Qwen3
1. Model Introduction
Qwen3 series are the most powerful vision-language models in the Qwen series to date, featuring advanced capabilities in multi-modal understanding, reasoning, and agentic applications.
This generation delivers comprehensive upgrades across the board:
- Stronger general intelligence: Significant improvements in instruction following, logical reasoning, text comprehension, mathematics, science, coding, and tool usage.
- Broader multilingual knowledge: Substantial gains in long-tail knowledge coverage across multiple languages.
- More helpful & aligned responses: Markedly better alignment with user preferences in subjective and open-ended tasks, enabling higher-quality, more useful text generation.
- Extended context length: Enhanced capabilities in understanding and reasoning over 256K-token long contexts.
- Stronger agent interaction capabilities: Improved tool use and search-based agent performance
- Flexible deployment options: Available in Dense and MoE architectures that scale from edge to cloud, with Instruct and reasoning-enhanced Thinking editions
For more details, please refer to the official Qwen3 GitHub Repository.
2. SGLang Installation
SGLang offers multiple installation methods. You can choose the most suitable installation method based on your hardware platform and requirements.
Please refer to the official SGLang installation guide for installation instructions.
3. Model Deployment
This section provides deployment configurations optimized for different hardware platforms and use cases.
3.1 Basic Configuration
The Qwen3 series offers models in various sizes and architectures, optimized for different hardware platforms. The recommended launch configurations vary by hardware and model size.
Interactive Command Generator: Use the configuration selector below to automatically generate the appropriate deployment command for your hardware platform, model size, quantization method, and thinking capabilities.
1Hardware Platform
2Model Size
3Quantization
4Capabilities
5Tool Call Parser
Generated Command
python -m sglang.launch_server \ --model Qwen/Qwen3-235B-A22B \ --tp 8
3.2 Configuration Tips
- Memory Management : Set lower
--context-lengthto conserve memory. A value of128000is sufficient for most scenarios, down from the default 262K. - Expert Parallelism : SGLang supports Expert Parallelism (EP) via
--ep, allowing experts in MoE models to be deployed on separate GPUs for better throughput. One thing to note is that, for quantized models, you need to set--epto a value that satisfies the requirement:(moe_intermediate_size / moe_tp_size) % weight_block_size_n == 0, where moe_tp_size is equal to tp_size divided by ep_size.Note that EP may perform worse in low concurrency scenarios due to additional communication overhead. Check out Expert Parallelism Deployment for more details. - Kernel Tuning : For MoE Triton kernel tuning on your specific hardware, refer to fused_moe_triton.
- Speculative Decoding: Using Speculative Decoding for latency-sensitive scenarios.
--speculative-algorithm EAGLE3: Speculative decoding algorithm--speculative-num-steps 3: Number of speculative verification rounds--speculative-eagle-topk 1: Top-k sampling for draft tokens--speculative-num-draft-tokens 4: Number of draft tokens per step--speculative-draft-model-path: The path of the draft model weights. This can be a local folder or a Hugging Face repo ID.
4. Model Invocation
4.1 Basic Usage
For basic API usage and request examples, please refer to:
4.2 Advanced Usage
4.2.1 Reasoning Parser
Qwen3-235B-A22B supports reasoning mode. Enable the reasoning parser during deployment to separate the thinking and content sections:
python -m sglang.launch_server \
--model Qwen/Qwen3-235B-A22B \
--reasoning-parser qwen3 \
--tp 8 \
--host 0.0.0.0 \
--port 8000
Streaming with Thinking Process:
from openai import OpenAI
client = OpenAI(
base_url="http://localhost:8000/v1",
api_key="EMPTY"
)
# Enable streaming to see the thinking process in real-time
response = client.chat.completions.create(
model="Qwen/Qwen3-235B-A22B",
messages=[
{"role": "user", "content": "Solve this problem step by step: What is 15% of 240?"}
],
temperature=0.7,
max_tokens=2048,
stream=True
)
# Process the stream
has_thinking = False
has_answer = False
thinking_started = False
for chunk in response:
if chunk.choices and len(chunk.choices) > 0:
delta = chunk.choices[0].delta
# Print thinking process
if hasattr(delta, 'reasoning_content') and delta.reasoning_content:
if not thinking_started:
print("=============== Thinking =================", flush=True)
thinking_started = True
has_thinking = True
print(delta.reasoning_content, end="", flush=True)
# Print answer content
if delta.content:
# Close thinking section and add content header
if has_thinking and not has_answer:
print("\n=============== Content =================", flush=True)
has_answer = True
print(delta.content, end="", flush=True)
print()
Output Example:
=============== Thinking =================
Okay, so I need to figure out what 15% of 240 is. Hmm, percentages can sometimes trip me up, but I think I remember some basics. Let me start by recalling that "percent" means "per hundred," so 15% is the same as 15 per 100, or 15/100. So, maybe I can convert 15% into a decimal first? Yeah, I think that's a common method.
