sglang 0.1.22__tar.gz → 0.1.24__tar.gz

{sglang-0.1.22/sglang.egg-info → sglang-0.1.24}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sglang
-Version: 0.1.22
+Version: 0.1.24
 Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License:                                  Apache License
                                    Version 2.0, January 2004
@@ -228,7 +228,7 @@ Requires-Dist: torch; extra == "srt"
 Requires-Dist: uvicorn; extra == "srt"
 Requires-Dist: uvloop; extra == "srt"
 Requires-Dist: zmq; extra == "srt"
-Requires-Dist: vllm==0.5.1; extra == "srt"
+Requires-Dist: vllm==0.5.3.post1; extra == "srt"
 Requires-Dist: outlines>=0.0.44; extra == "srt"
 Provides-Extra: openai
 Requires-Dist: openai>=1.0; extra == "openai"
@@ -282,6 +282,7 @@ The core features include:
 
 ### Method 1: With pip
 ```
+pip install --upgrade pip setuptools wheel
 pip install "sglang[all]"
 
 # Install FlashInfer CUDA kernels
@@ -293,6 +294,7 @@ pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.3/
 git clone https://github.com/sgl-project/sglang.git
 cd sglang
 
+pip install --upgrade pip
 pip install -e "python[all]"
 
 # Install FlashInfer CUDA kernels
@@ -390,15 +392,16 @@ python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct
 python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000 --mem-fraction-static 0.7
 ```
 - See [hyperparameter_tuning.md](docs/hyperparameter_tuning.md) on tuning hyperparameters for better performance.
-- Add `--nnodes 2` to run tensor parallelism on multiple nodes. If you have two nodes with two GPUs on each node and want to run TP=4, let `sgl-dev-1` be the hostname of the first node and `50000` be an available port.
+- Add `--nnodes 2` to run tensor parallelism on multiple nodes. 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.
 ```
 # Node 0
-python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --tp 4 --nccl-init sgl-dev-1:50000 --nnodes 2 --node-rank 0
+python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --tp 4 --nccl-init sgl-dev-0:50000 --nnodes 2 --node-rank 0
 
 # Node 1
-python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --tp 4 --nccl-init sgl-dev-1:50000 --nnodes 2 --node-rank 1
+python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --tp 4 --nccl-init sgl-dev-0:50000 --nnodes 2 --node-rank 1
 ```
 - If the model does not have a template in the Hugging Face tokenizer, you can specify a [custom chat template](docs/custom_chat_template.md).
+- To enable fp8 quantization, you can add `--quantization fp8` on a fp16 checkpoint or directly load a fp8 checkpoint without specifying any arguments.
 
 ### Supported Models
 
@@ -420,6 +423,7 @@ python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct
 - Grok
 - ChatGLM
 - InternLM 2
+- Mistral NeMo
 
 Instructions for supporting a new model are [here](https://github.com/sgl-project/sglang/blob/main/docs/model_support.md).
 

{sglang-0.1.22 → sglang-0.1.24}/README.md

@@ -37,6 +37,7 @@ The core features include:
 
 ### Method 1: With pip
 ```
+pip install --upgrade pip setuptools wheel
 pip install "sglang[all]"
 
 # Install FlashInfer CUDA kernels
@@ -48,6 +49,7 @@ pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.3/
 git clone https://github.com/sgl-project/sglang.git
 cd sglang
 
+pip install --upgrade pip
 pip install -e "python[all]"
 
 # Install FlashInfer CUDA kernels
@@ -145,15 +147,16 @@ python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct
 python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000 --mem-fraction-static 0.7
 ```
 - See [hyperparameter_tuning.md](docs/hyperparameter_tuning.md) on tuning hyperparameters for better performance.
-- Add `--nnodes 2` to run tensor parallelism on multiple nodes. If you have two nodes with two GPUs on each node and want to run TP=4, let `sgl-dev-1` be the hostname of the first node and `50000` be an available port.
+- Add `--nnodes 2` to run tensor parallelism on multiple nodes. 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.
 ```
 # Node 0
-python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --tp 4 --nccl-init sgl-dev-1:50000 --nnodes 2 --node-rank 0
+python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --tp 4 --nccl-init sgl-dev-0:50000 --nnodes 2 --node-rank 0
 
 # Node 1
-python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --tp 4 --nccl-init sgl-dev-1:50000 --nnodes 2 --node-rank 1
+python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --tp 4 --nccl-init sgl-dev-0:50000 --nnodes 2 --node-rank 1
 ```
 - If the model does not have a template in the Hugging Face tokenizer, you can specify a [custom chat template](docs/custom_chat_template.md).
+- To enable fp8 quantization, you can add `--quantization fp8` on a fp16 checkpoint or directly load a fp8 checkpoint without specifying any arguments.
 
