sglang 0.1.21__py3-none-any.whl → 0.1.22__py3-none-any.whl

sglang/srt/utils.py

@@ -3,9 +3,9 @@
 import base64
 import fcntl
 import logging
-import multiprocessing
 import os
 import random
+import resource
 import socket
 import struct
 import time
@@ -16,12 +16,11 @@ from typing import List, Optional
 import numpy as np
 import psutil
 import requests
-import rpyc
 import torch
+import torch.distributed as dist
 import triton
 from fastapi.responses import JSONResponse
 from packaging import version as pkg_version
-from rpyc.utils.server import ThreadedServer
 from starlette.middleware.base import BaseHTTPMiddleware
 
 logger = logging.getLogger(__name__)
@@ -148,7 +147,6 @@ def is_port_available(port):
 def allocate_init_ports(
     port: Optional[int] = None,
     additional_ports: Optional[List[int]] = None,
-    tp_size: int = 1,
     dp_size: int = 1,
 ):
     """Allocate ports for all connections."""
@@ -160,8 +158,8 @@ def allocate_init_ports(
     ret_ports = list(set(x for x in ret_ports if is_port_available(x)))
     cur_port = ret_ports[-1] + 1 if len(ret_ports) > 0 else 10000
 
-    # HTTP + Tokenizer + Controller + Detokenizer + dp_size * (nccl + tp_size)
-    num_ports_needed = 4 + dp_size * (1 + tp_size)
+    # HTTP + Tokenizer + Controller + Detokenizer + dp_size * 1 (nccl)
+    num_ports_needed = 4 + dp_size
     while len(ret_ports) < num_ports_needed:
         if cur_port not in ret_ports and is_port_available(cur_port):
             ret_ports.append(cur_port)
@@ -188,71 +186,6 @@ def get_int_token_logit_bias(tokenizer, vocab_size):
     return logit_bias
 
 
-def wrap_kernel_launcher(kernel):
-    """A faster launcher for triton kernels."""
-    if int(triton.__version__.split(".")[0]) >= 3:
-        return None
-
-    gpu_id = torch.cuda.current_device()
-    kernels = kernel.cache[gpu_id].values()
-    kernel = next(iter(kernels))
-
-    # Different trition versions use different low-level names
-    if hasattr(kernel, "cu_function"):
-        kfunction = kernel.cu_function
-    else:
-        kfunction = kernel.function
-
-    if hasattr(kernel, "c_wrapper"):
-        run = kernel.c_wrapper
-    else:
-        run = kernel.run
-
-    add_cluster_dim = True
-
-    def ret_func(grid, num_warps, *args):
-        nonlocal add_cluster_dim
-
-        try:
-            if add_cluster_dim:
-                run(
-                    grid[0],
-                    grid[1],
-                    grid[2],
-                    num_warps,
-                    1,
-                    1,
-                    1,
-                    1,
-                    kernel.shared,
-                    0,
-                    kfunction,
-                    None,
-                    None,
-                    kernel,
-                    *args,
-                )
-            else:
-                run(
-                    grid[0],
-                    grid[1],
-                    grid[2],
-                    num_warps,
-                    kernel.shared,
-                    0,
-                    kfunction,
-                    None,
-                    None,
-                    kernel,
-                    *args,
-                )
-        except TypeError:
-            add_cluster_dim = not add_cluster_dim
-            ret_func(grid, num_warps, *args)
-
-    return ret_func
-
-
 def is_multimodal_model(model):
     from sglang.srt.model_config import ModelConfig
 