...
So conclusion: The answer is 36.
=============== Content =================
To determine what 15% of 240 is, we can follow a systematic approach that involves converting the percentage to a decimal and then performing multiplication. Here's a step-by-step breakdown of the solution:
....
### Final Answer:
$$
\boxed{36}
$$
Thus, 15% of 240 is **36**.
Note: The reasoning parser captures the model's step-by-step thinking process, allowing you to see how the model arrives at its conclusions.
4.2.3 Tool Calling
Qwen3 supports tool calling capabilities. Enable the tool call parser:
python -m sglang.launch_server \
--model Qwen/Qwen3-235B-A22B \
--reasoning-parser qwen3 \
--tool-call-parser qwen3 \
--tp 8 \
--host 0.0.0.0 \
--port 8000
Python Example (with Thinking Process):
from openai import OpenAI
client = OpenAI(
base_url="http://localhost:8000/v1",
api_key="EMPTY"
)
# Define available tools
tools = [
{
"type": "function",
"function": {
"name": "get_weather",
"description": "Get the current weather for a location",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city name"
},
"unit": {
"type": "string",
"enum": ["celsius", "fahrenheit"],
"description": "Temperature unit"
}
},
"required": ["location"]
}
}
}
]
# Make request with streaming to see thinking process
response = client.chat.completions.create(
model="Qwen/Qwen3-235B-A22B",
messages=[
{"role": "user", "content": "What's the weather in Beijing?"}
],
tools=tools,
temperature=0.7,
stream=True
)
# Process streaming response
thinking_started = False
has_thinking = False
tool_calls_accumulator = {}
for chunk in response:
if chunk.choices and len(chunk.choices) > 0:
delta = chunk.choices[0].delta
# Print thinking process
if hasattr(delta, 'reasoning_content') and delta.reasoning_content:
if not thinking_started:
print("=============== Thinking =================", flush=True)
thinking_started = True
has_thinking = True
print(delta.reasoning_content, end="", flush=True)
# Accumulate tool calls
if hasattr(delta, 'tool_calls') and delta.tool_calls:
# Close thinking section if needed
if has_thinking and thinking_started:
print("\n=============== Content =================\n", flush=True)
thinking_started = False
for tool_call in delta.tool_calls:
index = tool_call.index
if index not in tool_calls_accumulator:
tool_calls_accumulator[index] = {
'name': None,
'arguments': ''
}
if tool_call.function:
if tool_call.function.name:
tool_calls_accumulator[index]['name'] = tool_call.function.name
if tool_call.function.arguments:
tool_calls_accumulator[index]['arguments'] += tool_call.function.arguments
# Print content
if delta.content:
print(delta.content, end="", flush=True)
# Print accumulated tool calls
for index, tool_call in sorted(tool_calls_accumulator.items()):
print(f"🔧 Tool Call: {tool_call['name']}")
print(f" Arguments: {tool_call['arguments']}")
print()
Output Example:
=============== Thinking =================
Okay, the user is asking for the weather in Beijing. Let me check the tools available. There's a function called get_weather that takes location and unit parameters. The location is required, so I need to specify Beijing as the location. The unit is optional and can be either celsius or fahrenheit. Since the user didn't specify the unit, maybe I should default to a common one. In China, they usually use celsius, so I'll set unit to celsius. I'll call the get_weather function with location: Beijing and unit: celsius. That should get the current weather for them.