 ### Supported Models
 
@@ -175,6 +178,7 @@ python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct
 - Grok
 - ChatGLM
 - InternLM 2
+- Mistral NeMo
 
 Instructions for supporting a new model are [here](https://github.com/sgl-project/sglang/blob/main/docs/model_support.md).
 

{sglang-0.1.22 → sglang-0.1.24}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "sglang"
-version = "0.1.22"
+version = "0.1.24"
 description = "SGLang is yet another fast serving framework for large language models and vision language models."
 readme = "README.md"
 requires-python = ">=3.8"
@@ -21,7 +21,7 @@ dependencies = [
 
 [project.optional-dependencies]
 srt = ["aiohttp", "fastapi", "hf_transfer", "huggingface_hub", "interegular", "packaging", "pillow",
-       "psutil", "pydantic", "torch", "uvicorn", "uvloop", "zmq", "vllm==0.5.1", "outlines>=0.0.44"]
+       "psutil", "pydantic", "torch", "uvicorn", "uvloop", "zmq", "vllm==0.5.3.post1", "outlines>=0.0.44"]
 openai = ["openai>=1.0", "tiktoken"]
 anthropic = ["anthropic>=0.20.0"]
 litellm = ["litellm>=1.0.0"]

{sglang-0.1.22 → sglang-0.1.24}/sglang/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "0.1.22"
+__version__ = "0.1.24"
 
 # SGL API Components
 from sglang.api import (

{sglang-0.1.22 → sglang-0.1.24}/sglang/bench_serving.py

@@ -5,6 +5,9 @@ Benchmark online serving.
 
 Usage:
 python3 -m sglang.bench_serving --backend sglang --num-prompt 10
+
+python3 -m sglang.bench_serving --backend sglang --dataset-name random --num-prompts 3000 --random-input 1024 --random-output 1024 --random-range-ratio 0.5
+python3 -m sglang.bench_serving --backend sglang --dataset-name random --request-rate-range 1,2,4,8,16,32 --random-input 4096 --random-output 1024 --random-range-ratio 0.125 --multi
 """
 
 import argparse
@@ -19,6 +22,7 @@ import traceback
 import warnings
 from argparse import ArgumentParser as FlexibleArgumentParser
 from dataclasses import dataclass, field
+from datetime import datetime
 from typing import AsyncGenerator, List, Optional, Tuple, Union
 
 import aiohttp
@@ -53,12 +57,80 @@ class RequestFuncOutput:
     itl: List[float] = field(default_factory=list)  # List of inter-token latencies
     prompt_len: int = 0
     error: str = ""
+    output_len: int = 0
 
 
 def remove_prefix(text: str, prefix: str) -> str:
     return text[len(prefix) :] if text.startswith(prefix) else text
 
