sglang 0.4.3__py3-none-any.whl → 0.4.3.post2__py3-none-any.whl

sglang/srt/models/qwen2_vl.py

@@ -31,8 +31,9 @@ import torch
 import torch.nn as nn
 import torch.nn.functional as F
 from einops import rearrange
+from transformers import Qwen2VLConfig
+from transformers.models.qwen2_vl.configuration_qwen2_vl import Qwen2VLVisionConfig
 
-from sglang.srt.configs import Qwen2VLConfig, Qwen2VLVisionConfig
 from sglang.srt.hf_transformers_utils import get_processor
 from sglang.srt.layers.activation import QuickGELU
 from sglang.srt.layers.attention.vision import VisionAttention

sglang/srt/openai_api/adapter.py

@@ -20,12 +20,14 @@ import os
 import time
 import uuid
 from http import HTTPStatus
-from typing import Dict, List, Optional
+from typing import Dict, List
 
 from fastapi import HTTPException, Request, UploadFile
 from fastapi.responses import ORJSONResponse, StreamingResponse
 from pydantic import ValidationError
 
+from sglang.lang.chat_template import get_chat_template_by_model_path
+
 try:
     from outlines.fsm.json_schema import convert_json_schema_to_str
 except ImportError:
@@ -92,7 +94,6 @@ file_id_response: Dict[str, FileResponse] = {}
 # map file id to file path in SGLang backend
 file_id_storage: Dict[str, str] = {}
 
-
 # backend storage directory
 storage_dir = None
 
@@ -116,12 +117,13 @@ def create_streaming_error_response(
     return json_str
 
 
-def load_chat_template_for_openai_api(tokenizer_manager, chat_template_arg):
+def load_chat_template_for_openai_api(tokenizer_manager, chat_template_arg, model_path):
     global chat_template_name
 
     logger.info(
         f"Use chat template for the OpenAI-compatible API server: {chat_template_arg}"
     )
+
     if not chat_template_exists(chat_template_arg):
         if not os.path.exists(chat_template_arg):
             raise RuntimeError(
@@ -163,6 +165,18 @@ def load_chat_template_for_openai_api(tokenizer_manager, chat_template_arg):
     else:
         chat_template_name = chat_template_arg
 
+    # check chat-template
+    chat_template = get_chat_template_by_model_path(model_path)
+    if chat_template is not None:
+        official_chat_template = chat_template.name
+        used_chat_template = chat_template_name
+        if official_chat_template != used_chat_template:
+            logger.warning(
+                f"Using a chat_template: '{used_chat_template}', "
+                f"which is different from official chat template: '{official_chat_template}', "
+                f"This discrepancy may lead to performance degradation."
+            )
+
 
 async def v1_files_create(file: UploadFile, purpose: str, file_storage_pth: str = None):
     try:

sglang/srt/server_args.py

@@ -262,14 +262,17 @@ class ServerArgs:
             )
 
         # Speculative Decoding
-        if self.speculative_algorithm == "EAGLE":
+        if (
+            self.speculative_algorithm == "EAGLE"
+            or self.speculative_algorithm == "NEXTN"
+        ):
             self.prefill_only_one_req = True
             self.disable_cuda_graph_padding = True
             self.disable_radix_cache = True
             self.disable_overlap_schedule = True
             self.chunked_prefill_size = -1
             logger.info(
-                "The radix cache, chunked prefill, and overlap scheduler are disabled because of using eagle speculative decoding."
+                f"The radix cache, chunked prefill, and overlap scheduler are disabled because of using {self.speculative_algorithm} speculative decoding."
             )
 
         # GGUF
@@ -705,7 +708,7 @@ class ServerArgs:
         parser.add_argument(
             "--speculative-algorithm",
             type=str,
-            choices=["EAGLE"],
+            choices=["EAGLE", "NEXTN"],
             help="Speculative algorithm.",
         )
         parser.add_argument(

sglang/srt/speculative/eagle_worker.py

@@ -24,6 +24,7 @@ from sglang.srt.speculative.eagle_utils import (
     fast_topk,
     select_top_k_tokens,
 )
+from sglang.srt.speculative.spec_info import SpeculativeAlgorithm
 
 logger = logging.getLogger(__name__)
 
@@ -57,11 +58,15 @@ class EAGLEWorker(TpModelWorker):
         # Parse arguments
         self.topk = server_args.speculative_eagle_topk
         self.speculative_num_steps = server_args.speculative_num_steps
+        self.speculative_algorithm = SpeculativeAlgorithm.from_string(
+            server_args.speculative_algorithm
+        )
         self.server_args = server_args
 
         # Share the embedding and lm_head
-        embed, head = self.target_worker.model_runner.model.get_embed_and_head()
-        self.model_runner.model.set_embed_and_head(embed, head)
+        if not self.speculative_algorithm.is_nextn():
+            embed, head = self.target_worker.model_runner.model.get_embed_and_head()
+            self.model_runner.model.set_embed_and_head(embed, head)
         self.model_runner.server_args.disable_cuda_graph = backup_disable_cuda_graph
 
         # Create multi-step attn backends and cuda graph runners

sglang/srt/speculative/spec_info.py

@@ -5,18 +5,28 @@ class SpeculativeAlgorithm(IntEnum):
     NONE = auto()
     EAGLE = auto()
 
+    # NEXTN spec decoding is for DeepSeek V3/R1
+    # currently it's implemented based on EAGLE
+    NEXTN = auto()
+
     def is_none(self):
         return self == SpeculativeAlgorithm.NONE
 
     def is_eagle(self):
-        return self == SpeculativeAlgorithm.EAGLE
+        return self == SpeculativeAlgorithm.EAGLE or self == SpeculativeAlgorithm.NEXTN
+
+    def is_nextn(self):
+        return self == SpeculativeAlgorithm.NEXTN
 
