chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[Docs] Add Falcon H1, Hunyuan-Large, Qwen3-Omni support and update Diffusion usage (#17888)

Browse Source 
Co-authored-by: Rishitshivam <164783543+Rishitshivam@users.noreply.github.com>
Co-authored-by: Ratish P <114130421+Ratish1@users.noreply.github.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Adarsh Shirawalmath <114558126+adarshxs@users.noreply.github.com>
Co-authored-by: zhaochenyang20 <zhaochen20@outlook.com>
This commit is contained in:
Rishit Shivam
2026-02-07 02:47:51 +05:30
committed by GitHub
parent f2e0048d06
commit c850a8a41a
25 changed files with 664 additions and 558 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
docs/supported_models/extending/index.rst Normal file
View File

@@ -0,0 +1,11 @@
Extending SGLang
================
Adding new models and alternative backends.
.. toctree::
:maxdepth: 1
support_new_models.md
transformers_fallback.md
modelscope.md

0
docs/supported_models/modelscope.md → docs/supported_models/extending/modelscope.md
View File

640
docs/supported_models/support_new_models.md → docs/supported_models/extending/support_new_models.md
View File

@@ -1,320 +1,320 @@
# How to Support New Models
This document explains how to add support for new language models and multimodal large language models (MLLMs) in
SGLang. It also covers how to test new models and register external implementations.
## How to Support a New Language Model
To support a new model in SGLang, you only need to add a single file under
the [SGLang Models Directory](https://github.com/sgl-project/sglang/tree/main/python/sglang/srt/models). You can learn
from existing model implementations and create a new file for your model. For most models, you should be able to find a
similar model to start with (e.g., starting from Llama). Also refer how
to [port a Model from vLLM to SGLang](#port-a-model-from-vllm-to-sglang)
## How to Support a New Multimodal Large Language Model
To support a new multimodal large language model (MLLM) in SGLang, there are several key components in addition to the
standard LLM support:
1. **Register your new model as multimodal**:
Extend `is_multimodal_model`
in [model_config.py](https://github.com/sgl-project/sglang/blob/0ab3f437aba729b348a683ab32b35b214456efc7/python/sglang/srt/configs/model_config.py#L561)
to return `True` for your model.
2. **Register a new chat-template**:
Only when your default chat-template is unable to accept images as input: Register a new chat template in [conversation.py](https://github.com/sgl-project/sglang/tree/main/python/sglang/srt/conversation.py) and the corresponding matching function.
3. **Multimodal Data Processor**:
Define a new `Processor` class that inherits from `BaseMultimodalProcessor` and register this processor as your
model’s dedicated processor.
See [multimodal_processor.py](https://github.com/sgl-project/sglang/tree/main/python/sglang/srt/multimodal/processors)
for more details.
4. **Handle Multimodal Tokens**:
Implement a `pad_input_ids` function for your new model. In this function, multimodal tokens in the prompt should be
expanded (if necessary) and padded with multimodal-data-hashes so that SGLang can recognize different multimodal data
with `RadixAttention`.
5. **Handle Image Feature Extraction**:
Implement a `get_image_feature` function for your new model, which extracts image features from raw image data and converts them into the embeddings used by the language model.
6. **Adapt to Vision Attention**:
Adapt the multi-headed `Attention` of ViT with SGLang’s `VisionAttention`.
You can refer to [Qwen2VL](https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/models/qwen2_vl.py) or
other mllm implementations. These models demonstrate how to correctly handle both multimodal and textual inputs.
## Testing and Debugging
Please note all your testing and benchmarking results in PR description.
### Interactive Debugging
For interactive debugging, compare the outputs of Hugging Face/Transformers and SGLang. The following two commands
should give the same text output and very similar prefill logits:
- Get the reference output:
```bash
python3 scripts/playground/reference_hf.py --model-path [new model] --model-type {text,mllm}
```
- Get the SGLang output:
```bash
python3 -m sglang.bench_one_batch --correct --model [new model]
```
### Add the Model to the Test Suite
To ensure the new model is well maintained, add it to the test suite by including it in the `ALL_OTHER_MODELS` list in
the [test_generation_models.py](https://github.com/sgl-project/sglang/blob/main/test/srt/models/test_generation_models.py)
file, test the new model on your local machine and report the results on demonstrative benchmarks (GSM8K, MMLU, MMMU,
MMMU-Pro, etc.) in your PR. \\
For VLMs, also include a test in `test_vision_openai_server_{x}.py` (e.g. [test_vision_openai_server_a.py](https://github.com/sgl-project/sglang/blob/main/test/srt/test_vision_openai_server_a.py), [test_vision_openai_server_b.py](https://github.com/sgl-project/sglang/blob/main/test/srt/test_vision_openai_server_b.py)).
This is an example command to run to test a new model on your local machine:
```bash
ONLY_RUN=Qwen/Qwen2-1.5B python3 -m unittest test_generation_models.TestGenerationModels.test_others
```
### Benchmark
- **(Required) MMMU**: follow MMMU benchmark [README.md](https://github.com/sgl-project/sglang/blob/main/benchmark/mmmu/README.md) to get SGLang vs. HF Transformer accuracy comparison. The accuracy score from SGLang run should not be much lower than that from HF Transformer run. Similarly, follow https://docs.sglang.io/developer_guide/benchmark_and_profiling.html to get performance comparison: TTFT and throughput must meet or exceed baselines (e.g., HF Transformer).
- **(Optional) Other evals**: If you ran other evals, please note the results in PR description.
## Port a Model from vLLM to SGLang
The [vLLM Models Directory](https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/models) is a valuable
resource, as vLLM covers many models. SGLang reuses vLLM’s interface and some layers, making it easier to port models
from vLLM to SGLang.
To port a model from vLLM to SGLang:
- Compare these two files for guidance:
- [SGLang Llama Implementation](https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/models/llama.py)
- [vLLM Llama Implementation](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llama.py)
- The major differences include:
- **Replace vLLM’s `Attention` with `RadixAttention`** (ensure you pass `layer_id` to `RadixAttention`).
- **Replace vLLM’s `LogitsProcessor` with SGLang’s `LogitsProcessor`.**
- **Replace the multi-headed `Attention` of ViT with SGLang’s `VisionAttention`.**
- **Replace other vLLM layers** (such as `RMSNorm`, `SiluAndMul`) with SGLang layers.
- **Remove `Sample`.**
- **Change the `forward()` functions** and add a `forward_batch()` method.
- **Add `EntryClass`** at the end.
- **Ensure that the new implementation uses only SGLang components** and does not rely on any vLLM components.
Note: make sure you add your new model to the supported models list in the supported models documentation.
## Registering an External Model Implementation
In addition to the methods above, you can register your new model with the `ModelRegistry` before launching the server.
This allows you to integrate your model without modifying the source code.
For example:
```python
from sglang.srt.models.registry import ModelRegistry
from sglang.srt.entrypoints.http_server import launch_server
# For a single model, add it to the registry:
ModelRegistry.models[model_name] = model_class
# For multiple models, you can imitate the import_model_classes() function:
from functools import lru_cache
@lru_cache()
def import_new_model_classes():
model_arch_name_to_cls = {}
# Populate model_arch_name_to_cls with your new model classes.
...
return model_arch_name_to_cls
ModelRegistry.models.update(import_new_model_classes())
# Launch the server with your server arguments:
launch_server(server_args)
```
## Example: Implementing and Serving a Llama Wrapper Model
Below is an introductory, step-by-step walkthrough on how to implement a new model end-to-end in SGLang and then run it via the [Offline Engine](https://github.com/sgl-project/sglang/blob/main/docs/basic_usage/offline_engine_api.ipynb).
### Implementing Our Model
To keep things simple, this new model will be a simple wrapper around [Llama 3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct), and our goal will be just to bias the output logits for each `forward` call by taking the square root of each individual logit.
Let's start by defining our model in a file called `llama_wrapper.py`.
The first step is to import the necessary libraries from SRT, which is SGLang's internal backend.
```python
# In the file `llama_wrapper.py`
import torch
from transformers import LlamaConfig
from typing import Optional
from sglang.srt.layers.logits_processor import LogitsProcessorOutput
from sglang.srt.layers.quantization.base_config import QuantizationConfig
from sglang.srt.model_executor.forward_batch_info import ForwardBatch, PPProxyTensors
from sglang.srt.models.llama import LlamaForCausalLM
```
Next, we declare a new `class` for our model and have it inherit from `LlamaForCausalLM`, which allows our model to access `LlamaForCausalLM`'s predefined modules and layers, such as `LlamaAttention` and `LlamaMLP`.
Note that almost all model implementations take in `config` and `quant_config` as arguments for their `__init__` method; `config` and `quant_config` are passed in via [`model_loader/loader.py`](https://github.com/sgl-project/sglang/blob/bf72b80122fd888bf619d17b96fa3e323ab809fc/python/sglang/srt/model_loader/loader.py#L219).
Because we have inherited from `LlamaForCausalLM`, we can pass our parameters directly to its constructor, which will set the member variables for us.