 
+# trt llm not support ignore_eos
+# https://github.com/triton-inference-server/tensorrtllm_backend/issues/505
+async def async_request_trt_llm(
+    request_func_input: RequestFuncInput,
+    pbar: Optional[tqdm] = None,
+) -> RequestFuncOutput:
+    api_url = request_func_input.api_url
+    assert api_url.endswith("generate_stream")
+
+    async with aiohttp.ClientSession(timeout=AIOHTTP_TIMEOUT) as session:
+        payload = {
+            "accumulate_tokens": True,
+            "text_input": request_func_input.prompt,
+            "temperature": 0.000001,
+            "top_p": 1.0,
+            "max_tokens": request_func_input.output_len,
+            "stream": True,
+            "min_length": request_func_input.output_len,
+            "end_id": 1048576,
+        }
+        output = RequestFuncOutput()
+        output.prompt_len = request_func_input.prompt_len
+
+        ttft = 0.0
+        st = time.perf_counter()
+        most_recent_timestamp = st
+        try:
+            async with session.post(url=api_url, json=payload) as response:
+                if response.status == 200:
+                    async for chunk_bytes in response.content:
+                        chunk_bytes = chunk_bytes.strip()
+                        if not chunk_bytes:
+                            continue
+
+                        chunk = remove_prefix(chunk_bytes.decode("utf-8"), "data:")
+
+                        data = json.loads(chunk)
+                        output.generated_text += data["text_output"]
+                        timestamp = time.perf_counter()
+                        # First token
+                        if ttft == 0.0:
+                            ttft = time.perf_counter() - st
+                            output.ttft = ttft
+
+                        # Decoding phase
+                        else:
+                            output.itl.append(timestamp - most_recent_timestamp)
+
+                        most_recent_timestamp = timestamp
+
+                    output.latency = most_recent_timestamp - st
+                    output.success = True
+                    output.output_len = request_func_input.output_len
+
+                else:
+                    output.error = response.reason or ""
+                    output.success = False
+        except Exception:
+            output.success = False
+            exc_info = sys.exc_info()
+            output.error = "".join(traceback.format_exception(*exc_info))
+
+        if pbar:
+            pbar.update(1)
+        return output
+
+
 # set ignore_eos True by default
 async def async_request_openai_completions(
     request_func_input: RequestFuncInput,
@@ -76,7 +148,7 @@ async def async_request_openai_completions(
             "temperature": 0.0,
             "best_of": 1,
             "max_tokens": request_func_input.output_len,
-            "stream": True,
+            "stream": not args.disable_stream,
             "ignore_eos": True,
         }
         headers = {"Authorization": f"Bearer {os.environ.get('OPENAI_API_KEY')}"}
@@ -99,8 +171,9 @@ async def async_request_openai_completions(
                             continue
 
                         chunk = remove_prefix(chunk_bytes.decode("utf-8"), "data: ")
+                        latency = time.perf_counter() - st
                         if chunk == "[DONE]":
-                            latency = time.perf_counter() - st
+                            pass
                         else:
                             data = json.loads(chunk)
 
@@ -123,6 +196,7 @@ async def async_request_openai_completions(
                     output.generated_text = generated_text
                     output.success = True
                     output.latency = latency
+                    output.output_len = request_func_input.output_len
                 else:
                     output.error = response.reason or ""
                     output.success = False
@@ -167,6 +241,7 @@ ASYNC_REQUEST_FUNCS = {
     "sglang": async_request_openai_completions,
     "vllm": async_request_openai_completions,
     "lmdeploy": async_request_openai_completions,
+    "trt": async_request_trt_llm,
 }
 
 
@@ -175,9 +250,11 @@ class BenchmarkMetrics:
     completed: int
     total_input: int
     total_output: int
+    total_output_retokenized: int
     request_throughput: float
     input_throughput: float
     output_throughput: float
+    output_throughput_retokenized: float
     mean_ttft_ms: float
     median_ttft_ms: float
     std_ttft_ms: float
@@ -190,6 +267,8 @@ class BenchmarkMetrics:
     median_itl_ms: float
     std_itl_ms: float
     p99_itl_ms: float
+    mean_e2e_latency_ms: float
+    median_e2e_latency_ms: float
 