     @staticmethod
     def from_string(name: str):
         name_map = {
             "EAGLE": SpeculativeAlgorithm.EAGLE,
+            "NEXTN": SpeculativeAlgorithm.NEXTN,
             None: SpeculativeAlgorithm.NONE,
         }
+        if name is not None:
+            name = name.upper()
         return name_map[name]
 
 

sglang/utils.py

@@ -306,22 +306,112 @@ def download_and_cache_file(url: str, filename: Optional[str] = None):
     return filename
 
 
-def execute_shell_command(command: str) -> subprocess.Popen:
+import fcntl
+
+
+def is_in_ci():
+    from sglang.test.test_utils import is_in_ci
+
+    return is_in_ci()
+
+
+LOCKFILE = os.path.expanduser("~/.sglang_port_lock")
+PORT_REGISTRY = os.path.expanduser("~/.sglang_port_registry.json")
+
+if not os.path.exists(LOCKFILE):
+    with open(LOCKFILE, "w") as f:
+        pass
+
+if not os.path.exists(PORT_REGISTRY):
+    with open(PORT_REGISTRY, "w") as f:
+        json.dump([], f)
+
+
+def print_highlight(html_content: str):
+    if is_in_ci():
+        html_content = str(html_content).replace("\n", "<br>")
+        display(HTML(f"<strong style='color: #00008B;'>{html_content}</strong>"))
+    else:
+        print(html_content)
+
+
+def init_port_registry():
+    """Initialize the port registry file if it doesn't exist."""
+    if not os.path.exists(PORT_REGISTRY):
+        with open(PORT_REGISTRY, "w") as f:
+            json.dump([], f)
+
+
+def reserve_port(start=30000, end=40000):
+    """
+    Reserve an available port using a file lock and a registry.
+    Returns the allocated port.
     """
-    Execute a shell command and return the process handle
+    init_port_registry()
+    with open(LOCKFILE, "w") as lock:
+        fcntl.flock(lock, fcntl.LOCK_EX)
+        try:
+            with open(PORT_REGISTRY, "r") as f:
+                used = json.load(f)
+        except Exception:
+            used = []
+        for port in range(start, end):
+            if port not in used:
+                used.append(port)
+                with open(PORT_REGISTRY, "w") as f:
+                    json.dump(used, f)
+                return port
+    raise RuntimeError("No free port available")
+
+
+def release_port(port):
+    """Release the reserved port by removing it from the registry."""
+    with open(LOCKFILE, "w") as lock:
+        fcntl.flock(lock, fcntl.LOCK_EX)
+        try:
+            with open(PORT_REGISTRY, "r") as f:
+                used = json.load(f)
+        except Exception:
+            used = []
+        if port in used:
+            used.remove(port)
+        with open(PORT_REGISTRY, "w") as f:
+            json.dump(used, f)
 
-    Args:
-        command: Shell command as a string (can include \\ line continuations)
-    Returns:
-        subprocess.Popen: Process handle
+
+def execute_shell_command(command: str) -> subprocess.Popen:
     """
-    # Replace \ newline with space and split
+    Execute a shell command and return its process handle.
+    """
+    # Replace newline continuations and split the command string.
     command = command.replace("\\\n", " ").replace("\\", " ")
     parts = command.split()
-
     return subprocess.Popen(parts, text=True, stderr=subprocess.STDOUT)
 
 
+def launch_server_cmd(command: str, host: str = "0.0.0.0", port: int = None):
+    """
+    Launch the server using the given command.
+    If no port is specified, a free port is reserved.
+    """
+    if port is None:
+        port = reserve_port()
+    full_command = f"{command} --port {port}"
+    process = execute_shell_command(full_command)
+    return process, port
+
+
+def terminate_process(process, port=None):
+    """
+    Terminate the process and, if a port was reserved, release it.
+    """
+    from sglang.srt.utils import kill_process_tree
+
+    kill_process_tree(process.pid)
+    if port is not None:
+        release_port(port)
+
+
 def wait_for_server(base_url: str, timeout: int = None) -> None:
     """Wait for the server to be ready by polling the /v1/models endpoint.
 
@@ -343,6 +433,7 @@ def wait_for_server(base_url: str, timeout: int = None) -> None:
                     NOTE: Typically, the server runs in a separate terminal.
                     In this notebook, we run the server and notebook code together, so their outputs are combined.
                     To improve clarity, the server logs are displayed in the original black color, while the notebook outputs are highlighted in blue.
+                    We are running those notebooks in a CI parallel environment, so the throughput is not representative of the actual performance.
                     """
                 )
                 break
@@ -353,17 +444,6 @@ def wait_for_server(base_url: str, timeout: int = None) -> None:
             time.sleep(1)
 
 
-def terminate_process(process):
-    from sglang.srt.utils import kill_process_tree
-
-    kill_process_tree(process.pid)
-
-
-def print_highlight(html_content: str):
-    html_content = str(html_content).replace("\n", "<br>")
-    display(HTML(f"<strong style='color: #00008B;'>{html_content}</strong>"))
-
-
 class TypeBasedDispatcher:
     def __init__(self, mapping: List[Tuple[Type, Callable]]):
         self._mapping = mapping

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.4.3"
+__version__ = "0.4.3.post2"

{sglang-0.4.3.dist-info → sglang-0.4.3.post2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: sglang
-Version: 0.4.3
+Version: 0.4.3.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
@@ -235,7 +235,7 @@ Requires-Dist: pyzmq>=25.1.2; extra == "runtime-common"
 Requires-Dist: torchao>=0.7.0; extra == "runtime-common"
 Requires-Dist: uvicorn; extra == "runtime-common"
 Requires-Dist: uvloop; extra == "runtime-common"
-Requires-Dist: xgrammar>=0.1.10; extra == "runtime-common"
+Requires-Dist: xgrammar==0.1.10; extra == "runtime-common"
 Requires-Dist: ninja; extra == "runtime-common"
 Provides-Extra: srt
 Requires-Dist: sglang[runtime_common]; extra == "srt"
@@ -243,7 +243,7 @@ Requires-Dist: cuda-python; extra == "srt"
 Requires-Dist: sgl-kernel>=0.0.3.post6; extra == "srt"
 Requires-Dist: torch; extra == "srt"
 Requires-Dist: vllm<=0.7.2,>=0.6.4.post1; extra == "srt"
-Requires-Dist: flashinfer_python>=0.2.1.post1; extra == "srt"
+Requires-Dist: flashinfer_python>=0.2.1.post2; extra == "srt"
 Requires-Dist: outlines<=0.1.11,>=0.0.44; extra == "srt"
 Provides-Extra: srt-hip
 Requires-Dist: sglang[runtime_common]; extra == "srt-hip"