@@ -371,49 +304,6 @@ def load_image(image_file):
     return image, image_size
 
 
-def connect_rpyc_service(host, port):
-    repeat_count = 0
-    while repeat_count < 20:
-        try:
-            con = rpyc.connect(
-                host,
-                port,
-                config={
-                    "allow_public_attrs": True,
-                    "allow_pickle": True,
-                    "sync_request_timeout": 3600,
-                },
-            )
-            break
-        except ConnectionRefusedError as e:
-            time.sleep(1)
-        repeat_count += 1
-    if repeat_count == 20:
-        raise RuntimeError(f"Connect rpyc error: {e}")
-
-    return con.root
-
-
-def start_rpyc_service(service: rpyc.Service, port: int):
-    t = ThreadedServer(
-        service=service,
-        port=port,
-        protocol_config={
-            "allow_public_attrs": True,
-            "allow_pickle": True,
-            "sync_request_timeout": 3600,
-        },
-    )
-    t.logger.setLevel(logging.WARN)
-    t.start()
-
-
-def start_rpyc_service_process(service: rpyc.Service, port: int):
-    proc = multiprocessing.Process(target=start_rpyc_service, args=(service, port))
-    proc.start()
-    return proc
-
-
 def suppress_other_loggers():
     from vllm.logger import logger as vllm_default_logger
 
@@ -445,7 +335,7 @@ def kill_parent_process():
     """Kill the parent process and all children of the parent process."""
     current_process = psutil.Process()
     parent_process = current_process.parent()
-    children = current_process.children(recursive=True)
+    children = parent_process.children(recursive=True)
     for child in children:
         if child.pid != current_process.pid:
             os.kill(child.pid, 9)
@@ -559,7 +449,6 @@ def get_ip_address(ifname):
 
 def send_addrs_to_rank_0(model_port_args, server_args):
     assert server_args.node_rank != 0 and server_args.dp_size == 1
-    import torch.distributed as dist
 
     ifname = os.environ.get(
         "SGLANG_SOCKET_IFNAME", os.environ.get("NCCL_SOCKET_IFNAME", "eth0")
@@ -591,7 +480,6 @@ def send_addrs_to_rank_0(model_port_args, server_args):
 
 def receive_addrs(model_port_args, server_args):
     assert server_args.node_rank == 0 and server_args.dp_size == 1
-    import torch.distributed as dist
 
     ifname = os.environ.get(
         "SGLANG_SOCKET_IFNAME", os.environ.get("NCCL_SOCKET_IFNAME", "eth0")
@@ -624,3 +512,14 @@ def receive_addrs(model_port_args, server_args):
 
     dist.barrier()
     dist.destroy_process_group()
+
+
+def set_ulimit(target_soft_limit=65535):
+    resource_type = resource.RLIMIT_NOFILE
+    current_soft, current_hard = resource.getrlimit(resource_type)
+
+    if current_soft < target_soft_limit:
+        try:
+            resource.setrlimit(resource_type, (target_soft_limit, current_hard))
+        except ValueError as e:
+            logger.warn(f"Fail to set RLIMIT_NOFILE: {e}")

sglang/test/test_conversation.py

@@ -1,5 +1,5 @@
 from sglang.srt.conversation import generate_chat_conv
-from sglang.srt.managers.openai_protocol import (
+from sglang.srt.managers.openai_api.protocol import (
     ChatCompletionMessageContentImagePart,
     ChatCompletionMessageContentImageURL,
     ChatCompletionMessageContentTextPart,

sglang/test/test_openai_protocol.py

@@ -1,4 +1,4 @@
-from sglang.srt.managers.openai_protocol import (
+from sglang.srt.managers.openai_api.protocol import (
     ChatCompletionMessageContentImagePart,
     ChatCompletionMessageContentImageURL,
     ChatCompletionMessageContentTextPart,

sglang/test/test_programs.py

@@ -306,7 +306,7 @@ def test_image_qa():
     assert (
         "taxi" in state.messages()[-1]["content"]
         or "car" in state.messages()[-1]["content"]
-    )
+    ), f"{state.messages()[-1]['content']}"
 
 
 def test_stream():

sglang/test/test_utils.py

@@ -6,9 +6,9 @@ from functools import partial
 import numpy as np
 import requests
 
