
+            if fixed_arg not in self.extra_args:
+                self.extra_args.append(fixed_arg)
 
 
 class ModelEvalMetrics:
@@ -1782,3 +1817,33 @@ def write_results_to_json(model, metrics, mode="a"):
 
     with open("results.json", "w") as f:
         json.dump(existing_results, f, indent=2)
+
+
+def intel_amx_benchmark(extra_args=None, min_throughput=None):
+    def decorator(test_func):
+        @wraps(test_func)
+        def wrapper(self):
+            common_args = [
+                "--attention-backend",
+                "intel_amx",
+                "--disable-radix",
+                "--trust-remote-code",
+            ]
+            full_args = common_args + (extra_args or [])
+
+            model = test_func(self)
+            prefill_latency, decode_throughput, decode_latency = run_bench_one_batch(
+                model, full_args
+            )
+
+            print(f"{model=}")
+            print(f"{prefill_latency=}")
+            print(f"{decode_throughput=}")
+            print(f"{decode_latency=}")
+
+            if is_in_ci() and min_throughput is not None:
+                self.assertGreater(decode_throughput, min_throughput)
+
+        return wrapper
+
+    return decorator

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.5.3rc2"
+__version__ = "0.5.4"

{sglang-0.5.3rc2.dist-info → sglang-0.5.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sglang
-Version: 0.5.3rc2
+Version: 0.5.4
 Summary: SGLang is a fast serving framework for large language models and vision language models.
 License:                                  Apache License
                                    Version 2.0, January 2004
@@ -218,10 +218,11 @@ Requires-Dist: blobfile==3.0.0
 Requires-Dist: build
 Requires-Dist: compressed-tensors
 Requires-Dist: cuda-python
+Requires-Dist: decord2
 Requires-Dist: datasets
 Requires-Dist: einops
 Requires-Dist: fastapi
-Requires-Dist: flashinfer_python==0.4.0rc3
+Requires-Dist: flashinfer_python==0.4.1
 Requires-Dist: hf_transfer
 Requires-Dist: huggingface_hub
 Requires-Dist: interegular
@@ -243,34 +244,37 @@ Requires-Dist: psutil
 Requires-Dist: py-spy
 Requires-Dist: pybase64
 Requires-Dist: pydantic
-Requires-Dist: pynvml
+Requires-Dist: nvidia-ml-py
 Requires-Dist: python-multipart
 Requires-Dist: pyzmq>=25.1.2
 Requires-Dist: requests
 Requires-Dist: scipy
 Requires-Dist: sentencepiece
 Requires-Dist: setproctitle
-Requires-Dist: sgl-kernel==0.3.14.post1
+Requires-Dist: sgl-kernel==0.3.16.post3
 Requires-Dist: soundfile==0.13.1
 Requires-Dist: tiktoken
 Requires-Dist: timm==1.0.16
 Requires-Dist: torch==2.8.0