{sglang-0.4.3.dist-info → sglang-0.4.3.post2.dist-info}/RECORD

@@ -9,20 +9,20 @@ sglang/check_env.py,sha256=lDVA3ybt1wOE33HIMpkkU7zGRgLWez1_ifRRJ8qxbtw,8445
 sglang/global_config.py,sha256=crt5cernXnDa1iQ8kGOq_ScTFclRlTQbJ-atFHM7I5I,1330
 sglang/launch_server.py,sha256=mDXfwha8LHpWQJekcCosR98QhCQsbmilsBlI5jAIgg0,420
 sglang/llama3_eval.py,sha256=gWSboDchIGybIce88bJlrCG0yiLZ513mw4gcutJlzGM,10017
-sglang/utils.py,sha256=7HpOrPBhMivWH719m7Dy1rjrAXOAsnqelpwNBBbvjqs,13319
-sglang/version.py,sha256=Nyg0pmk5ea9-SLCAFEIF96ByFx4-TJFtrqYPN-Zn6g4,22
+sglang/utils.py,sha256=9fm5ghtYPXqsWKjUzlQKJIoH5iFit6Rz21RhyaC3YL4,15673
+sglang/version.py,sha256=AJcJAUlaCr4igIr4vVaRiJeU678q5BdTdso-33Tq98k,28
 sglang/lang/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sglang/lang/chat_template.py,sha256=v4SyYViPHX3i3XT46F7vlARn4UaSiP3PBpTGtzO6uRY,17006
+sglang/lang/chat_template.py,sha256=0tZX67LgtYGrWopnSuTeqWVdxaw2deJOFWOBJpd6htU,17547
 sglang/lang/choices.py,sha256=-W1DVw9N9ZliVpvmWrzIXG4cswAah8eMQrHWzkS3D8o,6234
 sglang/lang/compiler.py,sha256=o1C6G3TzhjSlsH-doTPy5oiVehr57dxNTa5oZw5TTAI,7639
 sglang/lang/interpreter.py,sha256=r7x5mBxAOaEwmxjaMBMcn7N8HDFv6V6K9eINtffDygQ,33074
-sglang/lang/ir.py,sha256=dtA6rs5JIN0tMm3jhgRqdpRhH2Sckil-BMyLRMyBEIY,18494
+sglang/lang/ir.py,sha256=YQlEX2eYMAVHG12xJ2Jds6S6el45_O-udsXJumpEoEQ,18552
 sglang/lang/tracer.py,sha256=o-jLAPPSuy2vBfsGGrTAnbuWtORzQ50B4C_P5zvYkx8,8291
 sglang/lang/backend/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/lang/backend/anthropic.py,sha256=EXRX7xJgA5KZszX7toSLVnKzFQ5EO0Loj-YjHFtxSxg,2081
 sglang/lang/backend/base_backend.py,sha256=tdoh9YF3CyekY1BKiX9n7-aA4srDWIuA4RDJLM7q8qg,1985
 sglang/lang/backend/litellm.py,sha256=ugmL7sfUxkUHVbHtwNzHgdQAEd4UCjNQboFuE3KThcY,2450
-sglang/lang/backend/openai.py,sha256=ha9a2P6T80TmSgYlyIwB1qYawWkjcOgiOptkktkqa1U,15436
+sglang/lang/backend/openai.py,sha256=BQj1FHPXmSfFVQV-SIs7WW6v7tUDUckjtpvs9mhP8Ok,15645
 sglang/lang/backend/runtime_endpoint.py,sha256=gM97bi8Kv8sLzCDJnH5ZZTQ9I6t31CeVUve7qdTsopo,16755
 sglang/lang/backend/vertexai.py,sha256=O-iBLD-y3vq80UxnrAoJri7bxpgd-_eakZ88Cf8bEGA,4855
 sglang/srt/_custom_ops.py,sha256=7jL5BTcoS8PmR56y2Qsa3q8emI-tmrJuV4hLTwLVFBE,5040
@@ -30,21 +30,21 @@ sglang/srt/aio_rwlock.py,sha256=6LYtOdeTUY3hkfa1dmYkgsaF2ttrwIF3hUWz2AZ2fqw,2970
 sglang/srt/conversation.py,sha256=USUoYiJf5DdHz7Ouclu30k3QSxMiem4WgZrA148MpSA,21695
 sglang/srt/custom_op.py,sha256=M5oqlgh32vAVeStFCruydTUfi_blGFJihVTnQBEOvwo,1134
 sglang/srt/function_call_parser.py,sha256=YmagXt1BIuTbeiWmSleZwJFCFR5r5EFqVQqKnJDYXiE,19568
-sglang/srt/hf_transformers_utils.py,sha256=_24uqCkZ4dvS9Uc5p2cCzX0Q8ShUzrh_Hp6mvg7hxHY,7729
+sglang/srt/hf_transformers_utils.py,sha256=ymMz_MjaeHirDwzzCWz5ktPEzWdIoP3K9DiZqNtjs6k,7737
 sglang/srt/mm_utils.py,sha256=1ScBunw_x4W8ebM_AcJ62-1T2mfT8NlMJqdAhkF1lb0,12367
 sglang/srt/model_parallel.py,sha256=eLXZhvJ4wG6dh0FontNCIdVZvHYdWgaeY-5cu7TD9tE,6078
 sglang/srt/server.py,sha256=PrQb9r6L9syWHKlggbbiQYsKtpwSmECqozRbf8qnoV8,874
-sglang/srt/server_args.py,sha256=97YAjIbttBjty9Rv9CQ-yGERprFRSu2Jq0zX8Yx-QH8,41018
+sglang/srt/server_args.py,sha256=C7zyFuYidgt__ZaqK8tNV9zPByQNaLyUNMOogBzBjXM,41128
 sglang/srt/torch_memory_saver_adapter.py,sha256=--FgbrcvJxTcRe856plD9ktqgrHGPTE18eZCJlE50hY,1255
 sglang/srt/utils.py,sha256=RVU-OORgeVQICMPzj17KHxbDdSYGOKFBnNR4dZejP9A,46780
-sglang/srt/configs/__init__.py,sha256=Nvwtif0X9IYUtj0aL9XvAo_RRZcxTshsaliwc8djooU,347
+sglang/srt/configs/__init__.py,sha256=naCw3LwTLHOCsldy2UyRmxoIWrWfX3hgEP2Gt7frXaw,382
 sglang/srt/configs/chatglm.py,sha256=j-b0YkdYUmQm2y1kNmMJtKeACxWKmBbvNNkDWbs6kbI,2907
 sglang/srt/configs/dbrx.py,sha256=tdhIkXAQl1yr0MxqFmsDG1E0e2puRTTKm6UTyANBLac,11005