-from sglang.backend.openai import OpenAI
-from sglang.backend.runtime_endpoint import RuntimeEndpoint
 from sglang.global_config import global_config
+from sglang.lang.backend.openai import OpenAI
+from sglang.lang.backend.runtime_endpoint import RuntimeEndpoint
 from sglang.utils import get_exception_traceback
 
 

{sglang-0.1.21.dist-info → sglang-0.1.22.dist-info}/METADATA

@@ -1,7 +1,7 @@
 Metadata-Version: 2.1
 Name: sglang
-Version: 0.1.21
-Summary: A structured generation langauge for LLMs.
+Version: 0.1.22
+Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License: Apache License
                                    Version 2.0, January 2004
                                 http://www.apache.org/licenses/
@@ -236,7 +236,6 @@ Requires-Dist: packaging ; extra == 'srt'
 Requires-Dist: pillow ; extra == 'srt'
 Requires-Dist: psutil ; extra == 'srt'
 Requires-Dist: pydantic ; extra == 'srt'
-Requires-Dist: rpyc ; extra == 'srt'
 Requires-Dist: torch ; extra == 'srt'
 Requires-Dist: uvicorn ; extra == 'srt'
 Requires-Dist: uvloop ; extra == 'srt'
@@ -252,23 +251,29 @@ Requires-Dist: outlines >=0.0.44 ; extra == 'srt'
 
 | [**Blog**](https://lmsys.org/blog/2024-01-17-sglang/) | [**Paper**](https://arxiv.org/abs/2312.07104) |
 
-SGLang is a structured generation language designed for large language models (LLMs).
-It makes your interaction with LLMs faster and more controllable by co-designing the frontend language and the runtime system.
+SGLang is a fast serving framework for large language models and vision language models.
+It makes your interaction with models faster and more controllable by co-designing the backend runtime and frontend language.
 
 The core features include:
+- **Fast Backend Runtime**: Efficient serving with RadixAttention for prefix caching, jump-forward constrained decoding, continuous batching, token attention (paged attention), tensor parallelism, flashinfer kernels, and quantization (AWQ/FP8/GPTQ/Marlin).
 - **Flexible Frontend Language**: Enables easy programming of LLM applications with chained generation calls, advanced prompting, control flow, multiple modalities, parallelism, and external interactions.
-- **High-Performance Backend Runtime**: Features RadixAttention for accelerating complex LLM programs by reusing the KV cache across multiple calls. It can also serve as a standalone inference engine with all common techniques implemented (e.g., continuous batching and tensor parallelism).
 
 ## News
+- [2024/04] 🔥 SGLang is used by the official **LLaVA-NeXT (video)** release ([blog](https://llava-vl.github.io/blog/2024-04-30-llava-next-video/)).
 - [2024/02] 🔥 SGLang enables **3x faster JSON decoding** with compressed finite state machine ([blog](https://lmsys.org/blog/2024-02-05-compressed-fsm/)).
-- [2024/01] 🔥 SGLang powers the serving of the official **LLaVA v1.6** release demo ([usage](https://github.com/haotian-liu/LLaVA?tab=readme-ov-file#demo)).
 - [2024/01] SGLang provides up to **5x faster inference** with RadixAttention ([blog](https://lmsys.org/blog/2024-01-17-sglang/)).
 
+<details>
+<summary>More</summary>
+
+- [2024/01] SGLang powers the serving of the official **LLaVA v1.6** release demo ([usage](https://github.com/haotian-liu/LLaVA?tab=readme-ov-file#demo)).
+
+</details>
+
 ## Contents
 - [Install](#install)
-- [Quick Start](#quick-start)
-- [Frontend: Structured Generation Language (SGLang)](#frontend-structured-generation-language-sglang)
 - [Backend: SGLang Runtime (SRT)](#backend-sglang-runtime-srt)
+- [Frontend: Structured Generation Language (SGLang)](#frontend-structured-generation-language-sglang)
 - [Benchmark And Performance](#benchmark-and-performance)
 - [Roadmap](#roadmap)
 - [Citation And Acknowledgment](#citation-and-acknowledgment)
@@ -297,6 +302,16 @@ pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.3/
 ### Method 3: Using docker
 The docker images are available on Docker Hub as [lmsysorg/sglang](https://hub.docker.com/r/lmsysorg/sglang/tags).
 
