sglang 0.3.1__py3-none-any.whl → 0.3.1.post2__py3-none-any.whl

sglang/srt/server_args.py

@@ -21,6 +21,8 @@ import logging
 import random
 from typing import List, Optional, Union
 
+from sglang.srt.utils import is_hip
+
 logger = logging.getLogger(__name__)
 
 
@@ -59,6 +61,7 @@ class ServerArgs:
     tp_size: int = 1
     stream_interval: int = 1
     random_seed: Optional[int] = None
+    constrained_json_whitespace_pattern: Optional[str] = None
 
     # Logging
     log_level: str = "info"
@@ -94,11 +97,12 @@ class ServerArgs:
     disable_cuda_graph_padding: bool = False
     disable_disk_cache: bool = False
     disable_custom_all_reduce: bool = False
+    disable_mla: bool = False
     enable_mixed_chunk: bool = False
     enable_torch_compile: bool = False
+    max_torch_compile_bs: int = 32
     torchao_config: str = ""
     enable_p2p_check: bool = False
-    enable_mla: bool = False
     triton_attention_reduce_in_fp32: bool = False
 
     # LoRA
@@ -152,11 +156,12 @@ class ServerArgs:
             )
             self.sampling_backend = "pytorch"
 
-        # Default kernel backends
-        if self.enable_mla:
-            logger.info("MLA optimization is tunred on. Use triton backend.")
+        # ROCm: flashinfer available later
+        if is_hip():
             self.attention_backend = "triton"
+            self.sampling_backend = "pytorch"
 
+        # Default kernel backends
         if self.attention_backend is None:
             self.attention_backend = "flashinfer"
 
@@ -359,6 +364,12 @@ class ServerArgs:
             default=ServerArgs.random_seed,
             help="The random seed.",
         )
+        parser.add_argument(
+            "--constrained-json-whitespace-pattern",
+            type=str,
+            default=ServerArgs.constrained_json_whitespace_pattern,
+            help=r"Regex pattern for syntactic whitespaces allowed in JSON constrained output. For example, to allow the model generate consecutive whitespaces, set the pattern to [\n\t ]*",
+        )
         parser.add_argument(
             "--log-level",
             type=str,
@@ -488,6 +499,11 @@ class ServerArgs:
             default=False,
             help="Disable the custom all-reduce kernel and fall back to NCCL.",
         )
+        parser.add_argument(
+            "--disable-mla",
+            action="store_true",
+            help="Disable Multi-head Latent Attention (MLA) for DeepSeek-V2.",
+        )
         parser.add_argument(
             "--enable-mixed-chunk",
             action="store_true",
@@ -498,6 +514,12 @@ class ServerArgs:
             action="store_true",
             help="Optimize the model with torch.compile. Experimental feature.",
         )
+        parser.add_argument(
+            "--max-torch-compile-bs",
+            type=int,
+            default=ServerArgs.max_torch_compile_bs,
+            help="Set the maximum batch size when using torch compile.",
+        )
         parser.add_argument(
             "--torchao-config",
             type=str,
@@ -509,11 +531,6 @@ class ServerArgs:
             action="store_true",
             help="Enable P2P check for GPU access, otherwise the p2p access is allowed by default.",
         )
-        parser.add_argument(
-            "--enable-mla",
-            action="store_true",
-            help="Enable Multi-head Latent Attention (MLA) for DeepSeek-V2.",
-        )
         parser.add_argument(
             "--triton-attention-reduce-in-fp32",
             action="store_true",
@@ -532,7 +549,8 @@ class ServerArgs:
             type=str,
             nargs="*",
             default=None,
-            help="The list of LoRA adapters.",
+            action=LoRAPathAction,
+            help="The list of LoRA adapters. You can provide a list of either path in str or renamed path in the format {name}={path}",
         )
         parser.add_argument(
             "--max-loras-per-batch",
@@ -590,3 +608,14 @@ class PortArgs:
     controller_port: int
     detokenizer_port: int
     nccl_ports: List[int]
+
+
+class LoRAPathAction(argparse.Action):
+    def __call__(self, parser, namespace, values, option_string=None):
+        setattr(namespace, self.dest, {})
+        for lora_path in values:
+            if "=" in lora_path:
+                name, path = lora_path.split("=", 1)
+                getattr(namespace, self.dest)[name] = path
+            else:
+                getattr(namespace, self.dest)[lora_path] = lora_path

sglang/srt/utils.py

@@ -51,6 +51,11 @@ show_time_cost = False
 time_infos = {}
 
 
+# torch flag AMD GPU
+def is_hip() -> bool:
+    return torch.version.hip is not None
+
+
 def enable_show_time_cost():
     global show_time_cost
     show_time_cost = True
@@ -187,7 +192,7 @@ def allocate_init_ports(
         cur_port += 1
 
     if port is not None and ret_ports[0] != port:
-        logger.warn(
+        logger.warning(
             f"WARNING: Port {port} is not available. Use port {ret_ports[0]} instead."
         )
 