 sglang/srt/configs/device_config.py,sha256=kfmpPOECqYxcRoY-ko0QZRhyiBWUGP2CMF51DMUN5nU,435
 sglang/srt/configs/exaone.py,sha256=Duxd4yQoKy8GWEzZD_kCY_OzmN_67CTJL_Kgn0eXk3g,10731
 sglang/srt/configs/load_config.py,sha256=la2ezNRcUZs7qiTYta2KEXqZ0U4TcmWW3U0sjoHgQQ0,3107
-sglang/srt/configs/model_config.py,sha256=sQIOfslBRzhOjucZdd8zE8nO9PEOc7zc6cZMbguQgoY,16876
-sglang/srt/configs/qwen2vl.py,sha256=ZjLy9v2eZY4wptUfY3CWgYKg2B5DDrkfCSyTy_Zf_bg,4351
+sglang/srt/configs/model_config.py,sha256=MPC1XJox6wo0Ut1LJ-05flKWlA95ZuzVKaDP9il4hD4,17023
+sglang/srt/configs/qwen2_5_vl_config.py,sha256=J8jq6QwseIOgqXQ3nuEX_yRVMNbyYjleZbf4nEhniGk,48184
 sglang/srt/constrained/base_grammar_backend.py,sha256=JFQFiAZLSqV6vck-ewIEzEEyncWLbRz_gkvkqpC282k,3185
 sglang/srt/constrained/outlines_backend.py,sha256=yPYgz44n-rSCStGGkS1lGazFiQzN7gqwSvpJ2YG0co4,7081
 sglang/srt/constrained/outlines_jump_forward.py,sha256=iZWXeR3gNYoMubLGyFmLPO4V2YsN5DiGjD71Xk9iFaE,6418
@@ -61,7 +61,7 @@ sglang/srt/distributed/device_communicators/pynccl.py,sha256=G-Dut_QJHOUG0j7--Zq
 sglang/srt/distributed/device_communicators/pynccl_wrapper.py,sha256=LblisImY9d6EMz-oPS9J16WHo2Q_SRL1DtlJKK63Hfg,15349
 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/engine.py,sha256=3JL7aj0K5UHCBsqCOG2MTLAv-8IpwtN2K-3_xB6fNqk,16888
+sglang/srt/entrypoints/engine.py,sha256=t2UhSOnr22BBCyoIhDJZIcmthQOQcL1iaB06LZqqAnU,17555
 sglang/srt/entrypoints/http_server.py,sha256=TJlekPuw01_AvfAhDUdD-DaxCmmW_uH_rWL2CNv2OGE,19545
 sglang/srt/layers/activation.py,sha256=f9KGwGi2znUx5SFKH_vO8htpBkfQ550VZZIycFDfPlk,5602
 sglang/srt/layers/dp_attention.py,sha256=LLUMHIdphhQy1rNR52uwIFl85oDFPAsogMwYF3d83PU,1910
@@ -77,7 +77,7 @@ sglang/srt/layers/torchao_utils.py,sha256=Ws24FdRBSkTpyeyA6bQrdDm-W5wfDxKvSIPUSa
 sglang/srt/layers/vocab_parallel_embedding.py,sha256=txcjkuSDa6gZwESKj8X-HSLhAnMmDXL0FmFWY9SKqik,22155
 sglang/srt/layers/attention/__init__.py,sha256=KlQ0fl-o9v_NxBDhNZ4dPW2uQ2HeJjLm-0MTMWgaa28,2980
 sglang/srt/layers/attention/double_sparsity_backend.py,sha256=4mVyFPfZxPTwkQHGNCfI_4hQ8CbsWXJfxz-IQW77gAc,9143
-sglang/srt/layers/attention/flashinfer_backend.py,sha256=8bEnIXFXBB8lv0ruH_paqT8uLIHsKYKYhQ9ytzDcsYI,47681
+sglang/srt/layers/attention/flashinfer_backend.py,sha256=MnQCTvOOPXy4RXvvZNSMjm1nPHXoysQ1cWGVf_yilQU,50518
 sglang/srt/layers/attention/torch_native_backend.py,sha256=KrcAqTLVZLtwgOmB0xhwUUsX32M-5LYZpNxaRNT4VuA,9252
 sglang/srt/layers/attention/triton_backend.py,sha256=mbYaYKHYrUyL2zEXrPtDRIcvVNe6L-bmcdLhKF92V-0,21292
 sglang/srt/layers/attention/vision.py,sha256=zLjKmzUlkgq1RFcP3b4EPArOAKovoaDLgYfM5SyB2wM,13181
@@ -124,6 +124,7 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=RWBo3j5AzZls5eD2eaejetSfM
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json",sha256=kklgf2qLI5CQYiJJ5e9Gxx2gAfGxcyMDYpdJnIXPV8E,2748
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=800,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json",sha256=8e0tN_DHPwvh_HECVHx9oOF_4WWdaht4s6Nmd_K-aBU,2904
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=0aSYzpv_cBAlpWCPrfGgNTCfae1KdKQnT56E8XFQl7A,3262
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=XmKFaMheq7NNrsvYCJteul0w809l_l460ZiDQC9ToGs,3262
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=5ro9O8Nf-7MB8NGCQ3QV5kB2k3iSWe0rdTz2A4W72CA,3732
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=5ro9O8Nf-7MB8NGCQ3QV5kB2k3iSWe0rdTz2A4W72CA,3732
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=5ro9O8Nf-7MB8NGCQ3QV5kB2k3iSWe0rdTz2A4W72CA,3732
@@ -210,24 +211,30 @@ sglang/srt/layers/quantization/modelopt_quant.py,sha256=_VdVz77dTP-IczPeFrdH6Ttr
 sglang/srt/layers/quantization/w8a8_int8.py,sha256=PBapNo27rkUdX95E0ihUZeHTXP2Id3CwlNlMU8GIuZc,3344
 "sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=tkLjwLC_aVXhzuvo-2QHkojXZauPJsf3jNHFn1S7uRA,3244