-Requires-Dist: torch_memory_saver==0.0.9rc2
+Requires-Dist: torch_memory_saver==0.0.9
 Requires-Dist: torchao==0.9.0
 Requires-Dist: torchaudio==2.8.0
 Requires-Dist: torchvision
 Requires-Dist: tqdm
-Requires-Dist: transformers==4.57.0
+Requires-Dist: transformers==4.57.1
 Requires-Dist: uvicorn
 Requires-Dist: uvloop
-Requires-Dist: xgrammar==0.1.24
+Requires-Dist: xgrammar==0.1.25
 Requires-Dist: grpcio==1.75.1
 Requires-Dist: grpcio-tools==1.75.1
-Provides-Extra: decord
-Requires-Dist: decord; extra == "decord"
+Requires-Dist: grpcio-reflection==1.75.1
+Requires-Dist: grpcio-health-checking==1.75.1
+Provides-Extra: modelopt
+Requires-Dist: nvidia-modelopt; extra == "modelopt"
 Provides-Extra: test
 Requires-Dist: accelerate; extra == "test"
 Requires-Dist: expecttest; extra == "test"
+Requires-Dist: gguf; extra == "test"
 Requires-Dist: jsonlines; extra == "test"
 Requires-Dist: matplotlib; extra == "test"
 Requires-Dist: pandas; extra == "test"
@@ -278,24 +282,28 @@ Requires-Dist: peft; extra == "test"
 Requires-Dist: pytest; extra == "test"
 Requires-Dist: sentence_transformers; extra == "test"
 Requires-Dist: tabulate; extra == "test"
+Provides-Extra: checkpoint-engine
+Requires-Dist: checkpoint-engine==0.1.2; extra == "checkpoint-engine"
+Provides-Extra: all
+Provides-Extra: dev
+Requires-Dist: sglang[test]; extra == "dev"
+Provides-Extra: cu130
+Requires-Dist: torch==2.9.0; extra == "cu130"
+Requires-Dist: torchaudio==2.9.0; extra == "cu130"
+Requires-Dist: torchvision==0.24.0; extra == "cu130"
+Provides-Extra: cu130-all
+Requires-Dist: sglang[test]; extra == "cu130-all"
+Requires-Dist: sglang[decord]; extra == "cu130-all"
+Requires-Dist: sglang[cu130]; extra == "cu130-all"
 Provides-Extra: tracing
 Requires-Dist: opentelemetry-api; extra == "tracing"
 Requires-Dist: opentelemetry-exporter-otlp; extra == "tracing"
 Requires-Dist: opentelemetry-exporter-otlp-proto-grpc; extra == "tracing"
 Requires-Dist: opentelemetry-sdk; extra == "tracing"
-Provides-Extra: all
-Requires-Dist: sglang[test]; extra == "all"
-Requires-Dist: sglang[decord]; extra == "all"
-Provides-Extra: all-aarch64
-Requires-Dist: sglang[test]; extra == "all-aarch64"
-Provides-Extra: dev
-Requires-Dist: sglang[test]; extra == "dev"
-Requires-Dist: sglang[decord]; extra == "dev"
 Provides-Extra: blackwell
-Requires-Dist: sglang[test]; extra == "blackwell"
-Requires-Dist: sglang[decord]; extra == "blackwell"
+Requires-Dist: sglang[dev]; extra == "blackwell"
 Provides-Extra: blackwell-aarch64
-Requires-Dist: sglang[test]; extra == "blackwell-aarch64"
+Requires-Dist: sglang[dev]; extra == "blackwell-aarch64"
 Dynamic: license-file
 