```python
class LlamaWrapper(LlamaForCausalLM):
def __init__(
self,
config: LlamaConfig,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
) -> None:
super().__init__(config=config, quant_config=quant_config, prefix=prefix)
```
Now, we want to define the `forward` method, which is what will be called at inference time.
Note that the signature for `forward` is essentially the same for any model; you can take a look at the other models defined in the [`models` directory](https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/models/) for references.
To see where exactly `forward` is called in the SGLang runtime's internals, take a look at [`forward_decode`](https://github.com/sgl-project/sglang/blob/bf72b80122fd888bf619d17b96fa3e323ab809fc/python/sglang/srt/model_executor/model_runner.py#L1705) and [`forward_extend`](https://github.com/sgl-project/sglang/blob/bf72b80122fd888bf619d17b96fa3e323ab809fc/python/sglang/srt/model_executor/model_runner.py#L1724) in the [`ModelRunner` class](https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/model_executor/model_runner.py).
```python
@torch.no_grad()
def forward(
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
forward_batch: ForwardBatch,
pp_proxy_tensors: Optional[PPProxyTensors] = None,
input_embeds: Optional[torch.Tensor] = None,
get_embedding: bool = False,
) -> LogitsProcessorOutput:
```
We now call the `__call__` method for `self.model` (which is a member variable that `LlamaForCausalLM` defines in its `__init__` method), which eventually calls `LlamaForCausalLM`'s `forward` method.
After that, we feed the `hidden_states` into our model's `LogitsProcessor` (again defined in `LlamaForCausalLM`).
```python
hidden_states = self.model(
input_ids,
positions,
forward_batch,
input_embeds,
pp_proxy_tensors=pp_proxy_tensors,
)
res: LogitsProcessorOutput = self.logits_processor(
input_ids,
hidden_states,
self.lm_head,
forward_batch,
)
```
After receiving the logits for the next token, we can finally perform our biasing step.
```python
orig_logits = res.next_token_logits
res.next_token_logits = torch.where(
orig_logits > 0,
orig_logits.sqrt(),
orig_logits
)
return res
```
Now, our `LlamaWrapper` model is created and ready to be served!
### Serving Our Model Via SGLang's Offline Engine
The next step of this walkthrough involves hosting our new model offline, so that it can be served locally and without an HTTP server.
First, create a new file called `run.py`.
Now, we must ensure that SGLang's `ModelRegistry` can find our model.
To do this, we first download the model's configuration and weights from Huggingface.
```python
# In the file `run.py`
import asyncio
from functools import lru_cache
from huggingface_hub import snapshot_download
from llama_wrapper import LlamaWrapper # Make sure to import our new model!
import sglang as sgl
from sglang.srt.models.registry import ModelRegistry
# Make sure to request access to this model on Huggingface, then export your
# `HF_TOKEN` to download the model snapshot
llama_dir = snapshot_download(
repo_id="meta-llama/Llama-3.1-8B-Instruct",
local_dir="./llama_ckpt",
)
```
Now that we have our model on disk, we want to point it to `LlamaWrapper` by changing the `architectures` field in `./llama_ckpt/config.json` to be `LlamaWrapper`.
That way, when we pass in the path of our model checkpoint to SGLang, it will know that we want to use "LlamaWrapper" instead of "LlamaForCausalLM" as our model.
```python
{
"architectures": [
# "LlamaForCausalLM"
"LlamaWrapper"
],
...
}
```
However, if we don't link our `LlamaWrapper` class to the "LlamaWrapper" registry keyword, then SGLang won't be able to find our model.
Thus, to register our `LlamaWrapper`, we want to follow the steps in the above section titled "Registering an External Model Implementation".
```python
@lru_cache()
def import_new_model_classes():
model_arch_name_to_cls = {"LlamaWrapper": LlamaWrapper}
return model_arch_name_to_cls
ModelRegistry.models.update(import_new_model_classes())
```
Lastly, when we create our `Engine`, we just pass in the path to the local model directory.
Then, our `LlamaWrapper` is ready to be served; for this walkthrough, we will use SGLang `Engine`'s non-streaming asynchronous generation endpoint.
```python
def main():
llm = sgl.Engine(model_path="./llama_ckpt")
sampling_params = {"temperature": 0.2, "top_k": 5}
prompts = [
"Write a short, neutral self-introduction for a fictional character. Hello, my name is",
"Provide a concise factual statement about France’s capital city. The capital of France is",
"Explain possible future trends in artificial intelligence. The future of AI is",
]
asyncio.run(run_llm(llm, sampling_params, prompts))
llm.shutdown()
async def run_llm(
llm,
sampling_params,
prompts,
) -> None:
outputs = await llm.async_generate(prompts, sampling_params)
for prompt, output in zip(prompts, outputs):
print(f"\nPrompt: {prompt}")
print(f"Generated text: {output['text']}")
if __name__ == "__main__":
main()
```
Now, when we call `python run.py`, we will get the outputs of our newly created model!
## Documentation
Add to table of supported models in [generative_models.md](https://github.com/sgl-project/sglang/blob/main/docs/supported_models/generative_models.md) or [multimodal_language_models.md](https://github.com/sgl-project/sglang/blob/main/docs/supported_models/multimodal_language_models.md)
---
By following these guidelines, you can add support for new language models and multimodal large language models in
SGLang and ensure they are thoroughly tested and easily integrated into the system.
# How to Support New Models
This document explains how to add support for new language models and multimodal large language models (MLLMs) in
SGLang. It also covers how to test new models and register external implementations.
## How to Support a New Language Model
To support a new model in SGLang, you only need to add a single file under
the [SGLang Models Directory](https://github.com/sgl-project/sglang/tree/main/python/sglang/srt/models). You can learn
from existing model implementations and create a new file for your model. For most models, you should be able to find a
similar model to start with (e.g., starting from Llama). Also refer how
to [port a Model from vLLM to SGLang](#port-a-model-from-vllm-to-sglang)
## How to Support a New Multimodal Large Language Model
To support a new multimodal large language model (MLLM) in SGLang, there are several key components in addition to the
standard LLM support:
1. **Register your new model as multimodal**:
Extend `is_multimodal_model`
in [model_config.py](https://github.com/sgl-project/sglang/blob/0ab3f437aba729b348a683ab32b35b214456efc7/python/sglang/srt/configs/model_config.py#L561)
to return `True` for your model.
2. **Register a new chat-template**:
Only when your default chat-template is unable to accept images as input: Register a new chat template in [conversation.py](https://github.com/sgl-project/sglang/tree/main/python/sglang/srt/conversation.py) and the corresponding matching function.
3. **Multimodal Data Processor**:
Define a new `Processor` class that inherits from `BaseMultimodalProcessor` and register this processor as your
model’s dedicated processor.
See [multimodal_processor.py](https://github.com/sgl-project/sglang/tree/main/python/sglang/srt/multimodal/processors)
for more details.
4. **Handle Multimodal Tokens**:
Implement a `pad_input_ids` function for your new model. In this function, multimodal tokens in the prompt should be
expanded (if necessary) and padded with multimodal-data-hashes so that SGLang can recognize different multimodal data
with `RadixAttention`.
5. **Handle Image Feature Extraction**:
Implement a `get_image_feature` function for your new model, which extracts image features from raw image data and converts them into the embeddings used by the language model.
6. **Adapt to Vision Attention**:
Adapt the multi-headed `Attention` of ViT with SGLang’s `VisionAttention`.
You can refer to [Qwen2VL](https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/models/qwen2_vl.py) or
other mllm implementations. These models demonstrate how to correctly handle both multimodal and textual inputs.
## Testing and Debugging
Please note all your testing and benchmarking results in PR description.
### Interactive Debugging
For interactive debugging, compare the outputs of Hugging Face/Transformers and SGLang. The following two commands
should give the same text output and very similar prefill logits:
- Get the reference output:
```bash
python3 scripts/playground/reference_hf.py --model-path [new model] --model-type {text,mllm}
```
- Get the SGLang output:
```bash
python3 -m sglang.bench_one_batch --correct --model [new model]
```
### Add the Model to the Test Suite
To ensure the new model is well maintained, add it to the test suite by including it in the `ALL_OTHER_MODELS` list in
the [test_generation_models.py](https://github.com/sgl-project/sglang/blob/main/test/srt/models/test_generation_models.py)
file, test the new model on your local machine and report the results on demonstrative benchmarks (GSM8K, MMLU, MMMU,
MMMU-Pro, etc.) in your PR. \\
For VLMs, also include a test in `test_vision_openai_server_{x}.py` (e.g. [test_vision_openai_server_a.py](https://github.com/sgl-project/sglang/blob/main/test/srt/test_vision_openai_server_a.py), [test_vision_openai_server_b.py](https://github.com/sgl-project/sglang/blob/main/test/srt/test_vision_openai_server_b.py)).
This is an example command to run to test a new model on your local machine:
```bash
ONLY_RUN=Qwen/Qwen2-1.5B python3 -m unittest test_generation_models.TestGenerationModels.test_others
```