 
 default_sharegpt_path = "ShareGPT_V3_unfiltered_cleaned_split.json"
@@ -384,31 +463,36 @@ def calculate_metrics(
     outputs: List[RequestFuncOutput],
     dur_s: float,
     tokenizer: PreTrainedTokenizerBase,
+    backend: str,
 ) -> Tuple[BenchmarkMetrics, List[int]]:
-    actual_output_lens: List[int] = []
+    output_lens: List[int] = []
+    retokenized_output_lens: List[int] = []
     total_input = 0
     completed = 0
     itls: List[float] = []
     tpots: List[float] = []
     ttfts: List[float] = []
+    e2e_latencies: List[float] = []
     for i in range(len(outputs)):
         if outputs[i].success:
-            # We use the tokenizer to count the number of output tokens for all
-            # serving backends instead of looking at len(outputs[i].itl) since
-            # multiple output tokens may be bundled together
-            # Note : this may inflate the output token count slightly
-            output_len = len(
+            output_len = outputs[i].output_len
+            output_lens.append(output_len)
+            retokenized_output_len = len(
                 tokenizer(outputs[i].generated_text, add_special_tokens=False).input_ids
             )
-            actual_output_lens.append(output_len)
+            retokenized_output_lens.append(retokenized_output_len)
             total_input += input_requests[i][1]
             if output_len > 1:
                 tpots.append((outputs[i].latency - outputs[i].ttft) / (output_len - 1))
             itls += outputs[i].itl
             ttfts.append(outputs[i].ttft)
+
+            e2e_latencies.append(outputs[i].latency)
+
             completed += 1
         else:
-            actual_output_lens.append(0)
+            output_lens.append(0)
+            retokenized_output_lens.append(0)
 
     if completed == 0:
         warnings.warn(
@@ -419,10 +503,12 @@ def calculate_metrics(
     metrics = BenchmarkMetrics(
         completed=completed,
         total_input=total_input,
-        total_output=sum(actual_output_lens),
+        total_output=sum(output_lens),
+        total_output_retokenized=sum(retokenized_output_lens),
         request_throughput=completed / dur_s,
         input_throughput=total_input / dur_s,
-        output_throughput=sum(actual_output_lens) / dur_s,
+        output_throughput=sum(output_lens) / dur_s,
+        output_throughput_retokenized=sum(retokenized_output_lens) / dur_s,
         mean_ttft_ms=np.mean(ttfts or 0)
         * 1000,  # ttfts is empty if streaming is not supported by backend
         median_ttft_ms=np.median(ttfts or 0) * 1000,
@@ -436,9 +522,11 @@ def calculate_metrics(
         median_itl_ms=np.median(itls or 0) * 1000,
         std_itl_ms=np.std(itls or 0) * 1000,
         p99_itl_ms=np.percentile(itls or 0, 99) * 1000,
+        mean_e2e_latency_ms=np.mean(e2e_latencies) * 1000,
+        median_e2e_latency_ms=np.median(e2e_latencies) * 1000,
     )
 
-    return metrics, actual_output_lens
+    return metrics, output_lens
 
 
 async def benchmark(
@@ -449,6 +537,7 @@ async def benchmark(
     input_requests: List[Tuple[str, int, int]],
     request_rate: float,
     disable_tqdm: bool,
+    enable_multi: bool,
 ):
     if backend in ASYNC_REQUEST_FUNCS:
         request_func = ASYNC_REQUEST_FUNCS[backend]
@@ -498,19 +587,26 @@ async def benchmark(
 
     benchmark_duration = time.perf_counter() - benchmark_start_time
 
-    metrics, actual_output_lens = calculate_metrics(
+    metrics, output_lens = calculate_metrics(
         input_requests=input_requests,
         outputs=outputs,
         dur_s=benchmark_duration,
         tokenizer=tokenizer,
+        backend=backend,
     )
 