 "sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=CPo1WRF0HgsQMPBkvpoImElQMrfwpJLhEvL86e6fkPU,3247
+"sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=9qdyh6ki9LAyq7VDO9WMRmBOPWKSrZhU-I7z1E9bTKA,550
 "sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=xqJNygFgPCe4jDpkfVOWCXpuNMUjmssdD8oGhp2iXv8,3726
 "sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=xqJNygFgPCe4jDpkfVOWCXpuNMUjmssdD8oGhp2iXv8,3726
 "sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=9ya9f1Nt0g0RD-6sRRBZOaIPLSpSFZCz7jNvqTPrgFE,3732
 "sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Qoj9rLLRDbKM4IKBCXvN8RcxzSmNPd0TQUiM7CXDqHI,3241
 "sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=UZljnxxCSjwnZlX3OgKWZJGXCf5BWF_agEpNX8I4Zxc,3248
 "sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=4D3Ku4y7BCVEJzueKvQC_KvOR026w3ONWsxfsA_YrEc,3249
+"sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=6POXzQZHdNwcBDv1w6BJKbLMRDt0jbFUuMsMNf-ToEs,549
 "sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=7v4tp0RaT4vxF4urSBrkK5FR_5ikeFQ1htF3DwDl1lk,3249
 "sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=3matoCzEe4aexwoe7YTmkjyE4NA8khWXjL5EySuNwzA,3254
 "sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=0J2MFgaLkv-mfVE5x363lgVKYU6miLG_xRO3tJUga_M,3249
+"sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=JeXNLkbMAjdDKV-WpzQy87SXN06towo3xUofLtvYCQI,551
 "sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=4ubbhwSFX_XbefRLEkLoWxJkcetFWPzsszPu0X3_Wrw,3242
 "sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=9f8Ib4gLEFSfdNpO8IL8uiONImvqnlPbJrZ0HM3OB-o,3247
 "sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=FhyniGTx5QeCuVrBSVTQys6q05Pr5lPEcPykpAX7Iyo,3247
+"sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Boa83ZSPZ0LvzqtfLGvois5QK4TmJfwjA2n96c9ET58,549
 "sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=0v17v78pETXv6S2ZoibekxOVhiTmCm807DYG4DONUck,3259
 "sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=I44PvJj758-sw_fCOVROLTpG0NQ5_5PCYyQcpZC1YSY,3259
 "sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=tG5_iVeRBHTgHX-liOf79nWRjj_lUZ-NQWTbBrBgORQ,3246
 "sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=-CVHqClROli9FWe_FnlnuAG2LiFivDFK_nghH6t-BWc,3261
+"sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=9tCZxJ0eAD7AYMH7OqS3AGppJUllKnJLNvMq7FMXdsA,552
 "sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=GsLoYkaZ2p4Qu0Coj-X90s7JWyfZBOloIHPlyNKSIes,3246
 "sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=BpKweY4HfBx5xvqSvwNahy9x7R5FH-YK8j6rLFQdKwo,3248
+"sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=QMVfMXS0Yjgob8_9xps1xuZi6KnY5l2MeKxXLRjTeg4,548
 "sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=I6a5wQ0R3cBp8ix_PDamWZN2aJmV_1p0tg430L3Updg,3727
 "sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=I6a5wQ0R3cBp8ix_PDamWZN2aJmV_1p0tg430L3Updg,3727
 "sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=bSxTaptdcgj27mQGmdUmQtYTn4V_8EcmtRaVNigKjLA,3730
@@ -240,6 +247,7 @@ sglang/srt/layers/quantization/w8a8_int8.py,sha256=PBapNo27rkUdX95E0ihUZeHTXP2Id
 "sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Mtw7a9BSspj2TzC-aPxE82o1LEvwzgbUuIofwRxUNA0,3263
 "sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=B0lo3SuoQXhBEnojH2TwpVeurvlKD8yI8kQrJ5ORhWU,3249
 "sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=NHdx3tZnfLF7NplswMzcTRbQEQFLtChg4rd7GU9lMbM,3262
+"sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=ZYZ03ziPGwG2sCasEYLj7ZIP7vNO8UNBR5qNTmKgRMs,549
 "sglang/srt/layers/quantization/configs/N=36864,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=laYeH4w0iZOj2Yg3vDgtKoroNQnwBEX4GUGLrO9095I,3260