### Benchmark
- **(Required) MMMU**: follow MMMU benchmark [README.md](https://github.com/sgl-project/sglang/blob/main/benchmark/mmmu/README.md) to get SGLang vs. HF Transformer accuracy comparison. The accuracy score from SGLang run should not be much lower than that from HF Transformer run. Similarly, follow https://docs.sglang.io/developer_guide/benchmark_and_profiling.html to get performance comparison: TTFT and throughput must meet or exceed baselines (e.g., HF Transformer).
- **(Optional) Other evals**: If you ran other evals, please note the results in PR description.
## Port a Model from vLLM to SGLang
The [vLLM Models Directory](https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/models) is a valuable
resource, as vLLM covers many models. SGLang reuses vLLM’s interface and some layers, making it easier to port models
from vLLM to SGLang.
To port a model from vLLM to SGLang:
- Compare these two files for guidance:
- [SGLang Llama Implementation](https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/models/llama.py)
- [vLLM Llama Implementation](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llama.py)
- The major differences include:
- **Replace vLLM’s `Attention` with `RadixAttention`** (ensure you pass `layer_id` to `RadixAttention`).
- **Replace vLLM’s `LogitsProcessor` with SGLang’s `LogitsProcessor`.**
- **Replace the multi-headed `Attention` of ViT with SGLang’s `VisionAttention`.**
- **Replace other vLLM layers** (such as `RMSNorm`, `SiluAndMul`) with SGLang layers.
- **Remove `Sample`.**
- **Change the `forward()` functions** and add a `forward_batch()` method.
- **Add `EntryClass`** at the end.
- **Ensure that the new implementation uses only SGLang components** and does not rely on any vLLM components.
Note: make sure you add your new model to the supported models list in the supported models documentation.
## Registering an External Model Implementation
In addition to the methods above, you can register your new model with the `ModelRegistry` before launching the server.
This allows you to integrate your model without modifying the source code.
For example:
```python
from sglang.srt.models.registry import ModelRegistry
from sglang.srt.entrypoints.http_server import launch_server
# For a single model, add it to the registry:
ModelRegistry.models[model_name] = model_class
# For multiple models, you can imitate the import_model_classes() function:
from functools import lru_cache
@lru_cache()
def import_new_model_classes():
model_arch_name_to_cls = {}
# Populate model_arch_name_to_cls with your new model classes.
...
return model_arch_name_to_cls
ModelRegistry.models.update(import_new_model_classes())
# Launch the server with your server arguments:
launch_server(server_args)
```
## Example: Implementing and Serving a Llama Wrapper Model
Below is an introductory, step-by-step walkthrough on how to implement a new model end-to-end in SGLang and then run it via the [Offline Engine](https://github.com/sgl-project/sglang/blob/main/docs/basic_usage/offline_engine_api.ipynb).
### Implementing Our Model
To keep things simple, this new model will be a simple wrapper around [Llama 3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct), and our goal will be just to bias the output logits for each `forward` call by taking the square root of each individual logit.
Let's start by defining our model in a file called `llama_wrapper.py`.
The first step is to import the necessary libraries from SRT, which is SGLang's internal backend.
```python
# In the file `llama_wrapper.py`
import torch
from transformers import LlamaConfig
from typing import Optional
from sglang.srt.layers.logits_processor import LogitsProcessorOutput
from sglang.srt.layers.quantization.base_config import QuantizationConfig
from sglang.srt.model_executor.forward_batch_info import ForwardBatch, PPProxyTensors
from sglang.srt.models.llama import LlamaForCausalLM
```
Next, we declare a new `class` for our model and have it inherit from `LlamaForCausalLM`, which allows our model to access `LlamaForCausalLM`'s predefined modules and layers, such as `LlamaAttention` and `LlamaMLP`.
Note that almost all model implementations take in `config` and `quant_config` as arguments for their `__init__` method; `config` and `quant_config` are passed in via [`model_loader/loader.py`](https://github.com/sgl-project/sglang/blob/bf72b80122fd888bf619d17b96fa3e323ab809fc/python/sglang/srt/model_loader/loader.py#L219).
Because we have inherited from `LlamaForCausalLM`, we can pass our parameters directly to its constructor, which will set the member variables for us.
```python
class LlamaWrapper(LlamaForCausalLM):
def __init__(
self,
config: LlamaConfig,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
) -> None:
super().__init__(config=config, quant_config=quant_config, prefix=prefix)
```
Now, we want to define the `forward` method, which is what will be called at inference time.
Note that the signature for `forward` is essentially the same for any model; you can take a look at the other models defined in the [`models` directory](https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/models/) for references.
To see where exactly `forward` is called in the SGLang runtime's internals, take a look at [`forward_decode`](https://github.com/sgl-project/sglang/blob/bf72b80122fd888bf619d17b96fa3e323ab809fc/python/sglang/srt/model_executor/model_runner.py#L1705) and [`forward_extend`](https://github.com/sgl-project/sglang/blob/bf72b80122fd888bf619d17b96fa3e323ab809fc/python/sglang/srt/model_executor/model_runner.py#L1724) in the [`ModelRunner` class](https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/model_executor/model_runner.py).
```python
@torch.no_grad()
def forward(
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
forward_batch: ForwardBatch,
pp_proxy_tensors: Optional[PPProxyTensors] = None,
input_embeds: Optional[torch.Tensor] = None,
get_embedding: bool = False,
) -> LogitsProcessorOutput:
```
We now call the `__call__` method for `self.model` (which is a member variable that `LlamaForCausalLM` defines in its `__init__` method), which eventually calls `LlamaForCausalLM`'s `forward` method.
After that, we feed the `hidden_states` into our model's `LogitsProcessor` (again defined in `LlamaForCausalLM`).
```python
hidden_states = self.model(
input_ids,
positions,
forward_batch,
input_embeds,
pp_proxy_tensors=pp_proxy_tensors,
)
res: LogitsProcessorOutput = self.logits_processor(
input_ids,
hidden_states,
self.lm_head,
forward_batch,
)
```
After receiving the logits for the next token, we can finally perform our biasing step.
```python
orig_logits = res.next_token_logits
res.next_token_logits = torch.where(
orig_logits > 0,
orig_logits.sqrt(),
orig_logits
)
return res
```
Now, our `LlamaWrapper` model is created and ready to be served!
### Serving Our Model Via SGLang's Offline Engine
The next step of this walkthrough involves hosting our new model offline, so that it can be served locally and without an HTTP server.
First, create a new file called `run.py`.
Now, we must ensure that SGLang's `ModelRegistry` can find our model.
To do this, we first download the model's configuration and weights from Huggingface.
```python
# In the file `run.py`
import asyncio
from functools import lru_cache
from huggingface_hub import snapshot_download
from llama_wrapper import LlamaWrapper # Make sure to import our new model!
import sglang as sgl
from sglang.srt.models.registry import ModelRegistry
# Make sure to request access to this model on Huggingface, then export your
# `HF_TOKEN` to download the model snapshot
llama_dir = snapshot_download(
repo_id="meta-llama/Llama-3.1-8B-Instruct",
local_dir="./llama_ckpt",
)
```
Now that we have our model on disk, we want to point it to `LlamaWrapper` by changing the `architectures` field in `./llama_ckpt/config.json` to be `LlamaWrapper`.
That way, when we pass in the path of our model checkpoint to SGLang, it will know that we want to use "LlamaWrapper" instead of "LlamaForCausalLM" as our model.
```python
{
"architectures": [
# "LlamaForCausalLM"
"LlamaWrapper"
],
...
}
```
However, if we don't link our `LlamaWrapper` class to the "LlamaWrapper" registry keyword, then SGLang won't be able to find our model.
Thus, to register our `LlamaWrapper`, we want to follow the steps in the above section titled "Registering an External Model Implementation".
```python
@lru_cache()
def import_new_model_classes():
model_arch_name_to_cls = {"LlamaWrapper": LlamaWrapper}
return model_arch_name_to_cls
ModelRegistry.models.update(import_new_model_classes())
```
Lastly, when we create our `Engine`, we just pass in the path to the local model directory.
Then, our `LlamaWrapper` is ready to be served; for this walkthrough, we will use SGLang `Engine`'s non-streaming asynchronous generation endpoint.
```python
def main():
llm = sgl.Engine(model_path="./llama_ckpt")
sampling_params = {"temperature": 0.2, "top_k": 5}
prompts = [
"Write a short, neutral self-introduction for a fictional character. Hello, my name is",
"Provide a concise factual statement about France’s capital city. The capital of France is",
"Explain possible future trends in artificial intelligence. The future of AI is",
]
asyncio.run(run_llm(llm, sampling_params, prompts))
llm.shutdown()
async def run_llm(
llm,
sampling_params,
prompts,
) -> None:
outputs = await llm.async_generate(prompts, sampling_params)
for prompt, output in zip(prompts, outputs):
print(f"\nPrompt: {prompt}")
print(f"Generated text: {output['text']}")
if __name__ == "__main__":
main()
```
Now, when we call `python run.py`, we will get the outputs of our newly created model!
## Documentation
Add to table of supported models in [generative_models.md](../text_generation/generative_models.md) or [multimodal_language_models.md](../text_generation/multimodal_language_models.md)
---
By following these guidelines, you can add support for new language models and multimodal large language models in
SGLang and ensure they are thoroughly tested and easily integrated into the system.