@@ -623,56 +628,7 @@ def set_ulimit(target_soft_limit=65535):
         try:
             resource.setrlimit(resource_type, (target_soft_limit, current_hard))
         except ValueError as e:
-            logger.warn(f"Fail to set RLIMIT_NOFILE: {e}")
-
-
-def is_llama3_405b_fp8_head_16(model_config):
-    """Return whether the model is meta-llama/Meta-Llama-3.1-405B-FP8 with 16 kv heads."""
-    if (
-        model_config.hf_config.architectures[0] == "LlamaForCausalLM"
-        and model_config.hf_config.hidden_size == 16384
-        and model_config.hf_config.intermediate_size == 53248
-        and model_config.hf_config.num_hidden_layers == 126
-        and model_config.hf_config.num_key_value_heads == 16
-        and hasattr(model_config.hf_config, "quantization_config")
-        and model_config.hf_config.quantization_config["quant_method"] == "fbgemm_fp8"
-    ):
-        return True
-    return False
-
-
-def monkey_patch_vllm_qvk_linear_loader():
-    """A temporary hack to fix the num_heads for meta-llama/Meta-Llama-3.1-405B-FP8 checkpoints."""
-    from vllm.model_executor.layers.linear import QKVParallelLinear
-
-    origin_weight_loader = QKVParallelLinear.weight_loader
-
-    def get_original_weight(loaded_weight, head_dim):
-        n_kv_head = loaded_weight.shape[0] // (2 * head_dim)
-        dim = loaded_weight.shape[1]
-        for i in range(n_kv_head):
-            loaded_weight[i * head_dim : (i + 1) * head_dim, :] = loaded_weight[
-                2 * i * head_dim : (2 * i + 1) * head_dim, :
-            ]
-        original_kv_weight = loaded_weight[: n_kv_head * head_dim, :]
-        assert original_kv_weight.shape == (n_kv_head * head_dim, dim)
-        return original_kv_weight
-
-    def weight_loader_srt(
-        self,
-        param: Parameter,
-        loaded_weight: torch.Tensor,
-        loaded_shard_id: Optional[str] = None,
-    ):
-        if (
-            loaded_shard_id in ["k", "v"]
-            and loaded_weight.shape[0] == self.head_size * self.total_num_kv_heads * 2
-        ):
-            loaded_weight = get_original_weight(loaded_weight, self.head_size)
-
-        origin_weight_loader(self, param, loaded_weight, loaded_shard_id)
-
-    setattr(QKVParallelLinear, "weight_loader", weight_loader_srt)
+            logger.warning(f"Fail to set RLIMIT_NOFILE: {e}")
 
 
 def add_api_key_middleware(app, api_key: str):

sglang/test/few_shot_gsm8k.py

@@ -44,7 +44,7 @@ def get_answer_value(answer_str):
         return INVALID
 
 
-def main(args):
+def run_eval(args):
     # Select backend
     set_default_backend(RuntimeEndpoint(f"{args.host}:{args.port}"))
 
@@ -119,6 +119,12 @@ def main(args):
     # Dump results
     dump_state_text("tmp_output_gsm8k.txt", states)
 
+    return {
+        "accuracy": acc,
+        "latency": latency,
+        "output_throughput": output_throughput,
+    }
+
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
@@ -129,4 +135,4 @@ if __name__ == "__main__":
     parser.add_argument("--host", type=str, default="http://127.0.0.1")
     parser.add_argument("--port", type=int, default=30000)
     args = parser.parse_args()
-    main(args)
+    run_eval(args)

sglang/test/test_utils.py

@@ -22,6 +22,7 @@ from sglang.lang.backend.runtime_endpoint import RuntimeEndpoint
 from sglang.srt.utils import kill_child_process
 from sglang.utils import get_exception_traceback
 
+DEFAULT_FP8_MODEL_NAME_FOR_TEST = "neuralmagic/Meta-Llama-3.1-8B-FP8"
 DEFAULT_MODEL_NAME_FOR_TEST = "meta-llama/Meta-Llama-3.1-8B-Instruct"
 DEFAULT_MOE_MODEL_NAME_FOR_TEST = "mistralai/Mixtral-8x7B-Instruct-v0.1"
 DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH = 600
@@ -304,7 +305,6 @@ def add_common_sglang_args_and_parse(parser: argparse.ArgumentParser):
 def select_sglang_backend(args: argparse.Namespace):
     if args.backend.startswith("srt"):
         if args.backend == "srt-no-parallel":
-            global_config.enable_parallel_decoding = False
             global_config.enable_parallel_encoding = False
         backend = RuntimeEndpoint(f"{args.host}:{args.port}")
     elif args.backend.startswith("gpt-"):

