sglang 0.2.14.post2__py3-none-any.whl → 0.3.0__py3-none-any.whl

sglang/srt/sampling/sampling_batch_info.py

@@ -21,10 +21,59 @@ class SamplingBatchInfo:
     top_ps: torch.Tensor = None
     top_ks: torch.Tensor = None
     min_ps: torch.Tensor = None
-    penalizer_orchestrator: penaltylib.BatchedPenalizerOrchestrator = None
+
+    # Dispatch in CUDA graph
+    need_min_p_sampling: bool = False
+
+    # Bias Tensors
     logit_bias: torch.Tensor = None
     vocab_mask: torch.Tensor = None
 
+    # Penalizer
+    penalizer_orchestrator: penaltylib.BatchedPenalizerOrchestrator = None
+    linear_penalties: torch.Tensor = None
+    scaling_penalties: torch.Tensor = None
+
+    def can_run_in_cuda_graph(self):
+        # Vocab bias and min_ps are not supported in CUDA graph
+        return (
+            self.logit_bias is None
+            and self.vocab_mask is None
+            and self.linear_penalties is None
+            and self.scaling_penalties is None
+            and not self.need_min_p_sampling
+        )
+
+    @classmethod
+    def dummy_one(cls, max_bs: int, vocab_size: int):
+        ret = cls(vocab_size=vocab_size)
+        ret.temperatures = torch.ones((max_bs, 1), dtype=torch.float, device="cuda")
+        ret.top_ps = torch.ones((max_bs,), dtype=torch.float, device="cuda")
+        ret.top_ks = torch.ones((max_bs,), dtype=torch.int, device="cuda")
+        return ret
+
+    def __getitem__(self, key):
+        if isinstance(key, slice):
+            # NOTE:This method is only used in CUDA graph
+            assert self.can_run_in_cuda_graph()
+            return SamplingBatchInfo(
+                vocab_size=self.vocab_size,
+                temperatures=self.temperatures[key],
+                top_ps=self.top_ps[key],
+                top_ks=self.top_ks[key],
+            )
+        else:
+            raise NotImplementedError
+
+    def inplace_assign(self, bs: int, other: SamplingBatchInfo):
+        # NOTE:This method is only used in CUDA graph
+        assert self.can_run_in_cuda_graph()
+
+        self.vocab_size = other.vocab_size
+        self.temperatures[:bs] = other.temperatures
+        self.top_ps[:bs] = other.top_ps
+        self.top_ks[:bs] = other.top_ks
+
     @classmethod
     def from_schedule_batch(cls, batch: ScheduleBatch, vocab_size: int):
         device = "cuda"
@@ -45,6 +94,7 @@ class SamplingBatchInfo:
         ret.min_ps = torch.tensor(
             [r.sampling_params.min_p for r in reqs], dtype=torch.float, device=device
         )
+        ret.need_min_p_sampling = any(r.sampling_params.min_p > 0 for r in reqs)
 
         # Each penalizers will do nothing if they evaluate themselves as not required by looking at
         # the sampling_params of the requests (See {_is_required()} of each penalizers). So this
@@ -72,6 +122,25 @@ class SamplingBatchInfo:
 
         return ret
 
+    def prepare_penalties(self):
+        self.scaling_penalties = None
+        self.linear_penalties = None
+
+        for penalizer in self.penalizer_orchestrator.penalizers.values():
+            if isinstance(penalizer, penaltylib.BatchedRepetitionPenalizer):
+                if penalizer.is_prepared():
+                    self.scaling_penalties = penalizer.cumulated_repetition_penalties
+            else:
+                if penalizer.is_prepared():
+                    if self.linear_penalties is None:
+                        bs = self.penalizer_orchestrator.batch.batch_size()
+                        self.linear_penalties = torch.zeros(
+                            (bs, self.vocab_size),
+                            dtype=torch.float32,
+                            device="cuda",
+                        )
+                    self.linear_penalties = penalizer.apply(self.linear_penalties)
+
     def update_regex_vocab_mask(self, batch: ScheduleBatch):
         bs, reqs = batch.batch_size(), batch.reqs
         device = "cuda"
@@ -81,15 +150,15 @@ class SamplingBatchInfo:
         self.vocab_mask = None
 