0
docs/supported_models/transformers_fallback.md → docs/supported_models/extending/transformers_fallback.md
View File

241
docs/supported_models/diffusion_models.md → docs/supported_models/image_generation/diffusion_models.md
View File

@@ -1,5 +1,7 @@
# Diffusion Models
> This page covers **image and video generation**. For **text generation** using diffusion LLMs (e.g., LLaDA2.0), see [Diffusion Language Models](../text_generation/diffusion_language_models.md).
SGLang Diffusion is an inference framework for accelerated image and video generation using diffusion models. It provides an end-to-end unified pipeline with optimized kernels from sgl-kernel and an efficient scheduler loop.
## Key Features
@@ -93,33 +95,33 @@ default parameters when initializing and generating videos.
### Video Generation Models
| Model Name | Hugging Face Model ID | Resolutions | TeaCache | Sliding Tile Attn | Sage Attn | Video Sparse Attention (VSA) |
|:-----------------------------|:--------------------------------------------------|:--------------------|:--------:|:-----------------:|:---------:|:----------------------------:|
| FastWan2.1 T2V 1.3B | `FastVideo/FastWan2.1-T2V-1.3B-Diffusers` | 480p | ⭕ | ⭕ | ⭕ | ✅ |
| FastWan2.2 TI2V 5B Full Attn | `FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers` | 720p | ⭕ | ⭕ | ⭕ | ✅ |
| Wan2.2 TI2V 5B | `Wan-AI/Wan2.2-TI2V-5B-Diffusers` | 720p | ⭕ | ⭕ | ✅ | ⭕ |
| Wan2.2 T2V A14B | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` | 480p<br>720p | ❌ | ❌ | ✅ | ⭕ |
| Wan2.2 I2V A14B | `Wan-AI/Wan2.2-I2V-A14B-Diffusers` | 480p<br>720p | ❌ | ❌ | ✅ | ⭕ |
| HunyuanVideo | `hunyuanvideo-community/HunyuanVideo` | 720×1280<br>544×960 | ❌ | ✅ | ✅ | ⭕ |
| FastHunyuan | `FastVideo/FastHunyuan-diffusers` | 720×1280<br>544×960 | ❌ | ✅ | ✅ | ⭕ |
| Wan2.1 T2V 1.3B | `Wan-AI/Wan2.1-T2V-1.3B-Diffusers` | 480p | ✅ | ✅ | ✅ | ⭕ |
| Wan2.1 T2V 14B | `Wan-AI/Wan2.1-T2V-14B-Diffusers` | 480p, 720p | ✅ | ✅ | ✅ | ⭕ |
| Wan2.1 I2V 480P | `Wan-AI/Wan2.1-I2V-14B-480P-Diffusers` | 480p | ✅ | ✅ | ✅ | ⭕ |
| Wan2.1 I2V 720P | `Wan-AI/Wan2.1-I2V-14B-720P-Diffusers` | 720p | ✅ | ✅ | ✅ | ⭕ |
| :--------------------------- | :------------------------------------------------ | :------------------ | :------: | :---------------: | :-------: | :--------------------------: |
| FastWan2.1 T2V 1.3B | `FastVideo/FastWan2.1-T2V-1.3B-Diffusers` | 480p | ⭕ | ⭕ | ⭕ | ✅ |
| FastWan2.2 TI2V 5B Full Attn | `FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers` | 720p | ⭕ | ⭕ | ⭕ | ✅ |
| Wan2.2 TI2V 5B | `Wan-AI/Wan2.2-TI2V-5B-Diffusers` | 720p | ⭕ | ⭕ | ✅ | ⭕ |
| Wan2.2 T2V A14B | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` | 480p<br>720p | ❌ | ❌ | ✅ | ⭕ |
| Wan2.2 I2V A14B | `Wan-AI/Wan2.2-I2V-A14B-Diffusers` | 480p<br>720p | ❌ | ❌ | ✅ | ⭕ |
| HunyuanVideo | `hunyuanvideo-community/HunyuanVideo` | 720×1280<br>544×960 | ❌ | ✅ | ✅ | ⭕ |
| FastHunyuan | `FastVideo/FastHunyuan-diffusers` | 720×1280<br>544×960 | ❌ | ✅ | ✅ | ⭕ |
| Wan2.1 T2V 1.3B | `Wan-AI/Wan2.1-T2V-1.3B-Diffusers` | 480p | ✅ | ✅ | ✅ | ⭕ |
| Wan2.1 T2V 14B | `Wan-AI/Wan2.1-T2V-14B-Diffusers` | 480p, 720p | ✅ | ✅ | ✅ | ⭕ |
| Wan2.1 I2V 480P | `Wan-AI/Wan2.1-I2V-14B-480P-Diffusers` | 480p | ✅ | ✅ | ✅ | ⭕ |
| Wan2.1 I2V 720P | `Wan-AI/Wan2.1-I2V-14B-720P-Diffusers` | 720p | ✅ | ✅ | ✅ | ⭕ |
**Note**: Wan2.2 TI2V 5B has some quality issues when performing I2V generation. We are working on fixing this issue.
### Image Generation Models
| Model Name | HuggingFace Model ID | Resolutions |
|:-----------------|:----------------------------------------|:---------------|
| FLUX.1-dev | `black-forest-labs/FLUX.1-dev` | Any resolution |
| FLUX.2-dev | `black-forest-labs/FLUX.2-dev` | Any resolution |
| FLUX.2-Klein | `black-forest-labs/FLUX.2-klein-4B` | Any resolution |
| Z-Image-Turbo | `Tongyi-MAI/Z-Image-Turbo` | Any resolution |
| GLM-Image | `zai-org/GLM-Image` | Any resolution |
| Qwen Image | `Qwen/Qwen-Image` | Any resolution |
| Qwen Image 2512 | `Qwen/Qwen-Image-2512` | Any resolution |
| Qwen Image Edit | `Qwen/Qwen-Image-Edit` | Any resolution |
| Model Name | HuggingFace Model ID | Resolutions |
| :-------------- | :---------------------------------- | :------------- |
| FLUX.1-dev | `black-forest-labs/FLUX.1-dev` | Any resolution |
| FLUX.2-dev | `black-forest-labs/FLUX.2-dev` | Any resolution |
| FLUX.2-Klein | `black-forest-labs/FLUX.2-klein-4B` | Any resolution |
| Z-Image-Turbo | `Tongyi-MAI/Z-Image-Turbo` | Any resolution |
| GLM-Image | `zai-org/GLM-Image` | Any resolution |
| Qwen Image | `Qwen/Qwen-Image` | Any resolution |
| Qwen Image 2512 | `Qwen/Qwen-Image-2512` | Any resolution |
| Qwen Image Edit | `Qwen/Qwen-Image-Edit` | Any resolution |
## Verified LoRA Examples
@@ -132,21 +134,20 @@ This section lists example LoRAs that have been explicitly tested and verified w
### Verified LoRAs by Base Model
| Base Model | Supported LoRAs |
|:-----------------|:----------------|
| Wan2.2 | `lightx2v/Wan2.2-Distill-Loras`<br>`Cseti/wan2.2-14B-Arcane_Jinx-lora-v1` |
| Wan2.1 | `lightx2v/Wan2.1-Distill-Loras` |
| Z-Image-Turbo | `tarn59/pixel_art_style_lora_z_image_turbo`<br>`wcde/Z-Image-Turbo-DeJPEG-Lora` |
| Qwen-Image | `lightx2v/Qwen-Image-Lightning`<br>`flymy-ai/qwen-image-realism-lora`<br>`prithivMLmods/Qwen-Image-HeadshotX`<br>`starsfriday/Qwen-Image-EVA-LoRA` |
| Qwen-Image-Edit | `ostris/qwen_image_edit_inpainting`<br>`lightx2v/Qwen-Image-Edit-2511-Lightning` |
| Flux | `dvyio/flux-lora-simple-illustration`<br>`XLabs-AI/flux-furry-lora`<br>`XLabs-AI/flux-RealismLora` |
| Base Model | Supported LoRAs |
| :-------------- | :------------------------------------------------------------------------------------------------------------------------------------------------- |
| Wan2.2 | `lightx2v/Wan2.2-Distill-Loras`<br>`Cseti/wan2.2-14B-Arcane_Jinx-lora-v1` |
| Wan2.1 | `lightx2v/Wan2.1-Distill-Loras` |
| Z-Image-Turbo | `tarn59/pixel_art_style_lora_z_image_turbo`<br>`wcde/Z-Image-Turbo-DeJPEG-Lora` |
| Qwen-Image | `lightx2v/Qwen-Image-Lightning`<br>`flymy-ai/qwen-image-realism-lora`<br>`prithivMLmods/Qwen-Image-HeadshotX`<br>`starsfriday/Qwen-Image-EVA-LoRA` |
| Qwen-Image-Edit | `ostris/qwen_image_edit_inpainting`<br>`lightx2v/Qwen-Image-Edit-2511-Lightning` |
| Flux | `dvyio/flux-lora-simple-illustration`<br>`XLabs-AI/flux-furry-lora`<br>`XLabs-AI/flux-RealismLora` |
#### Special Requirements
> [!NOTE]