     print("\n{s:{c}^{n}}".format(s=" Serving Benchmark Result ", n=50, c="="))
+    print("{:<40} {:<10}".format("Backend:", backend))
     print("{:<40} {:<10}".format("Traffic request rate:", request_rate))
     print("{:<40} {:<10}".format("Successful requests:", metrics.completed))
     print("{:<40} {:<10.2f}".format("Benchmark duration (s):", benchmark_duration))
     print("{:<40} {:<10}".format("Total input tokens:", metrics.total_input))
     print("{:<40} {:<10}".format("Total generated tokens:", metrics.total_output))
+    print(
+        "{:<40} {:<10}".format(
+            "Total generated tokens (retokenized):", metrics.total_output_retokenized
+        )
+    )
     print(
         "{:<40} {:<10.2f}".format(
             "Request throughput (req/s):", metrics.request_throughput
@@ -526,6 +622,15 @@ async def benchmark(
             "Output token throughput (tok/s):", metrics.output_throughput
         )
     )
+    print("{s:{c}^{n}}".format(s="End-to-End Latency", n=50, c="-"))
+    print(
+        "{:<40} {:<10.2f}".format("Mean E2E Latency (ms):", metrics.mean_e2e_latency_ms)
+    )
+    print(
+        "{:<40} {:<10.2f}".format(
+            "Median E2E Latency (ms):", metrics.median_e2e_latency_ms
+        )
+    )
     print("{s:{c}^{n}}".format(s="Time to First Token", n=50, c="-"))
     print("{:<40} {:<10.2f}".format("Mean TTFT (ms):", metrics.mean_ttft_ms))
     print("{:<40} {:<10.2f}".format("Median TTFT (ms):", metrics.median_ttft_ms))
@@ -542,11 +647,53 @@ async def benchmark(
     print("{:<40} {:<10.2f}".format("P99 ITL (ms):", metrics.p99_itl_ms))
     print("=" * 50)
 
+    if (
+        metrics.median_ttft_ms is not None
+        and metrics.mean_itl_ms is not None
+        and metrics.output_throughput is not None
+    ):
+        result = {
+            "backend": args.backend,
+            "dataset_name": args.dataset_name,
+            "request_rate": request_rate,
+            "total_input": metrics.total_input,
+            "total_output": metrics.total_output,
+            "total_output_retokenized": metrics.total_output_retokenized,
+            "mean_e2e_latency": metrics.mean_e2e_latency_ms,
+            "median_e2e_latency": metrics.median_e2e_latency_ms,
+            "median_ttft": metrics.median_ttft_ms,
+            "median_itl": metrics.median_itl_ms,
+            "output_token_throughput": metrics.output_throughput,
+            "sharegpt_output_len": args.sharegpt_output_len,
+            "random_input_len": args.random_input_len,
+            "random_output_len": args.random_output_len,
+            "random_range_ratio": args.random_range_ratio,
+            "benchmark_duration": benchmark_duration,
+        }
+    else:
+        print(f"Error running benchmark for request rate: {request_rate}")
+        print("-" * 30)
+
+    # Determine output file name
+    if args.output_file:
+        output_file_name = args.output_file
+    else:
+        now = datetime.now().strftime("%m%d")
+        if args.dataset_name == "random":
+            output_file_name = f"{args.backend}_{now}_{args.num_prompts}_{args.random_input_len}_{args.random_output_len}.jsonl"
+        else:
+            output_file_name = f"{args.backend}_{now}_{args.num_prompts}_sharegpt.jsonl"
+
+    # Append results to a JSONL file
+    with open(output_file_name, "a") as file:
+        file.write(json.dumps(result) + "\n")
+
     result = {
         "duration": benchmark_duration,
         "completed": metrics.completed,
         "total_input_tokens": metrics.total_input,
         "total_output_tokens": metrics.total_output,
+        "total_output_tokens_retokenized": metrics.total_output_retokenized,
         "request_throughput": metrics.request_throughput,
         "input_throughput": metrics.input_throughput,
         "output_throughput": metrics.output_throughput,
@@ -563,15 +710,34 @@ async def benchmark(
         "std_itl_ms": metrics.std_itl_ms,
         "p99_itl_ms": metrics.p99_itl_ms,
         "input_lens": [output.prompt_len for output in outputs],
-        "output_lens": actual_output_lens,
+        "output_lens": output_lens,
         "ttfts": [output.ttft for output in outputs],
         "itls": [output.itl for output in outputs],
         "generated_texts": [output.generated_text for output in outputs],
         "errors": [output.error for output in outputs],
+        "mean_e2e_latency_ms": metrics.mean_e2e_latency_ms,
+        "median_e2e_latency_ms": metrics.median_e2e_latency_ms,
     }
     return result
 