=============== Content =================
🔧 Tool Call: get_weather
Arguments: {"location": "Beijing", "unit": "celsius"}
Note:
- The reasoning parser shows how the model decides to use a tool
- Tool calls are clearly marked with the function name and arguments
- You can then execute the function and send the result back to continue the conversation
Handling Tool Call Results:
# After getting the tool call, execute the function
def get_weather(location, unit="celsius"):
# Your actual weather API call here
return f"The weather in {location} is 22°{unit[0].upper()} and sunny."
# Send tool result back to the model
messages = [
{"role": "user", "content": "What's the weather in Beijing?"},
{
"role": "assistant",
"content": None,
"tool_calls": [{
"id": "call_123",
"type": "function",
"function": {
"name": "get_weather",
"arguments": '{"location": "Beijing", "unit": "celsius"}'
}
}]
},
{
"role": "tool",
"tool_call_id": "call_123",
"content": get_weather("Beijing", "celsius")
}
]
final_response = client.chat.completions.create(
model="Qwen/Qwen3-235B-A22B",
messages=messages,
temperature=0.7
)
print(final_response.choices[0].message.content)
# Output: "The current weather in Beijing is **22°C** and **sunny**. A perfect day to enjoy outdoor activities! 🌞"
5. Benchmark
5.1 Speed Benchmark
Test Environment:
- Hardware: NVIDIA B200 GPU (8x)
- Model: Qwen3-235B-A22B
- Tensor Parallelism: 8
- sglang version: 0.5.6
We use SGLang's built-in benchmarking tool to conduct performance evaluation on the ShareGPT_Vicuna_unfiltered dataset. This dataset contains real conversation data and can better reflect performance in actual use scenarios.
5.1.1 Standard Scenario Benchmark
- Model Deployment Command:
python -m sglang.launch_server \
--model Qwen/Qwen3-235B-A22B \
--tp 8
5.1.1.1 Low Concurrency
- Benchmark Command:
python3 -m sglang.bench_serving \
--backend sglang \
--model Qwen/Qwen3-235B-A22B \
--dataset-name random \
--random-input-len 1000 \
--random-output-len 1000 \
--num-prompts 10 \
--max-concurrency 1
- Test Results:
============ Serving Benchmark Result ============
Backend: sglang
Traffic request rate: inf
Max request concurrency: 1
Successful requests: 10
Benchmark duration (s): 44.16
Total input tokens: 6101
Total input text tokens: 6101
Total input vision tokens: 0
Total generated tokens: 4210
Total generated tokens (retokenized): 4210
Request throughput (req/s): 0.23
Input token throughput (tok/s): 138.14
Output token throughput (tok/s): 95.32
Peak output token throughput (tok/s): 98.00
Peak concurrent requests: 2
Total token throughput (tok/s): 233.47
Concurrency: 1.00
----------------End-to-End Latency----------------
Mean E2E Latency (ms): 4414.85
Median E2E Latency (ms): 3499.95
---------------Time to First Token----------------
Mean TTFT (ms): 96.27
Median TTFT (ms): 65.22
P99 TTFT (ms): 359.71
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 10.27
Median TPOT (ms): 10.28
P99 TPOT (ms): 10.30
---------------Inter-Token Latency----------------
Mean ITL (ms): 10.28
Median ITL (ms): 10.28
P95 ITL (ms): 10.34
P99 ITL (ms): 10.40
Max ITL (ms): 10.76
==================================================
5.1.1.2 Medium Concurrency
- Benchmark Command:
python3 -m sglang.bench_serving \
--backend sglang \
--model Qwen/Qwen3-235B-A22B \
--dataset-name random \
--random-input-len 1000 \
--random-output-len 1000 \
--num-prompts 80 \
--max-concurrency 16
- Test Results:
============ Serving Benchmark Result ============
Backend: sglang
Traffic request rate: inf
Max request concurrency: 16
Successful requests: 80
Benchmark duration (s): 48.01
Total input tokens: 39668
Total input text tokens: 39668
Total input vision tokens: 0