> Sliding Tile Attention: Currently, only Hopper GPUs (H100s) are supported.
---
# SGLang diffusion CLI Inference
@@ -158,7 +159,6 @@ The SGLang-diffusion CLI provides a quick way to access the inference pipeline f
- A working SGLang diffusion installation and the `sglang` CLI available in `$PATH`.
- Python 3.11+ if you plan to use the OpenAI Python SDK.
## Supported Arguments
### Server Arguments
@@ -173,7 +173,6 @@ The SGLang-diffusion CLI provides a quick way to access the inference pipeline f
- `--ulysses-degree {ULYSSES_DEGREE}`: The degree of DeepSpeed-Ulysses-style SP in USP
- `--ring-degree {RING_DEGREE}`: The degree of ring attention-style SP in USP
### Sampling Parameters
- `--prompt {PROMPT}`: Text description for the video you want to generate
@@ -181,7 +180,6 @@ The SGLang-diffusion CLI provides a quick way to access the inference pipeline f
- `--negative-prompt {PROMPT}`: Negative prompt to guide generation away from certain concepts
- `--seed {SEED}`: Random seed for reproducible generation
#### Image/Video Configuration
- `--height {HEIGHT}`: Height of the generated output
@@ -189,7 +187,6 @@ The SGLang-diffusion CLI provides a quick way to access the inference pipeline f
- `--num-frames {NUM_FRAMES}`: Number of frames to generate
- `--fps {FPS}`: Frames per second for the saved output, if this is a video-generation task
#### Output Options
- `--output-path {PATH}`: Directory to save the generated video
@@ -210,34 +207,31 @@ Example configuration file (config.json):
```json
{
"model_path": "FastVideo/FastHunyuan-diffusers",
"prompt": "A beautiful woman in a red dress walking down a street",
"output_path": "outputs/",
"num_gpus": 2,
"sp_size": 2,
"tp_size": 1,
"num_frames": 45,
"height": 720,
"width": 1280,
"num_inference_steps": 6,
"seed": 1024,
"fps": 24,
"precision": "bf16",
"vae_precision": "fp16",
"vae_tiling": true,
"vae_sp": true,
"vae_config": {
"load_encoder": false,
"load_decoder": true,
"tile_sample_min_height": 256,
"tile_sample_min_width": 256
},
"text_encoder_precisions": [
"fp16",
"fp16"
],
"mask_strategy_file_path": null,
"enable_torch_compile": false
"model_path": "FastVideo/FastHunyuan-diffusers",
"prompt": "A beautiful woman in a red dress walking down a street",
"output_path": "outputs/",
"num_gpus": 2,
"sp_size": 2,
"tp_size": 1,
"num_frames": 45,
"height": 720,
"width": 1280,
"num_inference_steps": 6,
"seed": 1024,
"fps": 24,
"precision": "bf16",
"vae_precision": "fp16",
"vae_tiling": true,
"vae_sp": true,
"vae_config": {
"load_encoder": false,
"load_decoder": true,
"tile_sample_min_height": 256,
"tile_sample_min_width": 256
},
"text_encoder_precisions": ["fp16", "fp16"],
"mask_strategy_file_path": null,
"enable_torch_compile": false
}
```
@@ -272,7 +266,6 @@ mask_strategy_file_path: null
enable_torch_compile: false
```
To see all the options, you can use the `--help` flag:
```bash
@@ -305,7 +298,6 @@ sglang serve "${SERVER_ARGS[@]}"
For detailed API usage, including Image, Video Generation and LoRA management, please refer to the [OpenAI API Documentation](#sglang-diffusion-openai-api).
## Generate
Run a one-off generation task without launching a persistent server.
@@ -346,15 +338,15 @@ SGLang diffusion supports a **diffusers backend** that allows you to run any dif
### Arguments
| Argument | Values | Description |
|----------|--------|-------------|
| `--backend` | `auto` (default), `sglang`, `diffusers` | `auto`: prefer native SGLang, fallback to diffusers. `sglang`: force native (fails if unavailable). `diffusers`: force vanilla diffusers pipeline. |
| Argument | Values | Description |
| ------------------------------- | ----------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `--backend` | `auto` (default), `sglang`, `diffusers` | `auto`: prefer native SGLang, fallback to diffusers. `sglang`: force native (fails if unavailable). `diffusers`: force vanilla diffusers pipeline. |
| `--diffusers-attention-backend` | `flash`, `_flash_3_hub`, `sage`, `xformers`, `native` | Attention backend for diffusers pipelines. See [diffusers attention backends](https://huggingface.co/docs/diffusers/main/en/optimization/attention_backends). |
| `--trust-remote-code` | flag | Required for models with custom pipeline classes (e.g., Ovis). |
| `--vae-tiling` | flag | Enable VAE tiling for large image support (decodes tile-by-tile). |
| `--vae-slicing` | flag | Enable VAE slicing for lower memory usage (decodes slice-by-slice). |
| `--dit-precision` | `fp16`, `bf16`, `fp32` | Precision for the diffusion transformer. |
| `--vae-precision` | `fp16`, `bf16`, `fp32` | Precision for the VAE. |
| `--trust-remote-code` | flag | Required for models with custom pipeline classes (e.g., Ovis). |
| `--vae-tiling` | flag | Enable VAE tiling for large image support (decodes tile-by-tile). |
| `--vae-slicing` | flag | Enable VAE slicing for lower memory usage (decodes slice-by-slice). |
| `--dit-precision` | `fp16`, `bf16`, `fp32` | Precision for the diffusion transformer. |
| `--vae-precision` | `fp16`, `bf16`, `fp32` | Precision for the VAE. |
### Example: Running Ovis-Image-7B
@@ -381,12 +373,12 @@ For pipeline-specific parameters not exposed via CLI, use `diffusers_kwargs` in
```json
{
"model_path": "AIDC-AI/Ovis-Image-7B",
"backend": "diffusers",
"prompt": "A beautiful landscape",
"diffusers_kwargs": {
"cross_attention_kwargs": {"scale": 0.5}
}
"model_path": "AIDC-AI/Ovis-Image-7B",
"backend": "diffusers",
"prompt": "A beautiful landscape",
"diffusers_kwargs": {
"cross_attention_kwargs": { "scale": 0.5 }
}
}
```
@@ -633,7 +625,8 @@ curl -sS -L "http://localhost:30010/v1/videos/<VIDEO_ID>/content" \
The server supports dynamic loading, merging, and unmerging of LoRA adapters.
**Important Notes:**
- Mutual Exclusion: Only one LoRA can be *merged* (active) at a time
- Mutual Exclusion: Only one LoRA can be _merged_ (active) at a time
- Switching: To switch LoRAs, you must first `unmerge` the current one, then `set` the new one
- Caching: The server caches loaded LoRA weights in memory. Switching back to a previously loaded LoRA (same path) has little cost
@@ -644,6 +637,7 @@ Loads one or more LoRA adapters and merges their weights into the model. Support
**Endpoint:** `POST /v1/set_lora`
**Parameters:**
- `lora_nickname` (string or list of strings, required): A unique identifier for the LoRA adapter(s). Can be a single string or a list of strings for multiple LoRAs