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.3.1"
+__version__ = "0.3.1.post2"

{sglang-0.3.1.dist-info → sglang-0.3.1.post2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sglang
-Version: 0.3.1
+Version: 0.3.1.post2
 Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License: Apache License
                                    Version 2.0, January 2004
@@ -269,7 +269,7 @@ Requires-Dist: peft; extra == "test"
 
 --------------------------------------------------------------------------------
 
-| [**Blog**](https://lmsys.org/blog/2024-07-25-sglang-llama3/) | [**Paper**](https://arxiv.org/abs/2312.07104) | [**Slack**](https://join.slack.com/t/sgl-fru7574/shared_invite/zt-2ngly9muu-t37XiH87qvD~6rVBTkTEHw) |
+| [**Blog**](https://lmsys.org/blog/2024-07-25-sglang-llama3/) | [**Paper**](https://arxiv.org/abs/2312.07104) | [**Join Slack**](https://join.slack.com/t/sgl-fru7574/shared_invite/zt-2ngly9muu-t37XiH87qvD~6rVBTkTEHw) | [**Join Weekly Development Meeting**](https://calendar.app.google/v2Tw3kuHkKYyp8VV7) |
 
 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.
@@ -278,7 +278,7 @@ The core features include:
 - **Fast Backend Runtime**: Provides efficient serving with RadixAttention for prefix caching, jump-forward constrained decoding, continuous batching, token attention (paged attention), tensor parallelism, FlashInfer kernels, chunked prefill, and quantization (INT4/FP8/AWQ/GPTQ).
 - **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 3, Gemma 2, Mistral, QWen, DeepSeek, LLaVA, etc.) and embedding models (e5-mistral), with easy extensibility for integrating new models.
-- **Active Community**: SGLang is open-source and backed by an active community with industry adoption, welcoming contributions to improve LLM and VLM serving.
+- **Active Community**: SGLang is open-source and backed by an active community with industry adoption.
 
 ## News
 - [2024/09] 🔥 SGLang v0.3 Release: 7x Faster DeepSeek MLA, 1.5x Faster torch.compile, Multi-Image/Video LLaVA-OneVision ([blog](https://lmsys.org/blog/2024-09-04-sglang-v0-3/)).
@@ -318,7 +318,7 @@ pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.4/
 ### Method 2: From source
 ```
 # Use the last release branch
-git clone -b v0.3.1 https://github.com/sgl-project/sglang.git
+git clone -b v0.3.1.post2 https://github.com/sgl-project/sglang.git
 cd sglang
 
 pip install --upgrade pip
@@ -483,7 +483,6 @@ python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct
 - To enable torch.compile acceleration, add `--enable-torch-compile`. It accelerates small models on small batch sizes.
 - To enable fp8 weight quantization, add `--quantization fp8` on a fp16 checkpoint or directly load a fp8 checkpoint without specifying any arguments.
 - To enable fp8 kv cache quantization, add `--kv-cache-dtype fp8_e5m2`.
-- To enable DeepSeek MLA acceleration, add `--enable-mla`.
 - If the model does not have a chat template in the Hugging Face tokenizer, you can specify a [custom chat template](docs/en/custom_chat_template.md).
 - To run tensor parallelism on multiple nodes, add `--nnodes 2`. If you have two nodes with two GPUs on each node and want to run TP=4, let `sgl-dev-0` be the hostname of the first node and `50000` be an available port.
 ```