Total generated tokens: 40725
Total generated tokens (retokenized): 40725
Request throughput (req/s): 1.67
Input token throughput (tok/s): 826.27
Output token throughput (tok/s): 848.29
Peak output token throughput (tok/s): 1152.00
Peak concurrent requests: 20
Total token throughput (tok/s): 1674.56
Concurrency: 13.58
----------------End-to-End Latency----------------
Mean E2E Latency (ms): 8150.37
Median E2E Latency (ms): 8608.58
---------------Time to First Token----------------
Mean TTFT (ms): 219.77
Median TTFT (ms): 76.69
P99 TTFT (ms): 817.57
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 15.88
Median TPOT (ms): 15.72
P99 TPOT (ms): 21.37
---------------Inter-Token Latency----------------
Mean ITL (ms): 15.61
Median ITL (ms): 14.73
P95 ITL (ms): 15.50
P99 ITL (ms): 53.54
Max ITL (ms): 335.18
==================================================
5.1.1.3 High Concurrency
- Benchmark Command:
python3 -m sglang.bench_serving \
--backend sglang \
--model Qwen/Qwen3-235B-A22B \
--dataset-name random \
--random-input-len 1000 \
--random-output-len 1000 \
--num-prompts 500 \
--max-concurrency 100
- Test Results:
============ Serving Benchmark Result ============
Backend: sglang
Traffic request rate: inf
Max request concurrency: 100
Successful requests: 500
Benchmark duration (s): 86.32
Total input tokens: 249831
Total input text tokens: 249831
Total input vision tokens: 0
Total generated tokens: 252162
Total generated tokens (retokenized): 252131
Request throughput (req/s): 5.79
Input token throughput (tok/s): 2894.15
Output token throughput (tok/s): 2921.15
Peak output token throughput (tok/s): 4200.00
Peak concurrent requests: 111
Total token throughput (tok/s): 5815.30
Concurrency: 89.76
----------------End-to-End Latency----------------
Mean E2E Latency (ms): 15496.34
Median E2E Latency (ms): 15116.24
---------------Time to First Token----------------
Mean TTFT (ms): 325.26
Median TTFT (ms): 98.25
P99 TTFT (ms): 1484.77
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 30.52
Median TPOT (ms): 31.62
P99 TPOT (ms): 35.22
---------------Inter-Token Latency----------------
Mean ITL (ms): 30.14
Median ITL (ms): 23.78
P95 ITL (ms): 63.71
P99 ITL (ms): 80.10
Max ITL (ms): 1019.75
==================================================
5.1.2 Reasoning Scenario Benchmark
- Model Deployment Command:
python -m sglang.launch_server \
--model Qwen/Qwen3-235B-A22B \
--tp 8
5.1.2.1 Low Concurrency
- Benchmark Command:
python3 -m sglang.bench_serving \
--backend sglang \
--model Qwen/Qwen3-235B-A22B \
--dataset-name random \
--random-input-len 1000 \
--random-output-len 8000 \
--num-prompts 10 \
--max-concurrency 1
- Test Results:
============ Serving Benchmark Result ============
Backend: sglang
Traffic request rate: inf
Max request concurrency: 1
Successful requests: 10
Benchmark duration (s): 461.12
Total input tokens: 6101
Total input text tokens: 6101
Total input vision tokens: 0
Total generated tokens: 44452
Total generated tokens (retokenized): 44419
Request throughput (req/s): 0.02
Input token throughput (tok/s): 13.23
Output token throughput (tok/s): 96.40
Peak output token throughput (tok/s): 98.00
Peak concurrent requests: 2
Total token throughput (tok/s): 109.63
Concurrency: 1.00
----------------End-to-End Latency----------------
Mean E2E Latency (ms): 46110.08
Median E2E Latency (ms): 49701.17
---------------Time to First Token----------------
Mean TTFT (ms): 91.23
Median TTFT (ms): 64.99
P99 TTFT (ms): 313.50
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 10.33
Median TPOT (ms): 10.34
P99 TPOT (ms): 10.38
---------------Inter-Token Latency----------------
Mean ITL (ms): 10.35
Median ITL (ms): 10.35
P95 ITL (ms): 10.45