- `lora_path` (string or list of strings/None, optional): Path to the `.safetensors` file(s) or Hugging Face repo ID(s). Required for the first load; optional if re-activating a cached nickname. If a list, must match the length of `lora_nickname`
- `target` (string or list of strings, optional): Which transformer(s) to apply the LoRA to. If a list, must match the length of `lora_nickname`. Valid values:
@@ -694,21 +688,22 @@ curl -X POST http://localhost:30010/v1/set_lora \
> [!NOTE]
> When using multiple LoRAs:
>
> - All list parameters (`lora_nickname`, `lora_path`, `target`, `strength`) must have the same length
> - If `target` or `strength` is a single value, it will be applied to all LoRAs
> - Multiple LoRAs applied to the same target will be merged in order
#### Merge LoRA Weights
Manually merges the currently set LoRA weights into the base model.
> [!NOTE]
> `set_lora` automatically performs a merge, so this is typically only needed if you have manually unmerged but want to re-apply the same LoRA without calling `set_lora` again.*
> `set_lora` automatically performs a merge, so this is typically only needed if you have manually unmerged but want to re-apply the same LoRA without calling `set_lora` again.\*
**Endpoint:** `POST /v1/merge_lora_weights`
**Parameters:**
- `target` (string, optional): Which transformer(s) to merge. One of "all" (default), "transformer", "transformer_2", "critic"
- `strength` (float, optional): LoRA strength for merge, default 1.0. Values < 1.0 reduce the effect, values > 1.0 amplify the effect
@@ -720,7 +715,6 @@ curl -X POST http://localhost:30010/v1/merge_lora_weights \
-d '{"strength": 0.8}'
```
#### Unmerge LoRA Weights
Unmerges the currently active LoRA weights from the base model, restoring it to its original state. This **must** be called before setting a different LoRA.
@@ -768,6 +762,7 @@ curl -sS -X GET "http://localhost:30010/v1/list_loras"
```
Notes:
- If LoRA is not enabled for the current pipeline, the server will return an error.
- `num_lora_layers_with_weights` counts only layers that have LoRA weights applied for the active adapter.
@@ -808,16 +803,16 @@ Backend selection is performed by the shared attention layers (e.g. `LocalAttent
The CLI accepts the lowercase names of `AttentionBackendEnum`. The table below lists the backends implemented by the built-in platforms. `fa3`/`fa4` are accepted as aliases for `fa`.
| CLI value | Enum value | Notes |
|---|---|---|
| `fa` / `fa3` / `fa4` | `FA` | FlashAttention. `fa3/fa4` are normalized to `fa` during argument parsing (`ServerArgs.__post_init__`). |
| `torch_sdpa` | `TORCH_SDPA` | PyTorch `scaled_dot_product_attention`. |
| `sliding_tile_attn` | `SLIDING_TILE_ATTN` | Sliding Tile Attention (STA). Requires `st_attn` and a mask-strategy config file set via the `SGLANG_DIFFUSION_ATTENTION_CONFIG` environment variable. |
| `sage_attn` | `SAGE_ATTN` | Requires `sageattention`. Upstream SageAttention CUDA extensions target SM80/SM86/SM89/SM90/SM120 (compute capability 8.0/8.6/8.9/9.0/12.0); see upstream `setup.py`: https://github.com/thu-ml/SageAttention/blob/main/setup.py. |
| `sage_attn_3` | `SAGE_ATTN_3` | Requires SageAttention3 installed per upstream instructions. |
| `video_sparse_attn` | `VIDEO_SPARSE_ATTN` | Requires `vsa`. |
| `vmoba_attn` | `VMOBA_ATTN` | Requires `kernel.attn.vmoba_attn.vmoba`. |
| `aiter` | `AITER` | Requires `aiter`. |
| CLI value | Enum value | Notes |
| -------------------- | ------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `fa` / `fa3` / `fa4` | `FA` | FlashAttention. `fa3/fa4` are normalized to `fa` during argument parsing (`ServerArgs.__post_init__`). |
| `torch_sdpa` | `TORCH_SDPA` | PyTorch `scaled_dot_product_attention`. |
| `sliding_tile_attn` | `SLIDING_TILE_ATTN` | Sliding Tile Attention (STA). Requires `st_attn` and a mask-strategy config file set via the `SGLANG_DIFFUSION_ATTENTION_CONFIG` environment variable. |
| `sage_attn` | `SAGE_ATTN` | Requires `sageattention`. Upstream SageAttention CUDA extensions target SM80/SM86/SM89/SM90/SM120 (compute capability 8.0/8.6/8.9/9.0/12.0); see upstream `setup.py`: https://github.com/thu-ml/SageAttention/blob/main/setup.py. |
| `sage_attn_3` | `SAGE_ATTN_3` | Requires SageAttention3 installed per upstream instructions. |
| `video_sparse_attn` | `VIDEO_SPARSE_ATTN` | Requires `vsa`. |
| `vmoba_attn` | `VMOBA_ATTN` | Requires `kernel.attn.vmoba_attn.vmoba`. |
| `aiter` | `AITER` | Requires `aiter`. |
## Selection priority
@@ -829,16 +824,16 @@ The selection order in `runtime/layers/attention/selector.py` is:
## Platform support matrix
| Backend | CUDA | ROCm | MPS | Notes |
|---|---:|---:|---:|---|
| `fa` | ✅ | ✅ | ❌ | CUDA requires SM80+ and fp16/bf16. FlashAttention is only used when the required runtime is installed; otherwise it falls back to `torch_sdpa`. |
| `torch_sdpa` | ✅ | ✅ | ✅ | Most compatible option across platforms. |
| `sliding_tile_attn` | ✅ | ❌ | ❌ | CUDA-only. Requires `st_attn` and `SGLANG_DIFFUSION_ATTENTION_CONFIG`. |
| `sage_attn` | ✅ | ❌ | ❌ | CUDA-only (optional dependency). |
| `sage_attn_3` | ✅ | ❌ | ❌ | CUDA-only (optional dependency). |
| `video_sparse_attn` | ✅ | ❌ | ❌ | CUDA-only. Requires `vsa`. |
| `vmoba_attn` | ✅ | ❌ | ❌ | CUDA-only. Requires `kernel.attn.vmoba_attn.vmoba`. |
| `aiter` | ✅ | ❌ | ❌ | Requires `aiter`. |
| Backend | CUDA | ROCm | MPS | Notes |
| ------------------- | ---: | ---: | --: | ----------------------------------------------------------------------------------------------------------------------------------------------- |
| `fa` | ✅ | ✅ | ❌ | CUDA requires SM80+ and fp16/bf16. FlashAttention is only used when the required runtime is installed; otherwise it falls back to `torch_sdpa`. |
| `torch_sdpa` | ✅ | ✅ | ✅ | Most compatible option across platforms. |
| `sliding_tile_attn` | ✅ | ❌ | ❌ | CUDA-only. Requires `st_attn` and `SGLANG_DIFFUSION_ATTENTION_CONFIG`. |
| `sage_attn` | ✅ | ❌ | ❌ | CUDA-only (optional dependency). |
| `sage_attn_3` | ✅ | ❌ | ❌ | CUDA-only (optional dependency). |
| `video_sparse_attn` | ✅ | ❌ | ❌ | CUDA-only. Requires `vsa`. |
| `vmoba_attn` | ✅ | ❌ | ❌ | CUDA-only. Requires `kernel.attn.vmoba_attn.vmoba`. |
| `aiter` | ✅ | ❌ | ❌ | Requires `aiter`. |
## Usage
@@ -906,7 +901,7 @@ sglang generate --model-path Qwen/Qwen-Image \
DBCache controls block-level caching behavior:
| Parameter | Env Variable | Default | Description |
|-----------|---------------------------|---------|------------------------------------------|
| --------- | ------------------------- | ------- | ---------------------------------------- |
| Fn | `SGLANG_CACHE_DIT_FN` | 1 | Number of first blocks to always compute |
| Bn | `SGLANG_CACHE_DIT_BN` | 0 | Number of last blocks to always compute |
| W | `SGLANG_CACHE_DIT_WARMUP` | 4 | Warmup steps before caching starts |
@@ -918,7 +913,7 @@ DBCache controls block-level caching behavior:
TaylorSeer improves caching accuracy using Taylor expansion:
| Parameter | Env Variable | Default | Description |
|-----------|-------------------------------|---------|---------------------------------|
| --------- | ----------------------------- | ------- | ------------------------------- |
| Enable | `SGLANG_CACHE_DIT_TAYLORSEER` | false | Enable TaylorSeer calibrator |
| Order | `SGLANG_CACHE_DIT_TS_ORDER` | 1 | Taylor expansion order (1 or 2) |
@@ -950,7 +945,7 @@ which to use cached results.
SCM is configured with presets:
| Preset | Compute Ratio | Speed | Quality |
|----------|---------------|----------|------------|
| -------- | ------------- | -------- | ---------- |
| `none` | 100% | Baseline | Best |
| `slow` | ~75% | ~1.3x | High |
| `medium` | ~50% | ~2x | Good |
@@ -981,7 +976,7 @@ sglang generate --model-path Qwen/Qwen-Image \
#### SCM Policy
| Policy | Env Variable | Description |
|-----------|---------------------------------------|---------------------------------------------|
| --------- | ------------------------------------- | ------------------------------------------- |
| `dynamic` | `SGLANG_CACHE_DIT_SCM_POLICY=dynamic` | Adaptive caching based on content (default) |
| `static` | `SGLANG_CACHE_DIT_SCM_POLICY=static` | Fixed caching pattern |
@@ -990,7 +985,7 @@ sglang generate --model-path Qwen/Qwen-Image \
All Cache-DiT parameters can be set via the following environment variables:
| Environment Variable | Default | Description |
|-------------------------------------|---------|------------------------------------------|
| ----------------------------------- | ------- | ---------------------------------------- |
| `SGLANG_CACHE_DIT_ENABLED` | false | Enable Cache-DiT acceleration |
| `SGLANG_CACHE_DIT_FN` | 1 | First N blocks to always compute |
| `SGLANG_CACHE_DIT_BN` | 0 | Last N blocks to always compute |
@@ -1008,14 +1003,14 @@ All Cache-DiT parameters can be set via the following environment variables:
SGLang Diffusion x Cache-DiT supports almost all models originally supported in SGLang Diffusion:
| Model Family | Example Models |
|--------------|-------------------------------------------|
| Wan | Wan2.1, Wan2.2 |
| Flux | FLUX.1-dev, FLUX.2-dev, FLUX.2-Klein |
| Z-Image | Z-Image-Turbo |
| Qwen | Qwen-Image, Qwen-Image-Edit |
| GLM | GLM-Image |
| Hunyuan | HunyuanVideo |
| Model Family | Example Models |
| ------------ | ------------------------------------ |
| Wan | Wan2.1, Wan2.2 |
| Flux | FLUX.1-dev, FLUX.2-dev, FLUX.2-Klein |
| Z-Image | Z-Image-Turbo |
| Qwen | Qwen-Image, Qwen-Image-Edit |
| GLM | GLM-Image |
| Hunyuan | HunyuanVideo |
## Performance Tips
@@ -1075,6 +1070,7 @@ sglang generate \
```
**Parameters:**
- `--profile`: Enable profiling for the denoising stage
- `--num-profiled-timesteps N`: Number of timesteps to profile after warmup (default: 5)
- Smaller values reduce trace file size
@@ -1094,6 +1090,7 @@ sglang generate \
```
**Parameters:**
- `--profile-all-stages`: Used with `--profile`, profile all pipeline stages instead of just denoising
### Output Location
@@ -1109,15 +1106,16 @@ The exact output file path will be shown in the console output, for example:
### View Traces
Load and visualize trace files at:
- https://ui.perfetto.dev/ (recommended)
- chrome://tracing (Chrome only)
For large trace files, reduce `--num-profiled-timesteps` or avoid using `--profile-all-stages`.
### `--perf-dump-path` (Stage/Step Timing Dump)
Besides profiler traces, you can also dump a lightweight JSON report that contains:
- stage-level timing breakdown for the full pipeline
- step-level timing breakdown for the denoising stage (per diffusion step)
@@ -1177,6 +1175,7 @@ nsys profile \
```
**Parameters:**
- `--delay N`: Wait N seconds before starting capture (skip initialization overhead)
- `--duration N`: Capture for N seconds (focus on specific stages)
- `--force-overwrite`: Overwrite existing output files
@@ -1202,16 +1201,19 @@ This guide outlines the requirements for contributing to the SGLang Diffusion mo
We follow a structured commit message format to maintain a clean history.
**Format:**
```text
[diffusion] <scope>: <subject>
```
**Examples:**
- `[diffusion] cli: add --perf-dump-path argument`
- `[diffusion] scheduler: fix deadlock in batch processing`
- `[diffusion] model: support Stable Diffusion 3.5`
**Rules:**
- **Prefix**: Always start with `[diffusion]`.
- **Scope** (Optional): `cli`, `scheduler`, `model`, `pipeline`, `docs`, etc.
- **Subject**: Imperative mood, short and clear (e.g., "add feature" not "added feature").
@@ -1223,11 +1225,13 @@ For PRs that impact **latency**, **throughput**, or **memory usage**, you **shou
### How to Generate a Report
1. **Baseline**: run the benchmark (for a single generation task)
```bash
$ sglang generate --model-path <model> --prompt "A benchmark prompt" --perf-dump-path baseline.json
```
2. **New**: run the same benchmark, without modifying any server_args or sampling_params
```bash
$ sglang generate --model-path <model> --prompt "A benchmark prompt" --perf-dump-path new.json
```
@@ -1238,7 +1242,7 @@ For PRs that impact **latency**, **throughput**, or **memory usage**, you **shou
### Performance Comparison Report
...
```
4. **Paste**: paste the table into the PR description
4. **Paste**: paste the table into the PR description
## 3. CI-Based Change Protection
@@ -1260,6 +1264,7 @@ SGLang diffusion uses a modular pipeline architecture built around two key conce
- **`PipelineStage`**: Modular components (prompt encoding, denoising loop, VAE decoding, etc.)
To add a new model, you'll need to define:
1. **`PipelineConfig`**: Static model configurations (paths, precision settings)
2. **`SamplingParams`**: Runtime generation parameters (prompt, guidance_scale, steps)
3. **`ComposedPipeline`**: Chain together pipeline stages