{sglang-0.3.1.dist-info → sglang-0.3.1.post2.dist-info}/RECORD

@@ -1,18 +1,19 @@
 sglang/__init__.py,sha256=T8MYdFfKFPZcgFKHMBpOCIlFbhjwmr77Nqm6mdE6bCY,1590
 sglang/api.py,sha256=pH4CjwOXUweL5MF1sIkFMddDxfnF7PyUxEHC5kvNVbI,6468
-sglang/bench_latency.py,sha256=EvmXpaREU-g25OTcOUTgAUPmA-txfnyjaqY-4hlq97w,16925
-sglang/bench_serving.py,sha256=6OM5JIDuoxJDg-VLE4ijGGcS8-6ViaidV05lIrZmSzo,36239
+sglang/bench_latency.py,sha256=bA50iUYOxEnLjzY2S4AgwxtSAqujUbGfQFwbLZj5XNc,17160
+sglang/bench_server_latency.py,sha256=KvFJgKQTSons7KOG0CBqnnOOx1gW29bBM1Z3GQO_6-E,5599
+sglang/bench_serving.py,sha256=3gIJ1O2x51Fwd4wYJjgwluTbWKXL-azckQte7YC5zIc,36261
 sglang/check_env.py,sha256=rGRABCgt-0SfUrow4px28b2P59aMn8eVTnN5eZc_a8s,5397
-sglang/global_config.py,sha256=KWpXd4OCCWW2TRQo-dShvLs4jb15ej9Ejhxr_wggzBg,1535
+sglang/global_config.py,sha256=38id86i3tRGCSOFZlN1LM01a3xt-V98xuNgKGG9boCk,1058
 sglang/launch_server.py,sha256=UnjNjYuZ8TtvmRtgYEsFImkbvCwvn_tQjk0V7cHy67E,450
 sglang/launch_server_llavavid.py,sha256=olPKyhozi1coCwoRMwBRYWsTFByrgus9CwPSeNmskgc,1002
 sglang/utils.py,sha256=NA_4xUrTI7KICQ3PEACfNWKE3nxSA5QvQZJNd4TQrDc,9395
-sglang/version.py,sha256=r4xAFihOf72W9TD-lpMi6ntWSTKTP2SlzKP1ytkjRbI,22
+sglang/version.py,sha256=U9F0UlFDynnYN5dX-kxehylWCwXo9a6E6W4FfDusfRg,28
 sglang/lang/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/lang/chat_template.py,sha256=uqI_I9zIKXGXg7-W-yjqvx1ZeS_TuwFCms6wkmC2QmY,13411
 sglang/lang/choices.py,sha256=-W1DVw9N9ZliVpvmWrzIXG4cswAah8eMQrHWzkS3D8o,6234
 sglang/lang/compiler.py,sha256=o1C6G3TzhjSlsH-doTPy5oiVehr57dxNTa5oZw5TTAI,7639
-sglang/lang/interpreter.py,sha256=M42SuOnijFaHWOe3Qyi-bNanRt-mYhSDa1wWn1J42Hw,30324
+sglang/lang/interpreter.py,sha256=rOquFbMzxry7IItZlAn5TwtQfxMy718JPxOkiXO-yrg,30234
 sglang/lang/ir.py,sha256=W3UfZikcGeT86PDDjDjw-yNzrKY2e2UYO4DTatMCfm0,17704
 sglang/lang/tracer.py,sha256=borJmlSJOhg1RUndGRnilnR60eEZz2Y9aU7BpftsOxU,8287
 sglang/lang/backend/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -25,82 +26,83 @@ sglang/lang/backend/vertexai.py,sha256=O-iBLD-y3vq80UxnrAoJri7bxpgd-_eakZ88Cf8bE
 sglang/srt/conversation.py,sha256=S5w5V6G1xigNxa3UQoSxRcMpQLWWDT9EPBoHBvHkSAk,19663
 sglang/srt/hf_transformers_utils.py,sha256=6HlqcmGPIvnSGaEEICeuzwag1QylSoSGbXRVvUdIMDo,6016
 sglang/srt/mm_utils.py,sha256=zox644S3IHUWmADdK4MnIbdTS2DWHOy0_Dq0gCU38QQ,12273
-sglang/srt/server.py,sha256=FNmTpX7E9fVWj_NFzp4AtE5ODaA_rg5Xm8uZ0FB0X4o,20041
-sglang/srt/server_args.py,sha256=5OHH3gaO1s5Y2UQw2_FnFxwxrsqnUQ_WNqP1R1IWUAA,21877
-sglang/srt/utils.py,sha256=pckOt7gyQfJaV3-h8FPurWyrPij5_EBUX_Xp7x6y6YM,24229
+sglang/srt/server.py,sha256=n4QRn36_t-HAH-lSME3tiZSCUGRQwqMUckgs0paHq5g,20179
+sglang/srt/server_args.py,sha256=3XjDt6SSjTfbOe0HSXA--2aUvrpWSnQmAHYwmeS1-M0,23159
+sglang/srt/utils.py,sha256=8yxiMRttCcfswynkNPWD3yZFNAGFz2P1PzSuxHCBGns,22340