         if has_regex:
+            self.vocab_mask = torch.zeros(
+                bs, self.vocab_size, dtype=torch.bool, device=device
+            )
             for i, req in enumerate(reqs):
                 if req.regex_fsm is not None:
-                    if self.vocab_mask is None:
-                        self.vocab_mask = torch.zeros(
-                            bs, self.vocab_size, dtype=torch.bool, device=device
-                        )
+                    self.vocab_mask[i].fill_(1)
                     self.vocab_mask[i][
                         req.regex_fsm.get_next_instruction(req.regex_fsm_state).tokens
-                    ] = 1
+                    ] = 0
 
     def filter(self, unfinished_indices: List[int], new_indices: torch.Tensor):
         self.penalizer_orchestrator.filter(unfinished_indices, new_indices)

sglang/srt/server.py

@@ -272,7 +272,7 @@ async def retrieve_file_content(file_id: str):
 
 def launch_server(
     server_args: ServerArgs,
-    model_overide_args: Optional[dict] = None,
+    model_override_args: Optional[dict] = None,
     pipe_finish_writer: Optional[mp.connection.Connection] = None,
 ):
     """Launch an HTTP server."""
@@ -317,7 +317,7 @@ def launch_server(
             tp_rank_range,
             server_args,
             ports[3],
-            model_overide_args,
+            model_override_args,
         )
 
         try:
@@ -328,7 +328,7 @@ def launch_server(
             return
 
     # Launch processes
-    tokenizer_manager = TokenizerManager(server_args, port_args, model_overide_args)
+    tokenizer_manager = TokenizerManager(server_args, port_args, model_override_args)
     if server_args.chat_template:
         load_chat_template_for_openai_api(tokenizer_manager, server_args.chat_template)
     pipe_controller_reader, pipe_controller_writer = mp.Pipe(duplex=False)
@@ -341,7 +341,7 @@ def launch_server(
 
     proc_controller = mp.Process(
         target=start_controller_process,
-        args=(server_args, port_args, pipe_controller_writer, model_overide_args),
+        args=(server_args, port_args, pipe_controller_writer, model_override_args),
     )
     proc_controller.start()
 
@@ -501,7 +501,7 @@ class Runtime:
     def __init__(
         self,
         log_level: str = "error",
-        model_overide_args: Optional[dict] = None,
+        model_override_args: Optional[dict] = None,
         *args,
         **kwargs,
     ):
@@ -525,7 +525,7 @@ class Runtime:
 
         proc = mp.Process(
             target=launch_server,
-            args=(self.server_args, model_overide_args, pipe_writer),
+            args=(self.server_args, model_override_args, pipe_writer),
         )
         proc.start()
         pipe_writer.close()

sglang/srt/utils.py

@@ -407,7 +407,6 @@ def monkey_patch_vllm_dummy_weight_loader():
         DummyModelLoader,
         LoRAConfig,
         ModelConfig,
-        MultiModalConfig,
         ParallelConfig,
         SchedulerConfig,
         _initialize_model,
@@ -422,7 +421,6 @@ def monkey_patch_vllm_dummy_weight_loader():
         model_config: ModelConfig,
         device_config: DeviceConfig,
         lora_config: Optional[LoRAConfig],
-        multimodal_config: Optional[MultiModalConfig],
         parallel_config: ParallelConfig,
         scheduler_config: SchedulerConfig,
         cache_config: CacheConfig,
@@ -433,7 +431,6 @@ def monkey_patch_vllm_dummy_weight_loader():
                     model_config,
                     self.load_config,
                     lora_config,
-                    multimodal_config,
                     cache_config,
                 )
 

sglang/test/runners.py

@@ -180,7 +180,7 @@ class SRTRunner:
             tp_size=tp_size,
             dtype=get_dtype_str(torch_dtype),
             port=port,
-            mem_fraction_static=0.7,
+            mem_fraction_static=0.69,
             trust_remote_code=False,
             is_embedding=not self.is_generation,
         )

sglang/test/test_programs.py

@@ -2,8 +2,12 @@
 
 import json
 import re
+import time
+
+import numpy as np
 
 import sglang as sgl
+from sglang.utils import fetch_and_cache_jsonl
 
 
 def test_few_shot_qa():
@@ -447,3 +451,67 @@ def test_chat_completion_speculative():
         )
 