 "sglang/srt/layers/quantization/configs/N=36864,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=TWcPDZ2miQMD6OWDC1FteRs80ND9RC-oJL3PLVmJbtI,3257
 "sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=6HvsrapGzLyCAZe_NE0VdcJTSJv1aztD1ZqFG7VODUA,3729
@@ -266,22 +274,28 @@ sglang/srt/layers/quantization/w8a8_int8.py,sha256=PBapNo27rkUdX95E0ihUZeHTXP2Id
 "sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=EgFTGyW_YuDwyEDUCoGglyI1ETdj9J7AR0UfJ86jMoI,3249
 "sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=BpKweY4HfBx5xvqSvwNahy9x7R5FH-YK8j6rLFQdKwo,3248
 "sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=NiorJgOotxkQcP49ID3z5al1UA4QQDrT8MvbCwAWL5Y,3248
+"sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=FcuzcmKFf2RbaUpAaAsuObUefcGMgNPMDbVdHXRkoGY,549
 "sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=bPQWtvaJrzOOIgI-R-MIxs_f4yC_FobkDydu3OkOFtg,3252
 "sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=h32lCTFilLlyKbMeuJvNWG1v0yJJzNj93kwSvlrHfaY,3249
 "sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=ZRgiuHZ2SFC6u-WV5DGwau4k1RiPLI67eENO0e-5Ylg,3253
+"sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=kjQ_kvF38bZGcmaeJGSJsSR0NcUjUOh3LZ2-5c4kPvE,550
 "sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=-hP_P8NM0K04mGzTmpGBNibQ5xxh5gPz5WtoMXhoz1E,3253
 "sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=0jX-z2lTgVw7ABLmWsIsQdqW4EjmbXKRDHye_XPLCAE,3245
 "sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=FB5Le4obvPoCgFSnC_3-Uh59n-Mt4Rol8saXVcK3RPw,3252
+"sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=mZGU39sobtUqNYKjtyIGjhOZyCOQFJMF3MinA1zjTJA,550
 "sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=kLviGvVngpgOuelfKtvv9Is7MWQ89rGxlomMRP6t0Ic,3250
 "sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=_exM3wJ3FMmGHweBcH-8IxwZBzaOmPaF3ScMM6KDpiY,3253
+"sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=ue2oWml2ouUTZelYx5Nt5pgCmY-ib3mLV1reJL9ZudE,550
 "sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=vlys0Zi_CaaU41OHGbWSBtbVglFi98bgqEySBMc9Sdg,3258
 "sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=YWyByOlKSqp5lbcUa8eu6N2dHRKJqJDbCDSjdDQJngg,3249
 "sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=ENRWYdUwI0ooHb6IwcHliupRWOPnw-7-WtxZB-qQGJI,3245
 "sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=j5PTW0IC4Z2yQIygcdICaOsvb639u6Mv-ZpJYkrBQ2k,3254
+"sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Nv9KP_KLGsRJdJF755dZBvbTws37u1GM2UigMRlAtl0,552
 "sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Ggy4hejkcWjiw5Bi-wGzSP5JLVuvOjip_rbjXFBJZbs,3257
 "sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Xy4mgZx5iiEvuv2ydO4dFNIT8s0jgBhNHE1vu93fGJM,3250
 "sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=TdWuE2RIsIyr4Im24MuWK3XyiNtbhO_hAiAXDz5gNUk,3246
 "sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=x476nFeltB_2iO9_6y-z2P_unAbh7ghLPFi5z2LOTOo,3253
+"sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=ZOWEo__oUy8AhJiAlRCuGNAZNdNweFdWBFptJYkwxs8,552
 "sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=sVbH4YRLTxBqvTh_6xbtXkj3orOrKytlwM-_4gtD6IY,3725
 "sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=sVbH4YRLTxBqvTh_6xbtXkj3orOrKytlwM-_4gtD6IY,3725
 "sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=sTvaJ0RiCaQem4F1z7oES6RVRJ2gKgBuccX13S1SqGc,3733
@@ -318,9 +332,9 @@ sglang/srt/managers/cache_controller.py,sha256=DXnIunJgtTws1WF2vZOYVQe56vacV7Mn4
 sglang/srt/managers/configure_logging.py,sha256=aY9xExurz7t_IdItd-9GuVuM7kEGB8_bRryhZxKdu9o,1542
 sglang/srt/managers/data_parallel_controller.py,sha256=b64aC6iLr5RolJyNQnT-yTQ_TSI9DDLtuABf_TPTUrM,9421
 sglang/srt/managers/detokenizer_manager.py,sha256=XC2INyykOgwmIrFEGc-zf6LGZ5mMt6oPZt1YRXW_cbY,9650