 sglang/srt/configs/__init__.py,sha256=292SuEorST-lAq2Uvsv2M7yC28uYZlssVvRDsF-bZCQ,86
 sglang/srt/configs/exaone.py,sha256=Duxd4yQoKy8GWEzZD_kCY_OzmN_67CTJL_Kgn0eXk3g,10731
 sglang/srt/configs/model_config.py,sha256=OqHrucJQHbH-wxgkGj-Dcx_B888uUGASpLRjz40HaLY,6651
 sglang/srt/constrained/__init__.py,sha256=ze8awDPvwAzdeMwzJ-25kXOQ4nVWoaP55jBDt5UOS_4,2070
 sglang/srt/constrained/base_tool_cache.py,sha256=5sazBMHHDpHMoqOjuY6itCxwTmIFCflIWEDXMtmrPVs,2006
-sglang/srt/constrained/fsm_cache.py,sha256=jItSvCu_XrAgltfejwgvdltaiT98-8lJGBe_84cSnTk,2786
+sglang/srt/constrained/fsm_cache.py,sha256=k7DRUAaiLTEX5UarfJ17gEYQ-QWQAGfvykjYFkM_Y2U,2982
 sglang/srt/constrained/jump_forward.py,sha256=9_HxmXtWjr5S6a5e0cBimbY3ZhiLiJC74V6jIqDXfuo,6575
-sglang/srt/layers/activation.py,sha256=JEXNTgqxoiU4N-gVm4XMjobhft4JKDcMrgTkfpsRUzM,4856
-sglang/srt/layers/attention_backend.py,sha256=39P3iMs7B1iEzCA3EHdqUp3BLafeIVFnFWGzpEhlTRk,18182
+sglang/srt/layers/activation.py,sha256=i3omgj3GdUIZBqJNUjpdJsMc2UM3Lx07FT2J1WICrqA,5171
+sglang/srt/layers/attention_backend.py,sha256=lqMsY4VaOO_szIWoTAinXf1DnP2UsbF32kzvwFySz9w,18119
 sglang/srt/layers/flashinfer_utils.py,sha256=jyaO7XiEisFZg_dfaCbfRCHSHSKYoM1wOzfHa0h1q14,7413
-sglang/srt/layers/layernorm.py,sha256=RXuS4UyksatqTF6lSK7VYyEiUEnBiNIBlEn8q4w84UA,3404
+sglang/srt/layers/layernorm.py,sha256=p_7bnmSpJ_slpoP0Gk5wQPpHtLllUu3imSIRBqGqTP0,3737
 sglang/srt/layers/logits_processor.py,sha256=Js2qSk1Z3uPL2cYO1ARai51f2i8OedV3qdwByQVSJtI,12439
 sglang/srt/layers/pooler.py,sha256=qNMG3Ycvt2yf9mk1Lcs-2K7oPeCuVeDYoHAxkMu9b_Q,1610
 sglang/srt/layers/radix_attention.py,sha256=EcVO0fUSmgvE_9R-MlpgJq0O_uT8ACuHzbMi19bANYc,1874
-sglang/srt/layers/sampler.py,sha256=1BKsZbSLBGFVtTJo1LsThuoRjOSOnsL1AiwFxJNIXRs,5800
+sglang/srt/layers/sampler.py,sha256=Y0o1bndTGRD713fHMbN5-LRUiyneBkb7bH_QlkkeqSs,3836
 sglang/srt/layers/torchao_utils.py,sha256=rTECwKSXhj_ylh_iSzfbopz9_lZOFHatquQrNJNLZlE,2703
 sglang/srt/layers/fused_moe/__init__.py,sha256=bWCrDdOy2ANEXTb8CHYO63O3Iu3eZnn0PJbgl0z5vvE,75
 sglang/srt/layers/fused_moe/fused_moe.py,sha256=1WM2cObWXcFWtqh_utGJFPnrT344rORwuQ9hJDaH2s0,23104
-sglang/srt/layers/fused_moe/layer.py,sha256=GT3r2UPx_PAufJd0SUMOXyh76ymAeYDubd0SM0H71bo,20977
+sglang/srt/layers/fused_moe/layer.py,sha256=raFyvPzjYz-Fv8B3IcOxQYKKCWqXis5mXwg1GFE61y4,22243
 sglang/srt/layers/triton_attention/decode_attention.py,sha256=XCQTX0kUttT1AG5FRMgfQbiXgvoempYD0UR2r6D_vJg,16711
 sglang/srt/layers/triton_attention/extend_attention.py,sha256=XTUTMrE-5jfMEufQUifZ-8NJQABSPcF47qhnNT5Z1iI,11050
 sglang/srt/layers/triton_attention/prefill_attention.py,sha256=QkXPcT02c13zha2M4mBm2S5dh_sS-Gc4FkkrcywRqvc,5377
 sglang/srt/lora/lora.py,sha256=ksj866lgDul6zxO30Jm7Nrjv-mFAMrzdvP8sez3Pl6U,14938
 sglang/srt/lora/lora_config.py,sha256=paVB7F7SIuxr_vodvKf8zzAlH2fdVYHhXxcXV62D0Vo,1411
-sglang/srt/lora/lora_manager.py,sha256=Q7rk1SMEZ75wda68rAZDGVyX_o8ZdIW2I5Fo_llaqHs,9475