     gen_character_spec().sync()
+
+
+def test_hellaswag_select():
+    """Benchmark the accuracy of sgl.select on the HellaSwag dataset."""
+
+    url = "https://raw.githubusercontent.com/rowanz/hellaswag/master/data/hellaswag_val.jsonl"
+    lines = fetch_and_cache_jsonl(url)
+
+    # Construct prompts
+    def get_one_example(lines, i, include_answer):
+        ret = lines[i]["activity_label"] + ": " + lines[i]["ctx"] + " "
+        if include_answer:
+            ret += lines[i]["endings"][lines[i]["label"]]
+        return ret
+
+    def get_few_shot_examples(lines, k):
+        ret = ""
+        for i in range(k):
+            ret += get_one_example(lines, i, True) + "\n\n"
+        return ret
+
+    num_questions = 200
+    num_shots = 20
+    few_shot_examples = get_few_shot_examples(lines, num_shots)
+
+    questions = []
+    choices = []
+    labels = []
+    for i in range(len(lines[:num_questions])):
+        questions.append(get_one_example(lines, i, False))
+        choices.append(lines[i]["endings"])
+        labels.append(lines[i]["label"])
+    arguments = [{"question": q, "choices": c} for q, c in zip(questions, choices)]
+
+    #####################################
+    ######### SGL Program Begin #########
+    #####################################
+
+    import sglang as sgl
+
+    @sgl.function
+    def few_shot_hellaswag(s, question, choices):
+        s += few_shot_examples + question
+        s += sgl.select("answer", choices=choices)
+
+    #####################################
+    ########## SGL Program End ##########
+    #####################################
+
+    # Run requests
+    tic = time.time()
+    rets = few_shot_hellaswag.run_batch(
+        arguments,
+        temperature=0,
+        num_threads=64,
+        progress_bar=True,
+    )
+    preds = [choices[i].index(rets[i]["answer"]) for i in range(len(rets))]
+    latency = time.time() - tic
+
+    # Compute accuracy
+    accuracy = np.mean(np.array(preds) == np.array(labels))
+
+    return accuracy, latency

sglang/test/test_utils.py

@@ -23,6 +23,10 @@ from sglang.utils import get_exception_traceback
 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
+DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_TP1 = "meta-llama/Meta-Llama-3.1-8B-Instruct,mistralai/Mistral-7B-Instruct-v0.3,deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct,google/gemma-2-27b-it"
+DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_TP2 = "meta-llama/Meta-Llama-3.1-70B-Instruct,mistralai/Mixtral-8x7B-Instruct-v0.1,Qwen/Qwen2-57B-A14B-Instruct"
+DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_FP8_TP1 = "neuralmagic/Meta-Llama-3.1-8B-Instruct-FP8,neuralmagic/Mistral-7B-Instruct-v0.3-FP8,neuralmagic/DeepSeek-Coder-V2-Lite-Instruct-FP8,neuralmagic/gemma-2-2b-it-FP8"
+DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_FP8_TP2 = "neuralmagic/Meta-Llama-3.1-70B-Instruct-FP8,neuralmagic/Mixtral-8x7B-Instruct-v0.1-FP8,neuralmagic/Qwen2-72B-Instruct-FP8,neuralmagic/Qwen2-57B-A14B-Instruct-FP8"
 
 if os.getenv("SGLANG_IS_IN_CI", "false") == "true":
     DEFAULT_PORT_FOR_SRT_TEST_RUNNER = 5157

sglang/utils.py

@@ -4,6 +4,7 @@ import base64
 import importlib
 import json
 import logging
+import os
 import signal
 import sys
 import traceback
@@ -15,6 +16,7 @@ from typing import Union
 
 import numpy as np
 import requests
+from tqdm import tqdm
 
 logger = logging.getLogger(__name__)
 