+```bash
+docker run --gpus all \
+    -p 30000:30000 \
+    -v ~/.cache/huggingface:/root/.cache/huggingface \
+    --env "HUGGING_FACE_HUB_TOKEN=<secret>" \
+    --ipc=host \
+    lmsysorg/sglang:latest \
+    python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B --host 0.0.0.0 --port 30000
+```
+
 ### Common Notes
 - If you see errors from the Triton compiler, please install the [Triton Nightly](https://triton-lang.org/main/getting-started/installation.html) by
 ```
@@ -306,13 +321,129 @@ pip install -U --index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/
 - If you cannot install FlashInfer, check out its [installation](https://docs.flashinfer.ai/installation.html#) page. If you still cannot install it, you can use the slower Triton kernels by adding `--disable-flashinfer` when launching the server.
 - If you only need to use the OpenAI backend, you can avoid installing other dependencies by using `pip install "sglang[openai]"`.
 
-## Quick Start
+## Backend: SGLang Runtime (SRT)
+The SGLang Runtime (SRT) is an efficient serving engine.
+
+### Quick Start
+Launch a server
+```
+python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000
+```
+
+Send a request
+```
+curl http://localhost:30000/generate \
+  -H "Content-Type: application/json" \
+  -d '{
+    "text": "Once upon a time,",
+    "sampling_params": {
+      "max_new_tokens": 16,
+      "temperature": 0
+    }
+  }'
+```
+Learn more about the argument format [here](docs/sampling_params.md).
+
+### OpenAI Compatible API
+In addition, the server supports OpenAI-compatible APIs.
+
+```python
+import openai
+client = openai.Client(
+    base_url="http://127.0.0.1:30000/v1", api_key="EMPTY")
+
+# Text completion
+response = client.completions.create(
+	model="default",
+	prompt="The capital of France is",
+	temperature=0,
+	max_tokens=32,
+)
+print(response)
+
+# Chat completion
+response = client.chat.completions.create(
+    model="default",
+    messages=[
+        {"role": "system", "content": "You are a helpful AI assistant"},
+        {"role": "user", "content": "List 3 countries and their capitals."},
+    ],
+    temperature=0,
+    max_tokens=64,
+)
+print(response)
+```
+
+It supports streaming, vision, and most features of the Chat/Completions/Models endpoints specified by the [OpenAI API Reference](https://platform.openai.com/docs/api-reference/).
+
+### Additional Server Arguments
+- Add `--tp 2` to enable tensor parallelism. If it indicates `peer access is not supported between these two devices`, add `--enable-p2p-check` option.
+```
+python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000 --tp 2
+```
+- Add `--dp 2` to enable data parallelism. It can also be used together with tp. Data parallelism is better for throughput if there is enough memory.
+```
+python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000 --dp 2 --tp 2
+```
+- If you see out-of-memory errors during serving, please try to reduce the memory usage of the KV cache pool by setting a smaller value of `--mem-fraction-static`. The default value is `0.9`
+```
+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.
+```
+# 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
+
+# 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
+```
+- 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).
+
+### Supported Models
+
+- Llama / Llama 2 / Llama 3
+- Mistral / Mixtral
+- Gemma / Gemma 2
+- Qwen / Qwen 2 / Qwen 2 MoE
+- LLaVA 1.5 / 1.6
+  - `python -m sglang.launch_server --model-path liuhaotian/llava-v1.5-7b --tokenizer-path llava-hf/llava-1.5-7b-hf --chat-template vicuna_v1.1 --port 30000`
+  - `python -m sglang.launch_server --model-path liuhaotian/llava-v1.6-vicuna-7b --tokenizer-path llava-hf/llava-1.5-7b-hf --chat-template vicuna_v1.1 --port 30000`
+  - `python -m sglang.launch_server --model-path liuhaotian/llava-v1.6-34b --tokenizer-path liuhaotian/llava-v1.6-34b-tokenizer --port 30000`
+- LLaVA-NeXT-Video
+  - see [examples/usage/llava_video](examples/usage/llava_video)
+- Yi-VL
+  - see [srt_example_yi_vl.py](examples/quick_start/srt_example_yi_vl.py).
+- StableLM
+- Command-R
+- DBRX
+- Grok
+- ChatGLM
+- InternLM 2
+
+Instructions for supporting a new model are [here](https://github.com/sgl-project/sglang/blob/main/docs/model_support.md).
+
+### Benchmark Performance
+
+- Benchmark a single static batch. Run the following command without launching a server. The arguments are the same as those for `launch_server.py`.
+  ```
+  python -m sglang.bench_latency --model-path meta-llama/Meta-Llama-3-8B-Instruct --batch 32 --input-len 256 --output-len 32
+  ```
+- Benchmark online serving. Launch a server first and run the following command.
+  ```
+  python3 -m sglang.bench_serving --backend sglang --num-prompt 10
+  ```
+
+## Frontend: Structured Generation Language (SGLang)
+The frontend language can be used with local models or API models.
+
+### Quick Start
 The example below shows how to use sglang to answer a mulit-turn question.
 
-### Using Local Models
+#### Using Local Models
 First, launch a server with
 ```
-python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000
+python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000
 ```
 
 Then, connect to the server and answer a multi-turn question.
@@ -341,7 +472,7 @@ for m in state.messages():
 print(state["answer_1"])
 ```
 