+sglang/srt/lora/lora_manager.py,sha256=7J7cGmyy1Ph4HCvLdM-ViAizAbV1snZqD-S7JLWXasI,9561
 sglang/srt/managers/controller_multi.py,sha256=KolZDso2WqH1ZhQw9p1eTmlFRgo4bcvzBxE44_sNE_o,6300
 sglang/srt/managers/controller_single.py,sha256=DiZALP_iIPZQMRx09a-LwT5_Dg7p-WU8HXyMoxJ9sRA,4955
 sglang/srt/managers/detokenizer_manager.py,sha256=yQkL5gLomLiy1qc6e9HNz8hcj7JQFHm1AfIrzpXaWJE,6852
-sglang/srt/managers/io_struct.py,sha256=bqmL3NDPLqOn6Au3WLF0NOe8Dh7ECMN7BTHCkEZ_Edk,11247
-sglang/srt/managers/policy_scheduler.py,sha256=tiBUi2GJU5eQEBK6HfsO1_YjWtFkougo40954DIp4dM,13026
-sglang/srt/managers/schedule_batch.py,sha256=QfixWzh7ks60eYE52mZHfUseXqcb89h4ZO1Aur3weLU,27340
+sglang/srt/managers/io_struct.py,sha256=yNV5BmeUzLPqv19j79kXQ50Iaqdk4vP-_TciiRf4OEE,11396
+sglang/srt/managers/policy_scheduler.py,sha256=PVo0DV0-5ODNN7FkPkeF1Y8BQ6uuLldPETOlB_YvvL4,11560
+sglang/srt/managers/schedule_batch.py,sha256=ns2qkaYAvzul-LCV1BEB6q1t5jKyftNsReMv62PC8M0,27386
 sglang/srt/managers/tokenizer_manager.py,sha256=ql-sObjl1oRigJwnLtqqTaaw-i7gPTDMoNXDEMftr40,29643
-sglang/srt/managers/tp_worker.py,sha256=Zbl_tFUAsD6Qv1fUEJCn_jyUc3JjDm33yI3Nmu1HY8w,39174
+sglang/srt/managers/tp_worker.py,sha256=0Y0k-roDrBxWZxD0axv5CCvUUW8vsJ8n78TANHLzEFs,39503
 sglang/srt/mem_cache/base_prefix_cache.py,sha256=qEQwEkG4E5rab2ZoTqcesf5pR_J4nV2jBxIHsBJHtIM,924
 sglang/srt/mem_cache/chunk_cache.py,sha256=CjZZYlqQzq7mYOiBMLWA5XNb6HIyh5lIMdY-K0OUZEc,2368
 sglang/srt/mem_cache/flush_cache.py,sha256=pTLKPRB17U6vl5RFJJvuJ4jCL2SyomgkUBNlkDpGRqo,978
 sglang/srt/mem_cache/memory_pool.py,sha256=4br3Ea2bfA-YsF_sPOVHlF2zQzYGd8fVaYTp197yZsE,7871
 sglang/srt/mem_cache/radix_cache.py,sha256=0AVr1BKKDOtTyybUkwxrz6PT8khDx-DpzgN5MgL27IE,10088
-sglang/srt/model_executor/cuda_graph_runner.py,sha256=LngmwtBcvobJ_9G8lD966SihjmMJlgMgHe_ZogK1kDg,10090
+sglang/srt/model_executor/cuda_graph_runner.py,sha256=gZ0Wukqz6u67MMIj4MC8JET9jcHdh0rotYzpuPlHruY,10512
 sglang/srt/model_executor/forward_batch_info.py,sha256=yvkhayY9Zu6gysoojcGT73lADGOtfHKkFKWdJLRyACI,6141
-sglang/srt/model_executor/model_runner.py,sha256=7jBSCdZxyDLWMOdwv1vRa7Oue-xbp8lA6I11ZPKFdAc,23457
+sglang/srt/model_executor/model_runner.py,sha256=X7AG1k9AI_kqS8q1i5Bfv-kFysIdqJAVWMGGZoAPThY,22726
 sglang/srt/models/baichuan.py,sha256=NrG1rMJXhemkrUCEf8xKOSDQVsOD-nN8RQz6MWHOg84,15124
 sglang/srt/models/chatglm.py,sha256=KwxLHBEvK02McXDvBS0gnRxfIvOAu2QP7lgibrj9Nbc,13371
 sglang/srt/models/commandr.py,sha256=2rAXRZRb4PkJZ4NWEqP_rIgsjxbdZyHpuoMOarqTWzQ,14163
 sglang/srt/models/dbrx.py,sha256=N_0Ku_p1NCsc29NktUBNqPv7Z33XhYxOZK5xN7nzW4s,14661
 sglang/srt/models/deepseek.py,sha256=7UJgde1EV9ey6d-CKRcEyTKh1_WhZdatpZiltIuqpik,16006
-sglang/srt/models/deepseek_v2.py,sha256=3D9WtPvVOu8U40x_KOksnmWBLmLIcgtV958go8NSj5Q,28307
+sglang/srt/models/deepseek_v2.py,sha256=1J0pt1jZRcBBGYbgt1wGiuxPcrdpfTEUEaGFqju6TVA,28431