 
+def parse_request_rate_range(request_rate_range):
+    if len(request_rate_range.split(",")) == 3:
+        start, stop, step = map(int, request_rate_range.split(","))
+        return list(range(start, stop, step))
+    else:
+        return list(map(int, request_rate_range.split(",")))
+
+
+def check_chat_template(model_path):
+    try:
+        tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
+        return "chat_template" in tokenizer.init_kwargs
+    except Exception as e:
+        print(f"Fail to load tokenizer config with error={e}")
+        return False
+
+
 def fire(args: argparse.Namespace):
     random.seed(args.seed)
     np.random.seed(args.seed)
@@ -581,6 +747,7 @@ def fire(args: argparse.Namespace):
             "sglang": 30000,
             "lmdeploy": 23333,
             "vllm": 8000,
+            "trt": 8000,
         }.get(args.backend, 30000)
 
     api_url = (
@@ -594,6 +761,16 @@ def fire(args: argparse.Namespace):
         else f"http://{args.host}:{args.port}/v1/models"
     )
 
+    if args.backend == "trt":
+        api_url = (
+            f"{args.base_url}/v2/models/ensemble/generate_stream"
+            if args.base_url
+            else f"http://{args.host}:{args.port}/v2/models/ensemble/generate_stream"
+        )
+        if args.model is None:
+            print("Please provide a model using `--model` when using `trt` backend.")
+            sys.exit(1)
+
     if args.model is None:
         try:
             response = requests.get(model_url)
@@ -610,6 +787,12 @@ def fire(args: argparse.Namespace):
         print("No model specified or found. Please provide a model using `--model`.")
         sys.exit(1)
 
+    if not check_chat_template(args.model):
+        print(
+            "\nWARNING It is recommended to use the `Chat` or `Instruct` model for benchmarking.\n"
+            "Because when the tokenizer counts the output tokens, if there is gibberish, it might count incorrectly.\n"
+        )
+
     print(f"{args}\n")
 
     backend = args.backend
@@ -637,17 +820,35 @@ def fire(args: argparse.Namespace):
     else:
         raise ValueError(f"Unknown dataset: {args.dataset_name}")
 
-    asyncio.run(
-        benchmark(
-            backend=backend,
-            api_url=api_url,
-            model_id=model_id,
-            tokenizer=tokenizer,
-            input_requests=input_requests,
-            request_rate=args.request_rate,
-            disable_tqdm=args.disable_tqdm,
+    if args.multi:
+        request_rates = parse_request_rate_range(args.request_rate_range)
+
+        for rate in request_rates:
+            asyncio.run(
+                benchmark(
+                    backend=backend,
+                    api_url=api_url,
+                    model_id=model_id,
+                    tokenizer=tokenizer,
+                    input_requests=input_requests,
+                    request_rate=rate,
+                    disable_tqdm=args.disable_tqdm,
+                    enable_multi=args.multi,
+                )
+            )
+    else:
+        asyncio.run(
+            benchmark(
+                backend=backend,
+                api_url=api_url,
+                model_id=model_id,
+                tokenizer=tokenizer,
+                input_requests=input_requests,
+                request_rate=args.request_rate,
+                disable_tqdm=args.disable_tqdm,
+                enable_multi=args.multi,
+            )
         )
-    )
 