-### Using OpenAI Models
+#### Using OpenAI Models
 Set the OpenAI API Key
 ```
 export OPENAI_API_KEY=sk-******
@@ -372,13 +503,12 @@ for m in state.messages():
 print(state["answer_1"])
 ```
 
-### More Examples
+#### More Examples
 
 Anthropic and VertexAI (Gemini) models are also supported.
 You can find more examples at [examples/quick_start](examples/quick_start).
 
-## Frontend: Structured Generation Language (SGLang)
-
+### Language Feature
 To begin with, import sglang.
 ```python
 import sglang as sgl
@@ -391,7 +521,7 @@ The system will manage the state, chat template, parallelism and batching for yo
 
 The complete code for the examples below can be found at [readme_examples.py](examples/usage/readme_examples.py)
 
-### Control Flow
+#### Control Flow
 You can use any Python code within the function body, including control flow, nested function calls, and external libraries.
 
 ```python
@@ -406,7 +536,7 @@ def tool_use(s, question):
         s += "The key word to search is" + sgl.gen("word")
 ```
 
-### Parallelism
+#### Parallelism
 Use `fork` to launch parallel prompts.
 Because `sgl.gen` is non-blocking, the for loop below issues two generation calls in parallel.
 
@@ -428,7 +558,7 @@ def tip_suggestion(s):
     s += "In summary" + sgl.gen("summary")
 ```
 
-### Multi Modality
+#### Multi Modality
 Use `sgl.image` to pass an image as input.
 
 ```python
@@ -440,7 +570,7 @@ def image_qa(s, image_file, question):
 
 See also [srt_example_llava.py](examples/quick_start/srt_example_llava.py).
 
-### Constrained Decoding
+#### Constrained Decoding
 Use `regex` to specify a regular expression as a decoding constraint.
 This is only supported for local models.
 
@@ -455,7 +585,7 @@ def regular_expression_gen(s):
     )
 ```
 
-### JSON Decoding
+#### JSON Decoding
 Use `regex` to specify a JSON schema with a regular expression.
 
 ```python
@@ -484,8 +614,7 @@ def character_gen(s, name):
 
 See also [json_decode.py](examples/usage/json_decode.py) for an additional example on specifying formats with Pydantic models.
 
-
-### Batching
+#### Batching
 Use `run_batch` to run a batch of requests with continuous batching.
 