 <div align="center" id="sglangtop">
@@ -320,18 +328,20 @@ Dynamic: license-file
 | [**Slides**](https://github.com/sgl-project/sgl-learning-materials?tab=readme-ov-file#slides) |
 
 ## News
+- [2025/09] 🔥 Deploying DeepSeek on GB200 NVL72 with PD and Large Scale EP (Part II): 3.8x Prefill, 4.8x Decode Throughput ([blog](https://lmsys.org/blog/2025-09-25-gb200-part-2/)).
+- [2025/09] 🔥 SGLang Day 0 Support for DeepSeek-V3.2 with Sparse Attention ([blog](https://lmsys.org/blog/2025-09-29-deepseek-V32/)).
 - [2025/08] 🔔 SGLang x AMD SF Meetup on 8/22: Hands-on GPU workshop, tech talks by AMD/xAI/SGLang, and networking ([Roadmap](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_sglang_roadmap.pdf), [Large-scale EP](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_sglang_ep.pdf), [Highlights](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_highlights.pdf), [AITER/MoRI](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_aiter_mori.pdf), [Wave](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_wave.pdf)).
-- [2025/08] 🔥 SGLang provides day-0 support for OpenAI gpt-oss model ([instructions](https://github.com/sgl-project/sglang/issues/8833))
-- [2025/06] 🔥 SGLang, the high-performance serving infrastructure powering trillions of tokens daily, has been awarded the third batch of the Open Source AI Grant by a16z ([a16z blog](https://a16z.com/advancing-open-source-ai-through-benchmarks-and-bold-experimentation/)).
-- [2025/06] 🔥 Deploying DeepSeek on GB200 NVL72 with PD and Large Scale EP (Part I): 2.7x Higher Decoding Throughput ([blog](https://lmsys.org/blog/2025-06-16-gb200-part-1/)).
-- [2025/05] 🔥 Deploying DeepSeek with PD Disaggregation and Large-scale Expert Parallelism on 96 H100 GPUs ([blog](https://lmsys.org/blog/2025-05-05-large-scale-ep/)).
-- [2025/03] Supercharge DeepSeek-R1 Inference on AMD Instinct MI300X ([AMD blog](https://rocm.blogs.amd.com/artificial-intelligence/DeepSeekR1-Part2/README.html))
+- [2025/08] SGLang provides day-0 support for OpenAI gpt-oss model ([instructions](https://github.com/sgl-project/sglang/issues/8833))
+- [2025/05] Deploying DeepSeek with PD Disaggregation and Large-scale Expert Parallelism on 96 H100 GPUs ([blog](https://lmsys.org/blog/2025-05-05-large-scale-ep/)).
 - [2025/03] SGLang Joins PyTorch Ecosystem: Efficient LLM Serving Engine ([PyTorch blog](https://pytorch.org/blog/sglang-joins-pytorch/))
 - [2024/12] v0.4 Release: Zero-Overhead Batch Scheduler, Cache-Aware Load Balancer, Faster Structured Outputs ([blog](https://lmsys.org/blog/2024-12-04-sglang-v0-4/)).
 
 <details>
 <summary>More</summary>
 
+- [2025/06] SGLang, the high-performance serving infrastructure powering trillions of tokens daily, has been awarded the third batch of the Open Source AI Grant by a16z ([a16z blog](https://a16z.com/advancing-open-source-ai-through-benchmarks-and-bold-experimentation/)).
+- [2025/06] Deploying DeepSeek on GB200 NVL72 with PD and Large Scale EP (Part I): 2.7x Higher Decoding Throughput ([blog](https://lmsys.org/blog/2025-06-16-gb200-part-1/)).
+- [2025/03] Supercharge DeepSeek-R1 Inference on AMD Instinct MI300X ([AMD blog](https://rocm.blogs.amd.com/artificial-intelligence/DeepSeekR1-Part2/README.html))
 - [2025/02] Unlock DeepSeek-R1 Inference Performance on AMD Instinct™ MI300X GPU ([AMD blog](https://rocm.blogs.amd.com/artificial-intelligence/DeepSeekR1_Perf/README.html))
 - [2025/01] SGLang provides day one support for DeepSeek V3/R1 models on NVIDIA and AMD GPUs with DeepSeek-specific optimizations. ([instructions](https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3), [AMD blog](https://www.amd.com/en/developer/resources/technical-articles/amd-instinct-gpus-power-deepseek-v3-revolutionizing-ai-development-with-sglang.html), [10+ other companies](https://x.com/lmsysorg/status/1887262321636221412))
 - [2024/10] The First SGLang Online Meetup ([slides](https://github.com/sgl-project/sgl-learning-materials?tab=readme-ov-file#the-first-sglang-online-meetup)).
@@ -344,14 +354,15 @@ Dynamic: license-file
 </details>
 
 ## About
-SGLang is a fast serving framework for large language models and vision language models.
-It makes your interaction with models faster and more controllable by co-designing the backend runtime and frontend language.
-The core features include:
+SGLang is a high-performance serving framework for large language models and vision-language models.
+It is designed to deliver low-latency and high-throughput inference across a wide range of setups, from a single GPU to large distributed clusters.
+Its core features include:
 