 
 # to avoid relying on SGLang's components
@@ -751,6 +952,23 @@ if __name__ == "__main__":
         action="store_true",
         help="Specify to disable tqdm progress bar.",
     )
+    parser.add_argument(
+        "--multi",
+        action="store_true",
+        help="Use request rate range rather than single value.",
+    )
+    parser.add_argument(
+        "--request-rate-range",
+        type=str,
+        default="2,34,2",
+        help="Range of request rates in the format start,stop,step. Default is 2,34,2. It also supports a list of request rates, requiring the parameters to not equal three.",
+    )
+    parser.add_argument("--output-file", type=str, help="Output JSONL file name.")
+    parser.add_argument(
+        "--disable-stream",
+        action="store_true",
+        help="Disable streaming mode.",
+    )
 
     set_ulimit()
 

{sglang-0.1.22 → sglang-0.1.24}/sglang/global_config.py

@@ -16,9 +16,9 @@ class GlobalConfig:
         self.wait_for_new_request_delay = 0.0006
 
         # Runtime constants: New generation token ratio estimation
-        self.base_new_token_ratio = 0.4
+        self.init_new_token_ratio = 0.7
         self.base_min_new_token_ratio = 0.2
-        self.new_token_ratio_decay = 0.0001
+        self.new_token_ratio_decay = 0.001
         self.new_token_ratio_recovery = 0.05
 
         # Runtime constants: The threshold (number of tokens) to trigger layer-wise cuda sync.
@@ -27,6 +27,7 @@ class GlobalConfig:
 
         # Runtime constants: others
         self.num_continue_decode_steps = 10
+        self.retract_decode_steps = 20
         self.flashinfer_workspace_size = 192 * 1024 * 1024
 
         # Output tokenization configs

{sglang-0.1.22 → sglang-0.1.24}/sglang/lang/interpreter.py

@@ -288,6 +288,7 @@ class StreamExecutor:
             exes[i].text_ = str(self.text_)
             exes[i].messages_ = list(self.messages_)
             exes[i].cur_role = self.cur_role
+            exes[i].cur_role_begin_pos = self.cur_role_begin_pos
             exes[i].fork_start_text_pos = len(self.text_)
             exes[i].images_ = list(self.images_)
 

{sglang-0.1.22 → sglang-0.1.24}/sglang/srt/hf_transformers_utils.py

@@ -4,19 +4,26 @@ import functools
 import json
 import os
 import warnings
-from typing import AbstractSet, Collection, Literal, Optional, Union
+from typing import AbstractSet, Collection, Dict, Literal, Optional, Type, Union
 
 from huggingface_hub import snapshot_download
 from transformers import (
     AutoConfig,
     AutoProcessor,
     AutoTokenizer,
+    PretrainedConfig,
     PreTrainedTokenizer,
     PreTrainedTokenizerFast,
 )
+from vllm.transformers_utils.configs import ChatGLMConfig, DbrxConfig
 
 from sglang.srt.utils import is_multimodal_model
 
+_CONFIG_REGISTRY: Dict[str, Type[PretrainedConfig]] = {
+    ChatGLMConfig.model_type: ChatGLMConfig,
+    DbrxConfig.model_type: DbrxConfig,
+}
+
 
 def download_from_hf(model_path: str):
     if os.path.exists(model_path):
@@ -40,6 +47,9 @@ def get_config(
     config = AutoConfig.from_pretrained(
         model, trust_remote_code=trust_remote_code, revision=revision
     )
+    if config.model_type in _CONFIG_REGISTRY:
+        config_class = _CONFIG_REGISTRY[config.model_type]
+        config = config_class.from_pretrained(model, revision=revision)
     if model_overide_args:
         config.update(model_overide_args)
     return config
@@ -63,6 +73,8 @@ def get_context_length(config):
     rope_scaling = getattr(config, "rope_scaling", None)
     if rope_scaling:
         rope_scaling_factor = config.rope_scaling["factor"]
+        if config.rope_scaling["rope_type"] == "llama3":
+            rope_scaling_factor = 1
     else:
         rope_scaling_factor = 1