 ```python
@@ -504,7 +633,7 @@ states = text_qa.run_batch(
 )
 ```
 
-### Streaming
+#### Streaming
 Add `stream=True` to enable streaming.
 
 ```python
@@ -523,139 +652,10 @@ for out in state.text_iter():
     print(out, end="", flush=True)
 ```
 
-### Tips and Implementation Details
+#### Tips and Implementation Details
 - The `choices` argument in `sgl.gen` is implemented by computing the [token-length normalized log probabilities](https://blog.eleuther.ai/multiple-choice-normalization/) of all choices and selecting the one with the highest probability.
 - The `regex` argument in `sgl.gen` is implemented through autoregressive decoding with logit bias masking, according to the constraints set by the regex. It is compatible with `temperature=0` and `temperature != 0`.
 
-## Backend: SGLang Runtime (SRT)
-The SGLang Runtime (SRT) is designed to work best with the SGLang frontend.
-However, it can also be used as a standalone API server.
-In this case, the [RadixAttention](https://arxiv.org/abs/2312.07104) can still greatly accelerate many use cases with automatic KV cache reuse.
-
-### Usage
-Launch a server
-```
-python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000
-```
-
-Send a request
-```
-curl http://localhost:30000/generate \
-  -H "Content-Type: application/json" \
-  -d '{
-    "text": "Once upon a time,",
-    "sampling_params": {
-      "max_new_tokens": 16,
-      "temperature": 0
-    }
-  }'
-```
-Learn more about the argument format [here](docs/sampling_params.md).
-
-### OpenAI Compatible API
-In addition, the server supports an experimental OpenAI-compatible API.
-
-```python
-import openai
-client = openai.Client(
-    base_url="http://127.0.0.1:30000/v1", api_key="EMPTY")
-
-# Text completion
-response = client.completions.create(
-	model="default",
-	prompt="The capital of France is",
-	temperature=0,
-	max_tokens=32,
-)
-print(response)
-
-# Chat completion
-response = client.chat.completions.create(
-    model="default",
-    messages=[
-        {"role": "system", "content": "You are a helpful AI assistant"},
-        {"role": "user", "content": "List 3 countries and their capitals."},
-    ],
-    temperature=0,
-    max_tokens=64,
-)
-print(response)
-```
-
-By default, the server uses the chat template specified in the model tokenizer from Hugging Face. It should just work for most official models such as Llama-2/Llama-3.
-
-If needed, you can also override the chat template when launching the server:
-
-```
-python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000 --chat-template llama-2
-```
-
-If the chat template you are looking for is missing, you are welcome to contribute it.
-Meanwhile, you can also temporarily register your chat template as follows:
-
-```json
-{
-  "name": "my_model",
-  "system": "<|im_start|>system",
-  "user": "<|im_start|>user",
-  "assistant": "<|im_start|>assistant",
-  "sep_style": "CHATML",
-  "sep": "<|im_end|>",
-  "stop_str": ["<|im_end|>", "<|im_start|>"]
-}
-```
-
-```
-python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000 --chat-template ./my_model_template.json
-```
-
-### Additional Arguments
-- Add `--tp 2` to enable tensor parallelism. If it indicates `peer access is not supported between these two devices`, add `--enable-p2p-check` option.
-```
-python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000 --tp 2
-```
-- Add `--dp 2` to enable data parallelism. It can also be used together with tp. Data parallelism is better for throughput if there is enough memory.
-```
-python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000 --dp 2 --tp 2