@@ -260,3 +262,40 @@ class LazyImport:
     def __call__(self, *args, **kwargs):
         module = self._load()
         return module(*args, **kwargs)
+
+
+def fetch_and_cache_jsonl(url, cache_file="cached_data.jsonl"):
+    """Read and cache a jsonl file from a url."""
+
+    # Check if the cache file already exists
+    if os.path.exists(cache_file):
+        print("Loading data from cache...")
+        with open(cache_file, "r") as f:
+            data = [json.loads(line) for line in f]
+    else:
+        print("Downloading data from URL...")
+        # Stream the response to show the progress bar
+        response = requests.get(url, stream=True)
+        response.raise_for_status()  # Check for request errors
+
+        # Total size of the file in bytes
+        total_size = int(response.headers.get("content-length", 0))
+        chunk_size = 1024  # Download in chunks of 1KB
+
+        # Use tqdm to display the progress bar
+        with open(cache_file, "wb") as f, tqdm(
+            desc=cache_file,
+            total=total_size,
+            unit="B",
+            unit_scale=True,
+            unit_divisor=1024,
+        ) as bar:
+            for chunk in response.iter_content(chunk_size=chunk_size):
+                f.write(chunk)
+                bar.update(len(chunk))
+
+        # Convert the data to a list of dictionaries
+        with open(cache_file, "r") as f:
+            data = [json.loads(line) for line in f]
+
+    return data

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.2.14.post2"
+__version__ = "0.3.0"

{sglang-0.2.14.post2.dist-info → sglang-0.3.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sglang
-Version: 0.2.14.post2
+Version: 0.3.0
 Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License: Apache License
                                    Version 2.0, January 2004
@@ -312,7 +312,7 @@ pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.4/
 ### Method 2: From source
 ```
 # Use the last release branch
-git clone -b v0.2.14.post2 https://github.com/sgl-project/sglang.git
+git clone -b v0.3.0 https://github.com/sgl-project/sglang.git
 cd sglang
 
 pip install --upgrade pip
@@ -461,7 +461,7 @@ It supports streaming, vision, and most features of the Chat/Completions/Models/
 ```
 python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000 --tp 2
 ```
-- Add `--dp 2` to enable multi-GPU data parallelism. It can also be used together with tensor parallelism. Data parallelism is better for throughput if there is enough memory.
+- Add `--dp 2` to enable multi-GPU data parallelism. Data parallelism is better for throughput if there is enough memory. It can also be used together with tensor parallelism. The following command uses 4 GPUs in total.
 ```
 python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000 --dp 2 --tp 2
 ```
@@ -489,13 +489,13 @@ python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct
 ### Supported Models
 
 **Generative Models**
-
 - Llama / Llama 2 / Llama 3 / Llama 3.1
 - Mistral / Mixtral / Mistral NeMo
 - Gemma / Gemma 2
 - Qwen / Qwen 2 / Qwen 2 MoE
 - DeepSeek / DeepSeek 2
 - [LLaVA-OneVision](https://llava-vl.github.io/blog/2024-08-05-llava-onevision/)
+  - `python3 -m sglang.launch_server --model-path lmms-lab/llava-onevision-qwen2-7b-ov --port=30000 --chat-template=chatml-llava`
   - `python3 -m sglang.launch_server --model-path lmms-lab/llava-onevision-qwen2-72b-ov --port=30000 --tp-size=8 --chat-template=chatml-llava`
   - Query the server with the [OpenAI Vision API](https://platform.openai.com/docs/guides/vision). See examples at [test/srt/test_vision_openai_server.py](test/srt/test_vision_openai_server.py)
 - LLaVA 1.5 / 1.6 / NeXT
@@ -509,6 +509,7 @@ python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct
 - Grok
 - ChatGLM
 - InternLM 2
+- Exaone 3
 
 **Embedding Models**
 
@@ -636,7 +637,7 @@ print(state["answer_1"])
 #### More Examples
 
 Anthropic and VertexAI (Gemini) models are also supported.
-You can find more examples at [examples/quick_start](examples/quick_start).
+You can find more examples at [examples/quick_start](examples/frontend_language/quick_start).
 
 ### Language Feature
 To begin with, import sglang.
@@ -649,7 +650,7 @@ You can implement your prompt flow in a function decorated by `sgl.function`.
 You can then invoke the function with `run` or `run_batch`.
 The system will manage the state, chat template, parallelism and batching for you.
 
-The complete code for the examples below can be found at [readme_examples.py](examples/usage/readme_examples.py)
+The complete code for the examples below can be found at [readme_examples.py](examples/frontend_language/usage/readme_examples.py)
 
 #### Control Flow
 You can use any Python code within the function body, including control flow, nested function calls, and external libraries.
@@ -698,7 +699,7 @@ def image_qa(s, image_file, question):
     s += sgl.assistant(sgl.gen("answer", max_tokens=256)
 ```
 
-See also [srt_example_llava.py](examples/quick_start/srt_example_llava.py).
+See also [srt_example_llava.py](examples/frontend_language/quick_start/local_example_llava_next.py).
 