-- **Fast Backend Runtime**: Provides efficient serving with RadixAttention for prefix caching, zero-overhead CPU scheduler, prefill-decode disaggregation, speculative decoding, continuous batching, paged attention, tensor/pipeline/expert/data parallelism, structured outputs, chunked prefill, quantization (FP4/FP8/INT4/AWQ/GPTQ), and multi-lora batching.
-- **Flexible Frontend Language**: Offers an intuitive interface for programming LLM applications, including chained generation calls, advanced prompting, control flow, multi-modal inputs, parallelism, and external interactions.
-- **Extensive Model Support**: Supports a wide range of generative models (Llama, Qwen, DeepSeek, Kimi, GPT, Gemma, Mistral, etc.), embedding models (e5-mistral, gte, mcdse) and reward models (Skywork), with easy extensibility for integrating new models.
-- **Active Community**: SGLang is open-source and backed by an active community with wide industry adoption.
+- **Fast Backend Runtime**: Provides efficient serving with RadixAttention for prefix caching, a zero-overhead CPU scheduler, prefill-decode disaggregation, speculative decoding, continuous batching, paged attention, tensor/pipeline/expert/data parallelism, structured outputs, chunked prefill, quantization (FP4/FP8/INT4/AWQ/GPTQ), and multi-LoRA batching.
+- **Extensive Model Support**: Supports a wide range of generative models (Llama, Qwen, DeepSeek, Kimi, GLM, GPT, Gemma, Mistral, etc.), embedding models (e5-mistral, gte, mcdse), and reward models (Skywork), with easy extensibility for integrating new models. Compatible with most Hugging Face models and OpenAI APIs.
+- **Extensive Hardware Support**: Runs on NVIDIA GPUs (GB200/B300/H100/A100/Spark), AMD GPUs (MI355/MI300), Intel Xeon CPUs, Google TPUs, Ascend NPUs, and more.
+- **Flexible Frontend Language**: Offers an intuitive interface for programming LLM applications, supporting chained generation calls, advanced prompting, control flow, multi-modal inputs, parallelism, and external interactions.
+- **Active Community**: SGLang is open-source and supported by a vibrant community with widespread industry adoption, powering over 300,000 GPUs worldwide.
 
 ## Getting Started
 - [Install SGLang](https://docs.sglang.ai/get_started/install.html)
@@ -367,7 +378,8 @@ Learn more in the release blogs: [v0.2 blog](https://lmsys.org/blog/2024-07-25-s
 [Development Roadmap (2025 H2)](https://github.com/sgl-project/sglang/issues/7736)
 
 ## Adoption and Sponsorship
-SGLang has been deployed at large scale, generating trillions of tokens in production each day. It is trusted and adopted by a wide range of leading enterprises and institutions, including xAI, AMD, NVIDIA, Intel, LinkedIn, Cursor, Oracle Cloud, Google Cloud, Microsoft Azure, AWS, Atlas Cloud, Voltage Park, Nebius, DataCrunch, Novita, InnoMatrix, MIT, UCLA, the University of Washington, Stanford, UC Berkeley, Tsinghua University, Jam & Tea Studios, Baseten, and other major technology organizations across North America and Asia. As an open-source LLM inference engine, SGLang has become the de facto industry standard, with deployments running on over 1,000,000 GPUs worldwide.
+SGLang has been deployed at large scale, generating trillions of tokens in production each day. It is trusted and adopted by a wide range of leading enterprises and institutions, including xAI, AMD, NVIDIA, Intel, LinkedIn, Cursor, Oracle Cloud, Google Cloud, Microsoft Azure, AWS, Atlas Cloud, Voltage Park, Nebius, DataCrunch, Novita, InnoMatrix, MIT, UCLA, the University of Washington, Stanford, UC Berkeley, Tsinghua University, Jam & Tea Studios, Baseten, and other major technology organizations across North America and Asia. As an open-source LLM inference engine, SGLang has become the de facto industry standard, with deployments running on over 300,000 GPUs worldwide.
+SGLang is currently hosted under the non-profit open-source organization [LMSYS](https://lmsys.org/about/).
 
 <img src="https://raw.githubusercontent.com/sgl-project/sgl-learning-materials/refs/heads/main/slides/adoption.png" alt="logo" width="800" margin="10px"></img>