-```
-- If you see out-of-memory errors during serving, please try to reduce the memory usage of the KV cache pool by setting a smaller value of `--mem-fraction-static`. The default value is `0.9`
-```
-python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --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.
-```
-# Node 0
-python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --tp 4 --nccl-init sgl-dev-1:50000 --nnodes 2 --node-rank 0
-
-# Node 1
-python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --tp 4 --nccl-init sgl-dev-1:50000 --nnodes 2 --node-rank 1
-```
-
-### Supported Models
-- Llama
-- Mistral
-- Mixtral
-- Qwen / Qwen 2 / Qwen 2 MoE
-- Gemma / Gemma 2
-  - `python -m sglang.launch_server --model-path google/gemma-7b-it --port 30000 --attention-reduce-in-fp32`
-- LLaVA
-  - `python3 -m sglang.launch_server --model-path liuhaotian/llava-v1.5-7b --tokenizer-path llava-hf/llava-1.5-7b-hf --chat-template vicuna_v1.1 --port 30000`
-  - `python3 -m sglang.launch_server --model-path liuhaotian/llava-v1.6-vicuna-7b --tokenizer-path llava-hf/llava-1.5-7b-hf --chat-template vicuna_v1.1 --port 30000`
-  - `python3 -m sglang.launch_server --model-path liuhaotian/llava-v1.6-34b --tokenizer-path liuhaotian/llava-v1.6-34b-tokenizer --port 3000`
-- LLaVA-NeXT-Video
-  - see [srt_example_llava_v.sh](examples/usage/llava_video/srt_example_llava_v.sh)
-- Yi-VL
-  - see [srt_example_yi_vl.py](examples/quick_start/srt_example_yi_vl.py).
-- StableLM
-- Command-R
-- DBRX
-- Grok
-- ChatGLM
-- AWQ/GPTQ/Marlin quantization
-
-Instructions for supporting a new model are [here](https://github.com/sgl-project/sglang/blob/main/docs/model_support.md).
-
 ## Benchmark And Performance
 - Llama-7B on NVIDIA A10G, FP16, Tensor Parallelism=1
 ![llama_7b](assets/llama_7b.jpg)
@@ -667,18 +667,8 @@ Instructions for supporting a new model are [here](https://github.com/sgl-projec
 - Synthetic latency and throughput benchmark [scripts](https://github.com/sgl-project/sglang/tree/main/benchmark/latency_throughput).
 
 ## Roadmap
-https://github.com/sgl-project/sglang/issues/157
+[Development Roadmap (2024 Q3)](https://github.com/sgl-project/sglang/issues/634)
 
 ## Citation And Acknowledgment
-```
-@misc{zheng2024sglang,
-      title={SGLang: Efficient Execution of Structured Language Model Programs},
-      author={Lianmin Zheng and Liangsheng Yin and Zhiqiang Xie and Chuyue Sun and Jeff Huang and Cody Hao Yu and Shiyi Cao and Christos Kozyrakis and Ion Stoica and Joseph E. Gonzalez and Clark Barrett and Ying Sheng},
-      year={2024},
-      eprint={2312.07104},
-      archivePrefix={arXiv},
-      primaryClass={cs.AI}
-}
-```
-
-We learned from the design and reused some code of the following projects: [Guidance](https://github.com/guidance-ai/guidance), [vLLM](https://github.com/vllm-project/vllm), [LightLLM](https://github.com/ModelTC/lightllm), [FlashInfer](https://github.com/flashinfer-ai/flashinfer), [Outlines](https://github.com/outlines-dev/outlines), [LMQL](https://github.com/eth-sri/lmql).
+Please cite our paper, [SGLang: Efficient Execution of Structured Language Model Programs](https://arxiv.org/abs/2312.07104), if you find the project useful.
+We also learned from the design and reused code from the following projects: [Guidance](https://github.com/guidance-ai/guidance), [vLLM](https://github.com/vllm-project/vllm), [LightLLM](https://github.com/ModelTC/lightllm), [FlashInfer](https://github.com/flashinfer-ai/flashinfer), [Outlines](https://github.com/outlines-dev/outlines), and [LMQL](https://github.com/eth-sri/lmql).