 #### Constrained Decoding
 Use `regex` to specify a regular expression as a decoding constraint.
@@ -742,7 +743,7 @@ def character_gen(s, name):
     s += sgl.gen("json_output", max_tokens=256, regex=character_regex)
 ```
 
-See also [json_decode.py](examples/usage/json_decode.py) for an additional example of specifying formats with Pydantic models.
+See also [json_decode.py](examples/frontend_language/usage/json_decode.py) for an additional example of specifying formats with Pydantic models.
 
 #### Batching
 Use `run_batch` to run a batch of requests with continuous batching.

sglang-0.3.0.dist-info/RECORD

@@ -0,0 +1,118 @@
+sglang/__init__.py,sha256=T8MYdFfKFPZcgFKHMBpOCIlFbhjwmr77Nqm6mdE6bCY,1590
+sglang/api.py,sha256=pH4CjwOXUweL5MF1sIkFMddDxfnF7PyUxEHC5kvNVbI,6468
+sglang/bench_latency.py,sha256=F7jMfKqMf1XFKJgkpR_yE33VJpsIhSr_SOJeRbngkb0,16758
+sglang/bench_serving.py,sha256=J_mMwnmDn0Jt07mzdGAuYOxpockHPLYJFL-kwoaqASY,36527
+sglang/check_env.py,sha256=rGRABCgt-0SfUrow4px28b2P59aMn8eVTnN5eZc_a8s,5397
+sglang/global_config.py,sha256=nwOjUflwqLQySPUMvk8Hk63TIS6mknh_ODSW3CZ1rJw,1704
+sglang/launch_server.py,sha256=FODfO0DW546dh-u1qDlWtrhsmj6hxkarXXv3cIdgkj8,549
+sglang/launch_server_llavavid.py,sha256=xnpSILJxsrbvqkERav5P26bErCQnhoTFmoKeScJltUA,1034
+sglang/utils.py,sha256=zxHwQhVxW_lWf-IH0wUw_pBTRLHLPypdRiU5M4XosMM,9669
+sglang/version.py,sha256=VrXpHDu3erkzwl_WXrqINBm9xWkcyUy53IQOj042dOs,22
+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=AC3tNNDwYfiu87jCldBWXYpFicCv6NMPJACMFEfCXu4,30331
+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
+sglang/lang/backend/anthropic.py,sha256=EXRX7xJgA5KZszX7toSLVnKzFQ5EO0Loj-YjHFtxSxg,2081
+sglang/lang/backend/base_backend.py,sha256=Q5HdiDtyBewQeoYH0kDtBRVL8KFiEPNq9dw7XmauHQ8,1985
+sglang/lang/backend/litellm.py,sha256=ugmL7sfUxkUHVbHtwNzHgdQAEd4UCjNQboFuE3KThcY,2450
+sglang/lang/backend/openai.py,sha256=qM7eVH_kMxnDd2rpxOH0v76KxtOJFlAwgLgWIKvFGCI,15060
+sglang/lang/backend/runtime_endpoint.py,sha256=hpezro0H6vG9KzLeKfYpPMwb4TaE0UanCIM0uG8Kdjw,9746
+sglang/lang/backend/vertexai.py,sha256=O-iBLD-y3vq80UxnrAoJri7bxpgd-_eakZ88Cf8bEGA,4855
+sglang/srt/conversation.py,sha256=2KDNe1suUPy6xqSkCx2xcO3pDPxTwqx5FaUxaqwCJ-M,19525
+sglang/srt/hf_transformers_utils.py,sha256=5UXJ-LdP92Sk_T843M9BHdnxRrcyiYfWH2IEg3dWgKI,6085
+sglang/srt/mm_utils.py,sha256=zox644S3IHUWmADdK4MnIbdTS2DWHOy0_Dq0gCU38QQ,12273