-sglang/srt/managers/image_processor.py,sha256=s1QH9cSzT_nnitc6idzFjuGDp-pDnMTpbVZoQfzdSXU,20671
+sglang/srt/managers/image_processor.py,sha256=AWtCjl_zCbcn5LD4Hp4NXmsu225lQE0gWixIhQuUMpE,23872
 sglang/srt/managers/io_struct.py,sha256=9jhu794cc_BljFmVL6kQseTHGZNwEzONdlGEy_wjAcA,18357
-sglang/srt/managers/schedule_batch.py,sha256=smqDrzohvA8j76CLgI53CvpduheW1m__26S0O8HcCf0,49187
+sglang/srt/managers/schedule_batch.py,sha256=70smg65ed4xbotRegd8NzRu1YKA6M0oKn2Q69i0qWgA,49246
 sglang/srt/managers/schedule_policy.py,sha256=Qero_lwPEb7bM87qjWtYijGyRhtY0mMwjWP6SbjvaUE,18260
 sglang/srt/managers/scheduler.py,sha256=w0FPjiU5MoyP58UdJoPBr-hf-WmlWPpqb-5TSJDJBLo,71908
 sglang/srt/managers/session_controller.py,sha256=WXRbtninVEVM0rQYiXFzOwsDph0TNj1L2sRCWQF0dSg,5571
@@ -336,8 +350,8 @@ sglang/srt/mem_cache/radix_cache.py,sha256=hVILXvc5PauHuLTeyZbm3NCf3AOimaAuXjll5
 sglang/srt/metrics/collector.py,sha256=_yl0_paSARxS1ypZgd-pLJ29tMizolHuwROX21dOXTk,7326
 sglang/srt/metrics/func_timer.py,sha256=VFyNRrbnKVCwnQsrlLin1lITJfjQpf9m8sGPqL5LIsQ,3438
 sglang/srt/model_executor/cuda_graph_runner.py,sha256=hH646E_c4UlclGEawPDjg4KHgTUEk70WrPl6C7nnltM,18774
-sglang/srt/model_executor/forward_batch_info.py,sha256=t1RlBgoeS-_Ikl28Xjvt-aouh1nNUc3eLM4iGY4_QqY,14988
-sglang/srt/model_executor/model_runner.py,sha256=uohQ2n2R1HcVyaHwbdwM6xDvFxZSLgxacjMSrrogLpw,33537
+sglang/srt/model_executor/forward_batch_info.py,sha256=cTyRuJVBTBmkP4LAfScRSRrpjLCq7UfmUKoXuU5LZUw,15098
+sglang/srt/model_executor/model_runner.py,sha256=FNi5BTrpQ5VQ-VsMHujaAuSno3Y3DYYjymzPJLJoOIM,33609
 sglang/srt/model_loader/__init__.py,sha256=zGZkOBz1zx-pkaIy47BasL3fjDlAcxAXUTjInOhXHAE,919
 sglang/srt/model_loader/loader.py,sha256=2d9fJNxC3Y6YWmQX4nVOB-b9Glc43ztlkJYJFX1_kxk,46811
 sglang/srt/model_loader/utils.py,sha256=0NaMR67fESFopaklmsleiL27XH1QUrjZW246MUu1EJ0,1369
@@ -347,7 +361,8 @@ sglang/srt/models/chatglm.py,sha256=n8uZpx9iHw6V-XCns9mtTf99Iqh35ZjPC5bFDYtkoes,
 sglang/srt/models/commandr.py,sha256=y8DFUW0NKbkoY2DP6nhgJ1f7F_ysjaHEkEnZYZW2zdk,14523
 sglang/srt/models/dbrx.py,sha256=-L9QkUr_xuMuI6mn0AzG_VE1MqRXoaaFtD4r8UuAzkY,14789
 sglang/srt/models/deepseek.py,sha256=KfcQ54BqlS73XQmtcG0sfnmm3VXOGwUIkd34WS6Gp0Y,15694
-sglang/srt/models/deepseek_v2.py,sha256=9_284fDT15WEmv8qfnH2EzKX8fxSZrNyiz4iQtgb0tI,39065
+sglang/srt/models/deepseek_nextn.py,sha256=QmzByVDFw8F5cJfBU4-VVryXovn4HxvGBwbBTfJavJg,11740
+sglang/srt/models/deepseek_v2.py,sha256=vjY8BznqlEjX4P_mZQp1Syv24YxVee9Q258O8KWqE8I,39429
 sglang/srt/models/exaone.py,sha256=Wvr6XofnH2feJ-TzAm5aD1YTyfcum6JdnKMG1S7Xy4g,13035
 sglang/srt/models/gemma.py,sha256=4Jvt9F-BNhPFiBi5H8aPqcYqKeJLI9KZKy2WpR96RpM,12123
 sglang/srt/models/gemma2.py,sha256=cyQfby-kp2OZPsUACmBh3-jsXkYwQg9Tj6xqtZ7mTwM,15947
@@ -363,7 +378,7 @@ sglang/srt/models/llama_classification.py,sha256=DwboM1xHXdf3Fddf7xGnrfdOLJwXdiJ
 sglang/srt/models/llama_eagle.py,sha256=88DzR54DKBIKJ1h-bkIa8mc1qJnlkdZ1eGYY3c5mpBY,4442
 sglang/srt/models/llama_embedding.py,sha256=rh-AiczPY_pTpzcACHvSMVjh1hsV_MZBBwP0LQxPsGM,3130
 sglang/srt/models/llama_reward.py,sha256=oPxh5E2UkxLULNdR68dFvt2I7j33CJFN6nyA-8L2_cg,4516
-sglang/srt/models/llava.py,sha256=xrkg8sht8tBOID7427IEZtHL-KKWfEivDe2NqGjTSAs,26373
+sglang/srt/models/llava.py,sha256=Qbh26DcC6djw5G8olq0AC0WqzkkRVsiuT8I6RPCpH0o,26384
 sglang/srt/models/llavavid.py,sha256=dYUkKfHoE15vF_VXA_s_ICCTUMSmSgvP181fk8dUi0g,12185
 sglang/srt/models/minicpm.py,sha256=hVWri0-3sAiuGOMcIhGL2GphQZ13qBcLXuLTsQVALGY,13720
 sglang/srt/models/minicpm3.py,sha256=DZ7LltHsyDq8iE7nMi5C9gLzYcQrAIZYkRmx6lCuAgo,24683
@@ -378,16 +393,17 @@ sglang/srt/models/olmoe.py,sha256=luqgdyCYJTFyhaRfZElWSFV17ee6FjfU0CpemMmsTS8,15
 sglang/srt/models/phi3_small.py,sha256=jVKH2twKfELtqyjMWjH8CnyXlCKEkYtiUUnx18k9OLQ,14799