 sglang/srt/models/exaone.py,sha256=3I5ZoiLotf7U-8c9QJRubpgf6JDx9I_z-ViXQlCC-x8,13087
 sglang/srt/models/gemma.py,sha256=GkwgGFHgGlXgBZN7s7Wooz5tMyCp1YtgLahU2NOo66M,12273
 sglang/srt/models/gemma2.py,sha256=sFfCNEm0_OOWElRSTDuroRv8wNMX8v_81Uko9m546KA,14923
 sglang/srt/models/gpt_bigcode.py,sha256=kzHYogeGXZF4KHpkXA-RGqvs016mA-6klWxD2QJTi9E,10195
 sglang/srt/models/grok.py,sha256=6I4OwQwNyAbh5GF24_SRm12XYBvM9iGWB-T4TSTJ0wU,14929
 sglang/srt/models/internlm2.py,sha256=6j7JH0p3yib8GZDH8Cmrs-pgwfH3eOlAK6V3Cq64O7w,12202
-sglang/srt/models/llama.py,sha256=tjdjlIxJr31vgbzGBP_el9RgYxw1kzvmqnVinnTVVUw,15259
-sglang/srt/models/llama_classification.py,sha256=A2ABTUD5u4XoWv1dsIPU7wcCQP3jhbDJblMhLgaiFBA,3402
+sglang/srt/models/llama.py,sha256=nbJwRcG9DnurVNSGLKJjnmBmTXP1_5WZpudth_0PVpw,15216
+sglang/srt/models/llama_classification.py,sha256=HF-69J9qIYdfX0R5wEtIgvafMzprKcXdvF3W_orl_kA,3394
 sglang/srt/models/llama_embedding.py,sha256=RI2mpYheP5WwhuTINU-6IrU61usuMyCK9h2zDEyLW4g,3458
 sglang/srt/models/llava.py,sha256=O4XGdl70Hh4tM_OHapFGHbReC82mbe9xLw6GELKWKhU,24881
 sglang/srt/models/llavavid.py,sha256=ou5uIuskBoBo0lXvqFFfDLBYYVfehx27n-Lu8X9gpLs,11992
 sglang/srt/models/minicpm.py,sha256=ioqCsTCE_oF8xqGF5fm5cK9dclK5Y0EQ1UJfyteIDDo,13825
-sglang/srt/models/minicpm3.py,sha256=S7bNeCAsfvL44Vn350KLaqX674SCb4CpUuDnhjLjr3U,25113
+sglang/srt/models/minicpm3.py,sha256=McPWyy2fQqfHUhi9Nk36rkvvPAS8RmLOY7Vh4ah5c1w,25216
 sglang/srt/models/mistral.py,sha256=tiYoKjyYVzlQl52QUZ33odD2yCxj9dxcqln474VuZOw,744
 sglang/srt/models/mixtral.py,sha256=oRC7mKBrPJhvzkWSabrbeQQQac-jtF4EV6H2Sgjc5JY,13897
 sglang/srt/models/mixtral_quant.py,sha256=wMACJq78OTWj7HlqPDRNEh8cjrVAjKqJEsOG3CO5xow,14072
+sglang/srt/models/olmoe.py,sha256=d0ECpU-IXXwGYg9tkVeMARUbqVcqEnWfpH3rrNiGKA0,15336
 sglang/srt/models/qwen.py,sha256=nqSRzkiZzpRVG6WGQ1MBUclQnXyw8jlvoOq-euM8j5s,9954
 sglang/srt/models/qwen2.py,sha256=9_M-VkHN1_T1XN-gsl_L636QMQ9BLF2WqvTcx_1L6aw,12432
 sglang/srt/models/qwen2_moe.py,sha256=s7b5XnSvsBYtZZUkjPp442m59CqPJ3HxGUIwXBVWsXw,17153
 sglang/srt/models/stablelm.py,sha256=30ngpc0Xq3VxzXJlf6svP1oax8Q3krMJkxM8PVKtZWU,11359
-sglang/srt/models/xverse.py,sha256=luhp_90ZNkTpXHDCURO4MZBy1vbvHTVCwSe4PYYLWBs,13701
-sglang/srt/models/xverse_moe.py,sha256=YR--WZ33G7XEMsS7ZJl1cQ62Q8PDo9gWqpvJBY_cb-M,15886
+sglang/srt/models/xverse.py,sha256=L3g32-je_7JmzF2-hztaIVshHYCIv7jOM3oFs-fb2MY,13658
+sglang/srt/models/xverse_moe.py,sha256=CgDD9cR83UVfTsPU6WcbHVYBrkYKv_kTdwncTIx7Q7U,15842
 sglang/srt/models/yivl.py,sha256=B6MELthWIm5KdSzX3o2tbbpApY8XdjUdmcQSD4dQe_I,4835
 sglang/srt/openai_api/adapter.py,sha256=CJ47YftRHAip1FMcHIhtCorBtzlIkv7F0Wz_JUcI4T4,51032
 sglang/srt/openai_api/protocol.py,sha256=rdSwUAoO5-KLemJOE50xwSUagxY4T1QIiNyCYsTtCi0,9868
-sglang/srt/sampling/sampling_batch_info.py,sha256=vkwy59Jt51FESYukmwDKwPbCM45WMb16dx_408B3oqc,7900