+sglang/srt/model_config.py,sha256=68QQ8iUWQHPv01RBeH23mvay6iJg9DWmCogC_vUgFLk,6371
+sglang/srt/server.py,sha256=yi8prs9_M0P0dOInrQLkHKiZ-oTigk_uzW8otEHImbU,19846
+sglang/srt/server_args.py,sha256=GiDyPWCvYA_98mSE9LuvUoEodo9gRnNPPIPn0nFkxUs,18259
+sglang/srt/utils.py,sha256=JJOlqRPbN_tSSNWj63syQpfz4v7hUwNvzWvOUpBh9SM,23746
+sglang/srt/configs/__init__.py,sha256=292SuEorST-lAq2Uvsv2M7yC28uYZlssVvRDsF-bZCQ,86
+sglang/srt/configs/exaone.py,sha256=Duxd4yQoKy8GWEzZD_kCY_OzmN_67CTJL_Kgn0eXk3g,10731
+sglang/srt/constrained/__init__.py,sha256=NLpZGj9RIx83ejDrM_pfaRtqGgaPq_ggJszPQENUJ2E,2037
+sglang/srt/constrained/base_tool_cache.py,sha256=5sazBMHHDpHMoqOjuY6itCxwTmIFCflIWEDXMtmrPVs,2006
+sglang/srt/constrained/fsm_cache.py,sha256=wigJs9PeTt-vYPJQEeUZwEKl6MFIfb5xy8uIg18bDbM,3132
+sglang/srt/constrained/jump_forward.py,sha256=LWRsmGPQcH6KT87wXwCRqtblU3pcAVCEzO0nWPxevs0,6636
+sglang/srt/layers/activation.py,sha256=JEXNTgqxoiU4N-gVm4XMjobhft4JKDcMrgTkfpsRUzM,4856
+sglang/srt/layers/decode_attention.py,sha256=TPD_608ZX9fQ_HDImifkxG_qcEYmimbEYY8lCBIjFuM,16628
+sglang/srt/layers/extend_attention.py,sha256=XIXm3p2cvKrDg10Po4qYGaEkXJOJBtCIhTB_lTyjAFE,14390
+sglang/srt/layers/layernorm.py,sha256=RXuS4UyksatqTF6lSK7VYyEiUEnBiNIBlEn8q4w84UA,3404
+sglang/srt/layers/logits_processor.py,sha256=Zx4eFAkFlThPrmz_-HuCN9SqGLanARm0wdZSVDyASAc,13085
+sglang/srt/layers/pooler.py,sha256=qNMG3Ycvt2yf9mk1Lcs-2K7oPeCuVeDYoHAxkMu9b_Q,1610
+sglang/srt/layers/prefill_attention.py,sha256=y7vdcuX8lMa9Qf_jQYNDvQO9PVCBQSs3hb5LV2DFgpU,5256
+sglang/srt/layers/radix_attention.py,sha256=o5a8r3XQ-oRwaxBlAgzJGv7p3dMbu0LrYsDc4uvpPgA,8338
+sglang/srt/layers/sampler.py,sha256=zPVa3PHc-tjDM_oP-1XFeHSRIErx844SLoe6MG8Qef0,6418
+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/managers/controller_multi.py,sha256=z3rguY1YYlSvVqLjKuurgJW1h0dxwPgIdPCQdJsVzYs,6478
+sglang/srt/managers/controller_single.py,sha256=5brrZ8vZxjvrSJHWrm5H3qGEZShN4EROG5r1o3pSjps,5124
+sglang/srt/managers/detokenizer_manager.py,sha256=yQkL5gLomLiy1qc6e9HNz8hcj7JQFHm1AfIrzpXaWJE,6852
+sglang/srt/managers/io_struct.py,sha256=Bd91cydX9_960NNP2xngqK-lsIaDB3oMYd56QddN4_Q,10722
+sglang/srt/managers/policy_scheduler.py,sha256=7HNUxBKJE444s_bHcPpbnHCygsnH-NIXYNSC2q6mRmc,8584
+sglang/srt/managers/schedule_batch.py,sha256=i68O-e9I_gDlme96xSBDjA2xDF1p-XBKvJRiJ9CsgcY,26423
+sglang/srt/managers/tokenizer_manager.py,sha256=8aHR5h9nYZsfdZE80uBc9egDFOQgKvjxmp-30Ha4ELk,29463