 sglang/srt/models/qwen.py,sha256=dg_sVrh7I58Q_LevvO2d5dFZi1T19V2czNh8-9nPUaE,9901
 sglang/srt/models/qwen2.py,sha256=igq-a61CQgH26xnim6c3yeWUCHiN_Nboxg4iu7oy7bo,15072
+sglang/srt/models/qwen2_5_vl.py,sha256=uSZEoCdyOlaANjnP21LxE7K_DqfG10JQ5sUkK6Ase2A,28045
 sglang/srt/models/qwen2_eagle.py,sha256=KTtejEezdLfd_odg3Na1i5kBk7W-YFg9hImfWyrMgVc,4288
 sglang/srt/models/qwen2_moe.py,sha256=GWi5nuaQWifPmyC3ld2G1wZJS5Xva6-1yjCUrNcGhkY,16539
-sglang/srt/models/qwen2_vl.py,sha256=d8jCZyoJ6sMZ8-Pw9bHb3VbwiZlbFzJvffA4yYBRj0U,23466
+sglang/srt/models/qwen2_vl.py,sha256=1LM4iyE4rHFRgP58hSFpKgZdaew_OSdwGRwwy3NiOzo,23523
 sglang/srt/models/registry.py,sha256=inKh9iwOp3LFYm3nqujg-OtABClOP-ifc1stA9cZegA,3434
 sglang/srt/models/stablelm.py,sha256=dO6EwFFiBWn-8yxV9tb3OtjNe9D0dF57Z298g7SmrhU,11308
 sglang/srt/models/torch_native_llama.py,sha256=X0AvlREIysazwFezqndRza7ZCWQ-R1hePoLW0brH4As,19131
 sglang/srt/models/xverse.py,sha256=sYSSbwB_VC6uGzxkzNHluaJzvSfQXCxQG_OsrIWLWvU,13549
 sglang/srt/models/xverse_moe.py,sha256=vN486GkRHvgyRgSW2e_zTOQHDkWx86lthahtKxl6M10,15511
 sglang/srt/models/yivl.py,sha256=88OubtuZ38Dxb2LzfV_MTPBI4wKhh4NJqFu--efbhFM,4809
-sglang/srt/openai_api/adapter.py,sha256=yQaAiZ43w3OIsNj8nC6BEjt9DPE5IZ-ovLAd0r7LbcY,61716
+sglang/srt/openai_api/adapter.py,sha256=tPsZ6cHlEofwJU7Cmfi3KtwSqvd3sv6EyeV6BfkdAcU,62349
 sglang/srt/openai_api/protocol.py,sha256=UInFUKQqS8KWLrCzA6s5_uaNC6xAUAAJ4WepQzQ7xpo,11845
 sglang/srt/sampling/custom_logit_processor.py,sha256=tDvoLgLqn-sy1qcY6vSrpbnHCeqbdk0uhMOO-uy4p4E,1099
 sglang/srt/sampling/sampling_batch_info.py,sha256=Ry1N79T9QQY_HJ8GjM50_W4tzKFxMtTfV4GccT7NQ0w,15129
@@ -401,8 +417,8 @@ sglang/srt/sampling/penaltylib/penalizers/repetition_penalty.py,sha256=l1DyU8kC8
 sglang/srt/speculative/build_eagle_tree.py,sha256=zWthboIgzPzSOXcGxDpDv0rBOQP55HYGrBKGqm2gWF0,20732
 sglang/srt/speculative/eagle_draft_cuda_graph_runner.py,sha256=FY4hcwd0Blx7AXbeX6quaXPNgWA8WGIqVcQiEgHyERk,8002
 sglang/srt/speculative/eagle_utils.py,sha256=ypjVmVTVzCGclOVHRMJxdLUSPkf1-7bNXQS0oP6dn5U,25644
-sglang/srt/speculative/eagle_worker.py,sha256=33zC6txEsNp9hD48iy-_67ov83Pf4iASulLg8GHLy5U,12898
-sglang/srt/speculative/spec_info.py,sha256=D7A27UU1iOwIBEjXTgAxZ7jdftbTiVlMCvK8GmYr2zg,488
+sglang/srt/speculative/eagle_worker.py,sha256=w7sLcW-EeE_iWyMJQhBuSo5Zvq6iPe-3m73-OIP1b-E,13153
+sglang/srt/speculative/spec_info.py,sha256=RWG4ik4Dah_V74mgP0gza6UaYFtN-BRV6aJZsHHGGtE,827
 sglang/test/few_shot_gsm8k.py,sha256=7yDbEQe49gZeJhz2wFFX-gf_59ThDKsCS1xwfogNc7k,4034
 sglang/test/few_shot_gsm8k_engine.py,sha256=QQbrwOX6-cJDD3RZC_e7zPnt6aSo8JdF8X_lRHSjdDM,3886
 sglang/test/run_eval.py,sha256=9yO0hXZOcn4abEOs96T-XPguDEklK16Ltco0pGF3zCg,4020
@@ -419,8 +435,8 @@ sglang/test/test_layernorm.py,sha256=IacByD5d-stXjzBz8Ypamc7povlcedpKPbb_4JLgo3c
 sglang/test/test_programs.py,sha256=aUV9Ex_B714ph7ytv6W3J7sdGDKC6lGIhUy95Yg6AHQ,18878
 sglang/test/test_utils.py,sha256=BU6lAX3bu3TNQZqVC9UPnyq3I7iV5kigHQKJx7UNlOQ,26192
 sglang/test/srt/sampling/penaltylib/utils.py,sha256=CjxHgywh0hx_87iynzQt_ztHu6zBVuE-YrZ-XPmW6U4,12906
-sglang-0.4.3.dist-info/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
-sglang-0.4.3.dist-info/METADATA,sha256=cpA0ecZd4jfaThOrStEpOGbXDTorUxqYdU4catzo2t4,23815
-sglang-0.4.3.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-sglang-0.4.3.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
-sglang-0.4.3.dist-info/RECORD,,
+sglang-0.4.3.post2.dist-info/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
+sglang-0.4.3.post2.dist-info/METADATA,sha256=kChnxBYuvq-HiSlLlTj6l3bOcIj_mLFuOsSQpUMGkZE,23821
+sglang-0.4.3.post2.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+sglang-0.4.3.post2.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.4.3.post2.dist-info/RECORD,,