+sglang/srt/sampling/sampling_batch_info.py,sha256=GewqyxCrW2PFwuzGHaCR59Pvw6j0n2dKGrlJWYQWwW4,6149
 sglang/srt/sampling/sampling_params.py,sha256=ggOXxafqfCD-xrGYcM57byLZ79CIeBP4AD5F44L_CW0,5635
 sglang/srt/sampling/penaltylib/__init__.py,sha256=5vQw0Y5DSzmsoFg1IdMIKLwFVhYZ5ArADHVBYbSmOec,513
 sglang/srt/sampling/penaltylib/orchestrator.py,sha256=WkTNeDhj9H9rtp2ZZeX6MS2sdKSGlLboE6FcuKrwUo0,10815
@@ -108,7 +110,7 @@ sglang/srt/sampling/penaltylib/penalizers/frequency_penalty.py,sha256=IvYioX53Vq
 sglang/srt/sampling/penaltylib/penalizers/min_new_tokens.py,sha256=XJZP0C4NFyXgcODbIWXxrgVEjmRgqLdZuVAtoN-LveY,3565
 sglang/srt/sampling/penaltylib/penalizers/presence_penalty.py,sha256=0PlANTrR959foTA3Nj5qBE7ndaOZgG-9X6LhzlmEUc8,2533
 sglang/srt/sampling/penaltylib/penalizers/repetition_penalty.py,sha256=v9jOgA0-I31WcrhIydiFbpy2ZJPLytFLGM98NRPd2sU,2820
-sglang/test/few_shot_gsm8k.py,sha256=uSHEPvUFbAgWKtaqxkhBpQrQV_SlTk0HN9FhjNLpL4g,3731
+sglang/test/few_shot_gsm8k.py,sha256=To7Sdg-DLF8poIQLwiOBYKbkz-1C_gn6H79vIbyPR-o,3860
 sglang/test/run_eval.py,sha256=NWxeLWmInBgkCvC9Jr_QzF7GfAiBve3Gf1JQrEOlNlU,3899
 sglang/test/runners.py,sha256=ZoWhT1TDXfLBVdbivXx1KUu9dhPlGjL_xrP18WLzVLo,11404
 sglang/test/simple_eval_common.py,sha256=r0G-9QLycs2ax3RMc44T_61fzMxlpTzv6pececC7lyY,12379
@@ -120,10 +122,10 @@ sglang/test/simple_eval_mmlu.py,sha256=FkwamjGMjueTixymkedF-YiPloSLiy4ftILFUrKZ9
 sglang/test/test_activation.py,sha256=jkdNRzJnbd5OgZliQaIXpxovlcky17UrweomcOcMxoE,1442
 sglang/test/test_layernorm.py,sha256=IacByD5d-stXjzBz8Ypamc7povlcedpKPbb_4JLgo3c,3720
 sglang/test/test_programs.py,sha256=3-XKnppQdCNWjaJb6jwib5Z9OSpgKvH8SFLJbE4J9qI,17001
-sglang/test/test_utils.py,sha256=iBs07MBFxOidipTG1-s2hrCvcURFJVXo7gg10pzAQX8,17168
+sglang/test/test_utils.py,sha256=dsHRd1xLzcjlarxUnDIz2XEHfut7HvqVPwx2Fn7vf10,17179
 sglang/test/srt/sampling/penaltylib/utils.py,sha256=-0p0rV-P4lNo7xAe3rQSBHTubc50a-DFyOQmLGAkgkQ,12515
-sglang-0.3.1.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-sglang-0.3.1.dist-info/METADATA,sha256=QKZQ7PjuK22x_QlQy1LqPX6y4zLgJJ9FPoNNSkw3cEk,38125
-sglang-0.3.1.dist-info/WHEEL,sha256=cVxcB9AmuTcXqmwrtPhNK88dr7IR_b6qagTj0UvIEbY,91
-sglang-0.3.1.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
-sglang-0.3.1.dist-info/RECORD,,
+sglang-0.3.1.post2.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+sglang-0.3.1.post2.dist-info/METADATA,sha256=WxMy8Ur_rjPxqVOoWSFoM3eBHWt0cKGyrtwOUfWL-Vc,38114
+sglang-0.3.1.post2.dist-info/WHEEL,sha256=GV9aMThwP_4oNCtvEC2ec3qUYutgWeAzklro_0m4WJQ,91
+sglang-0.3.1.post2.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.3.1.post2.dist-info/RECORD,,

{sglang-0.3.1.dist-info → sglang-0.3.1.post2.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (74.1.2)
+Generator: setuptools (75.1.0)
 Root-Is-Purelib: true
 Tag: py3-none-any