+sglang/srt/managers/tp_worker.py,sha256=4UuaBLzV6NMsG4XEIcpa4xMcOKIFvTan51ynKz85HXg,36842
+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=4vIUqVQpnHNhwWrokMVmGM4Dp5JFPHyXIvpEQsi2pNU,12862
+sglang/srt/model_executor/forward_batch_info.py,sha256=fSLhatN8vCgxn0Mft9D-r0pNi3SN0EQSTJmgaOtrqJc,16471
+sglang/srt/model_executor/model_runner.py,sha256=93YCStmZfdZlY0r-GGIVi0Xw66VwF77dEtGVmQf1VfU,23893
+sglang/srt/models/chatglm.py,sha256=PPOaeqipbkcsTUhMPbLb1HItWgW7KntefUfjEoMSxUM,13585
+sglang/srt/models/commandr.py,sha256=k86ykwWOlxLGaBbGUoMSaXngUxCbMVRbY5AoMOWpbU8,14377
+sglang/srt/models/dbrx.py,sha256=goLJ9Yt-9vxkwhCUFBidvP41H_dYTFsvrMZ4xm4FqGA,14875
+sglang/srt/models/deepseek.py,sha256=aYP6HUgxQbhcQGQEF4vX0ronBF8AirqIFG98EQn0YzY,16220
+sglang/srt/models/deepseek_v2.py,sha256=Htw_HDju9huYU5gBu2dqq6bKVao-AsifxfkGl2xRx-8,28521
+sglang/srt/models/exaone.py,sha256=ZFr0G0WITxg3dDfV_-vWqZpK_wMmiZi4r0vOT0gO9V4,13301
+sglang/srt/models/gemma.py,sha256=Ya_u2lKPKAc9iHEsW_HAEfCDgYTbxUOCzBI0LDuoOYs,12489
+sglang/srt/models/gemma2.py,sha256=MCmzzRAAafEQuQj6aGtB-TF4jH0RWrXcOPxSz6LRsXs,15137
+sglang/srt/models/gpt_bigcode.py,sha256=HEhMRO1Y37JfZtP7mDp0MexWj5h6XT9rKvxorOMKoQA,10409
+sglang/srt/models/grok.py,sha256=ZcJ4E11rKh-xo4k_j-H1XRreJWWv8yii-bMYC1lO2R8,15143
+sglang/srt/models/internlm2.py,sha256=VtWATs2eLIqbadYXTPY_vycFIstVk4zg3kxycA9H0Qw,12416
+sglang/srt/models/llama.py,sha256=MfDnlVWoJUG9DxgGYPiwhoU-0ZeRbhp6UmBR2ZAJSNk,13402
+sglang/srt/models/llama_classification.py,sha256=oSeROs633Gnak8vrbnWnCWDxfgP_zmKGO1A_43ukEQ4,4029
+sglang/srt/models/llama_embedding.py,sha256=RI2mpYheP5WwhuTINU-6IrU61usuMyCK9h2zDEyLW4g,3458
+sglang/srt/models/llava.py,sha256=OXmlOVIjFnMRKGwLweYB1N-xlfpZlTlZpqhsbwUCY6Y,23471
+sglang/srt/models/llavavid.py,sha256=4R2t8BZJKN85IrTLsLFb4yZuKVI2Cwp7kY8AJ-nEVoE,12012
+sglang/srt/models/minicpm.py,sha256=7RZEJ2TCqBL1JmMFVJ3J9DmZHRw0q90st49Wkh-sdL4,14039
+sglang/srt/models/mistral.py,sha256=tiYoKjyYVzlQl52QUZ33odD2yCxj9dxcqln474VuZOw,744
+sglang/srt/models/mixtral.py,sha256=KIsvruhXNq3Fwrs4_YE7J6fx54ObfnMuRNxgScE3Bmo,13830
+sglang/srt/models/mixtral_quant.py,sha256=O_97UKDYZokFhIBnamWfw0HLhln9_BUk_KfQ-sQnd8s,14286
+sglang/srt/models/qwen.py,sha256=geK88AyEyPbbDvMHJNY8XMSNpsCeu8g9kxnKyiJBpK4,10168
+sglang/srt/models/qwen2.py,sha256=WGYy3wcRY3f8Drd9I8GblXfv0bbHluRKVhnnhEZf584,12654