P99 ITL (ms): 10.48
Max ITL (ms): 11.68
==================================================
5.1.2.2 Medium Concurrency
- Benchmark Command:
python3 -m sglang.bench_serving \
--backend sglang \
--model Qwen/Qwen3-235B-A22B \
--dataset-name random \
--random-input-len 1000 \
--random-output-len 8000 \
--num-prompts 80 \
--max-concurrency 16
- Test Results:
============ Serving Benchmark Result ============
Backend: sglang
Traffic request rate: inf
Max request concurrency: 16
Successful requests: 80
Benchmark duration (s): 340.59
Total input tokens: 39668
Total input text tokens: 39668
Total input vision tokens: 0
Total generated tokens: 318226
Total generated tokens (retokenized): 318120
Request throughput (req/s): 0.23
Input token throughput (tok/s): 116.47
Output token throughput (tok/s): 934.33
Peak output token throughput (tok/s): 1136.00
Peak concurrent requests: 18
Total token throughput (tok/s): 1050.80
Concurrency: 13.84
----------------End-to-End Latency----------------
Mean E2E Latency (ms): 58925.28
Median E2E Latency (ms): 58874.55
---------------Time to First Token----------------
Mean TTFT (ms): 174.97
Median TTFT (ms): 72.43
P99 TTFT (ms): 596.87
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 14.83
Median TPOT (ms): 14.90
P99 TPOT (ms): 15.26
---------------Inter-Token Latency----------------
Mean ITL (ms): 14.77
Median ITL (ms): 14.77
P95 ITL (ms): 15.30
P99 ITL (ms): 15.67
Max ITL (ms): 150.37
==================================================
5.1.2.3 High Concurrency
- Benchmark Command:
python3 -m sglang.bench_serving \
--backend sglang \
--model Qwen/Qwen3-235B-A22B \
--dataset-name random \
--random-input-len 1000 \
--random-output-len 8000 \
--num-prompts 320 \
--max-concurrency 64
- Test Results:
============ Serving Benchmark Result ============
Backend: sglang
Traffic request rate: inf
Max request concurrency: 64
Successful requests: 320
Benchmark duration (s): 534.33
Total input tokens: 158939
Total input text tokens: 158939
Total input vision tokens: 0
Total generated tokens: 1300705
Total generated tokens (retokenized): 1300016
Request throughput (req/s): 0.60
Input token throughput (tok/s): 297.45
Output token throughput (tok/s): 2434.26
Peak output token throughput (tok/s): 2944.00
Peak concurrent requests: 68
Total token throughput (tok/s): 2731.71
Concurrency: 56.33
----------------End-to-End Latency----------------
Mean E2E Latency (ms): 94064.17
Median E2E Latency (ms): 96651.55
---------------Time to First Token----------------
Mean TTFT (ms): 291.24
Median TTFT (ms): 86.55
P99 TTFT (ms): 1218.49
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 23.22
Median TPOT (ms): 23.49
P99 TPOT (ms): 24.46
---------------Inter-Token Latency----------------
Mean ITL (ms): 23.08
Median ITL (ms): 22.85
P95 ITL (ms): 23.59
P99 ITL (ms): 59.35
Max ITL (ms): 650.89
==================================================
5.1.3 Summarization Scenario Benchmark
5.1.3.1 Low Concurrency
- Benchmark Command:
python3 -m sglang.bench_serving \
--backend sglang \
--model Qwen/Qwen3-235B-A22B \
--dataset-name random \
--random-input-len 8000 \
--random-output-len 1000 \
--num-prompts 10 \
--max-concurrency 1
- Test Results:
============ Serving Benchmark Result ============
Backend: sglang
Traffic request rate: inf
Max request concurrency: 1
Successful requests: 10
Benchmark duration (s): 45.24
Total input tokens: 41941
Total input text tokens: 41941
Total input vision tokens: 0
Total generated tokens: 4210
Total generated tokens (retokenized): 4210
Request throughput (req/s): 0.22
Input token throughput (tok/s): 927.04
Output token throughput (tok/s): 93.06
Peak output token throughput (tok/s): 98.00
Peak concurrent requests: 2