+sglang/srt/models/qwen2_moe.py,sha256=b0gd42GBWyvDmUu8BZbD9ZJO_ExbXBLQZRvu61UuXOA,17086
+sglang/srt/models/stablelm.py,sha256=9feHoiDEXSIe0WCrt4AfWXqxliJwRvr8w4XSnk6ipSI,11573
+sglang/srt/models/yivl.py,sha256=B6MELthWIm5KdSzX3o2tbbpApY8XdjUdmcQSD4dQe_I,4835
+sglang/srt/openai_api/adapter.py,sha256=3EeqASZXogpUkOP4xj7Rg_LfOLiIMUrZ9uFdeAy_pcc,50144
+sglang/srt/openai_api/protocol.py,sha256=onhnCjXpXCysvx_dLgOEmXz5XHHYB1t772cvHcK1GlY,9538
+sglang/srt/sampling/sampling_batch_info.py,sha256=CIoD0SzHSWCe7Wc4jkJj5vIPHGnOdfbgkC6fG5KQxOw,7551
+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
+sglang/srt/sampling/penaltylib/penalizers/frequency_penalty.py,sha256=IvYioX53Vq_ji-0Zhcz_r5mUa3T3GaIydVS6K4FhWfE,2557
+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/run_eval.py,sha256=NWxeLWmInBgkCvC9Jr_QzF7GfAiBve3Gf1JQrEOlNlU,3899
+sglang/test/runners.py,sha256=7N2g4vyqN98o6F0Lem5LUNAlW9ShEVxZxZuzSjmc0i4,7688
+sglang/test/simple_eval_common.py,sha256=r0G-9QLycs2ax3RMc44T_61fzMxlpTzv6pececC7lyY,12379
+sglang/test/simple_eval_gpqa.py,sha256=8Xt9Bw05c7SZTYrCZgB68OZUqUbLo69ywiyx0bTvSUk,3220
+sglang/test/simple_eval_humaneval.py,sha256=7lTi841NT58smNOtRwCedrdX9IWWypdLkOtaQOBy-GI,5687
+sglang/test/simple_eval_math.py,sha256=6kGKNwNbLN-Af3Wj8WTimWhH-Xp3enDmSvvSjsgWUpk,2550
+sglang/test/simple_eval_mgsm.py,sha256=wfbqJW9Rkc66vzq2fEMF6jchmoA8mw1OUiGU55cZ2B0,10261
+sglang/test/simple_eval_mmlu.py,sha256=FkwamjGMjueTixymkedF-YiPloSLiy4ftILFUrKZ9XI,4357
+sglang/test/test_activation.py,sha256=jkdNRzJnbd5OgZliQaIXpxovlcky17UrweomcOcMxoE,1442
+sglang/test/test_layernorm.py,sha256=IacByD5d-stXjzBz8Ypamc7povlcedpKPbb_4JLgo3c,3720
+sglang/test/test_programs.py,sha256=l21J8N91QTMO9TOvXPWNvPZVT0DgxYxOPHh1pOoFV_k,16927
+sglang/test/test_utils.py,sha256=3tt-BBv-lx7BT3whbVTMyRz6sh5jIbdBEbLZ08m2Ms8,15132
+sglang/test/srt/sampling/penaltylib/utils.py,sha256=-0p0rV-P4lNo7xAe3rQSBHTubc50a-DFyOQmLGAkgkQ,12515
+sglang-0.3.0.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+sglang-0.3.0.dist-info/METADATA,sha256=muukBuN4kq_4mCG_r_RFY94pQliDcVh-WuXNMApXoak,37383
+sglang-0.3.0.dist-info/WHEEL,sha256=uCRv0ZEik_232NlR4YDw4Pv3Ajt5bKvMH13NUU7hFuI,91
+sglang-0.3.0.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.3.0.dist-info/RECORD,,

{sglang-0.2.14.post2.dist-info → sglang-0.3.0.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (74.0.0)
+Generator: setuptools (74.1.1)
 Root-Is-Purelib: true
 Tag: py3-none-any