Total token throughput (tok/s): 1020.09
Concurrency: 1.00
----------------End-to-End Latency----------------
Mean E2E Latency (ms): 4522.50
Median E2E Latency (ms): 3640.13
---------------Time to First Token----------------
Mean TTFT (ms): 159.87
Median TTFT (ms): 147.35
P99 TTFT (ms): 334.09
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 10.38
Median TPOT (ms): 10.39
P99 TPOT (ms): 10.46
---------------Inter-Token Latency----------------
Mean ITL (ms): 10.39
Median ITL (ms): 10.39
P95 ITL (ms): 10.47
P99 ITL (ms): 10.50
Max ITL (ms): 11.29
==================================================
5.1.3.2 Medium Concurrency
- Benchmark Command:
python3 -m sglang.bench_serving \
--backend sglang \
--model Qwen/Qwen3-235B-A22B \
--dataset-name random \
--random-input-len 8000 \
--random-output-len 1000 \
--num-prompts 80 \
--max-concurrency 16
- Test Results:
============ Serving Benchmark Result ============
Backend: sglang
Traffic request rate: inf
Max request concurrency: 16
Successful requests: 80
Benchmark duration (s): 53.01
Total input tokens: 300020
Total input text tokens: 300020
Total input vision tokens: 0
Total generated tokens: 41589
Total generated tokens (retokenized): 41579
Request throughput (req/s): 1.51
Input token throughput (tok/s): 5659.39
Output token throughput (tok/s): 784.51
Peak output token throughput (tok/s): 1152.00
Peak concurrent requests: 20
Total token throughput (tok/s): 6443.90
Concurrency: 13.93
----------------End-to-End Latency----------------
Mean E2E Latency (ms): 9234.00
Median E2E Latency (ms): 9575.11
---------------Time to First Token----------------
Mean TTFT (ms): 388.55
Median TTFT (ms): 182.46
P99 TTFT (ms): 1926.02
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 17.46
Median TPOT (ms): 17.41
P99 TPOT (ms): 28.28
---------------Inter-Token Latency----------------
Mean ITL (ms): 17.05
Median ITL (ms): 14.28
P95 ITL (ms): 14.88
P99 ITL (ms): 100.67
Max ITL (ms): 1431.38
==================================================
5.1.3.3 High Concurrency
- Benchmark Command:
python3 -m sglang.bench_serving \
--backend sglang \
--model Qwen/Qwen3-235B-A22B \
--dataset-name random \
--random-input-len 8000 \
--random-output-len 1000 \
--num-prompts 320 \
--max-concurrency 64
- Test Results:
============ Serving Benchmark Result ============
Backend: sglang
Traffic request rate: inf
Max request concurrency: 64
Successful requests: 320
Benchmark duration (s): 103.23
Total input tokens: 1273893
Total input text tokens: 1273893
Total input vision tokens: 0
Total generated tokens: 169680
Total generated tokens (retokenized): 169679
Request throughput (req/s): 3.10
Input token throughput (tok/s): 12340.40
Output token throughput (tok/s): 1643.72
Peak output token throughput (tok/s): 2750.00
Peak concurrent requests: 71
Total token throughput (tok/s): 13984.11
Concurrency: 59.06
----------------End-to-End Latency----------------
Mean E2E Latency (ms): 19051.11
Median E2E Latency (ms): 18542.35
---------------Time to First Token----------------
Mean TTFT (ms): 896.85
Median TTFT (ms): 215.93
P99 TTFT (ms): 6302.90
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 35.36
Median TPOT (ms): 36.07
P99 TPOT (ms): 57.59
---------------Inter-Token Latency----------------
Mean ITL (ms): 34.30
Median ITL (ms): 22.38
P95 ITL (ms): 110.00
P99 ITL (ms): 165.88
Max ITL (ms): 5855.77
==================================================
5.2 Accuracy Benchmark
5.2.1 GSM8K Benchmark
- Benchmark Command:
python3 -m sglang.test.few_shot_gsm8k --num-questions 200
-
Results:
- Qwen/Qwen3-235B-A22B
Accuracy: 0.945
Invalid: 0.000
Latency: 11.980 s
Output throughput: 2358.105 token/s
- Qwen/Qwen3-235B-A22B