9
docs/supported_models/image_generation/index.rst Normal file
View File

@@ -0,0 +1,9 @@
Image Generation
================
Models for generating images and videos using diffusion.
.. toctree::
:maxdepth: 1
diffusion_models.md

14
docs/supported_models/index.rst Normal file
View File

@@ -0,0 +1,14 @@
Supported Models
================
SGLang supports a wide variety of model architectures for different use cases.
Browse by category below to find models suited for your needs.
.. toctree::
:maxdepth: 2
text_generation/index
image_generation/index
retrieval_ranking/index
specialized/index
extending/index

151
docs/supported_models/mindspore_models.md
View File

@@ -1,151 +0,0 @@
# MindSpore Models
## Introduction
MindSpore is a high-performance AI framework optimized for Ascend NPUs. This doc guides users to run MindSpore models in SGLang.
## Requirements
MindSpore currently only supports Ascend NPU devices. Users need to first install CANN 8.5.
The CANN software packages can be downloaded from the [Ascend Official Website](https://www.hiascend.com).
## Supported Models
Currently, the following models are supported:
- **Qwen3**: Dense and MoE models
- **DeepSeek V3/R1**
- *More models coming soon...*
## Installation
> **Note**: Currently, MindSpore models are provided by an independent package `sgl-mindspore`. Support for MindSpore is built upon current SGLang support for Ascend NPU platform. Please first [install SGLang for Ascend NPU](../platforms/ascend_npu.md) and then install `sgl-mindspore`:
```shell
git clone https://github.com/mindspore-lab/sgl-mindspore.git
cd sgl-mindspore
pip install -e .
```
## Run Model
Current SGLang-MindSpore supports Qwen3 and DeepSeek V3/R1 models. This doc uses Qwen3-8B as an example.
### Offline inference
Use the following script for offline inference:
```python
import sglang as sgl
# Initialize the engine with MindSpore backend
llm = sgl.Engine(
model_path="/path/to/your/model", # Local model path
device="npu", # Use NPU device
model_impl="mindspore", # MindSpore implementation
attention_backend="ascend", # Attention backend
tp_size=1, # Tensor parallelism size
dp_size=1 # Data parallelism size
)
# Generate text
prompts = [
"Hello, my name is",
"The capital of France is",
"The future of AI is"
]
sampling_params = {"temperature": 0, "top_p": 0.9}
outputs = llm.generate(prompts, sampling_params)
for prompt, output in zip(prompts, outputs):
print(f"Prompt: {prompt}")
print(f"Generated: {output['text']}")
print("---")
```
### Start server
Launch a server with MindSpore backend:
```bash
# Basic server startup
python3 -m sglang.launch_server \
--model-path /path/to/your/model \
--host 0.0.0.0 \
--device npu \
--model-impl mindspore \
--attention-backend ascend \
--tp-size 1 \
--dp-size 1
```
For distributed server with multiple nodes:
```bash
# Multi-node distributed server
python3 -m sglang.launch_server \
--model-path /path/to/your/model \
--host 0.0.0.0 \
--device npu \
--model-impl mindspore \
--attention-backend ascend \
--dist-init-addr 127.0.0.1:29500 \
--nnodes 2 \
--node-rank 0 \
--tp-size 4 \
--dp-size 2
```
## Troubleshooting
#### Debug Mode
Enable sglang debug logging by log-level argument.
```bash
python3 -m sglang.launch_server \
--model-path /path/to/your/model \
--host 0.0.0.0 \
--device npu \
--model-impl mindspore \
--attention-backend ascend \
--log-level DEBUG
```
Enable mindspore info and debug logging by setting environments.
```bash
export GLOG_v=1 # INFO
export GLOG_v=0 # DEBUG
```
#### Explicitly select devices
Use the following environment variable to explicitly select the devices to use.
```shell
export ASCEND_RT_VISIBLE_DEVICES=4,5,6,7 # to set device
```
#### Some communication environment issues
In case of some environment with special communication environment, users need set some environment variables.
```shell
export MS_ENABLE_LCCL=off # current not support LCCL communication mode in SGLang-MindSpore
```
#### Some dependencies of protobuf
In case of some environment with special protobuf version, users need set some environment variables to avoid binary version mismatch.
```shell
export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python # to avoid protobuf binary version mismatch
```
## Support
For MindSpore-specific issues:
- Refer to the [MindSpore documentation](https://www.mindspore.cn/)

0
docs/supported_models/classify_models.md → docs/supported_models/retrieval_ranking/classify_models.md
View File

0
docs/supported_models/embedding_models.md → docs/supported_models/retrieval_ranking/embedding_models.md
View File

11
docs/supported_models/retrieval_ranking/index.rst Normal file
View File

@@ -0,0 +1,11 @@
Retrieval & Ranking
===================
Models for embeddings, reranking, and classification.
.. toctree::
:maxdepth: 1
embedding_models.md
rerank_models.md
classify_models.md

0
docs/supported_models/rerank_models.md → docs/supported_models/retrieval_ranking/rerank_models.md
View File

9
docs/supported_models/specialized/index.rst Normal file
View File

@@ -0,0 +1,9 @@
Specialized Models
==================
Models for specialized tasks like reward modeling.
.. toctree::
:maxdepth: 1
reward_models.md

56
docs/supported_models/reward_models.md → docs/supported_models/specialized/reward_models.md
View File

@@ -1,28 +1,28 @@
# Reward Models
These models output a scalar reward score or classification result, often used in reinforcement learning or content moderation tasks.
```{important}
They are executed with `--is-embedding` and some may require `--trust-remote-code`.
```
## Example launch Command
```shell
python3 -m sglang.launch_server \
--model-path Qwen/Qwen2.5-Math-RM-72B \ # example HF/local path
--is-embedding \
--host 0.0.0.0 \
--tp-size=4 \ # set for tensor parallelism
--port 30000 \
```
## Supported models
| Model Family (Reward) | Example HuggingFace Identifier | Description |
|---------------------------------------------------------------------------|-----------------------------------------------------|---------------------------------------------------------------------------------|
| **Llama (3.1 Reward / `LlamaForSequenceClassification`)** | `Skywork/Skywork-Reward-Llama-3.1-8B-v0.2` | Reward model (preference classifier) based on Llama 3.1 (8B) for scoring and ranking responses for RLHF. |
| **Gemma 2 (27B Reward / `Gemma2ForSequenceClassification`)** | `Skywork/Skywork-Reward-Gemma-2-27B-v0.2` | Derived from Gemma‑2 (27B), this model provides human preference scoring for RLHF and multilingual tasks. |
| **InternLM 2 (Reward / `InternLM2ForRewardMode`)** | `internlm/internlm2-7b-reward` | InternLM 2 (7B)–based reward model used in alignment pipelines to guide outputs toward preferred behavior. |
| **Qwen2.5 (Reward - Math / `Qwen2ForRewardModel`)** | `Qwen/Qwen2.5-Math-RM-72B` | A 72B math-specialized RLHF reward model from the Qwen2.5 series, tuned for evaluating and refining responses. |
| **Qwen2.5 (Reward - Sequence / `Qwen2ForSequenceClassification`)** | `jason9693/Qwen2.5-1.5B-apeach` | A smaller Qwen2.5 variant used for sequence classification, offering an alternative RLHF scoring mechanism. |
# Reward Models
These models output a scalar reward score or classification result, often used in reinforcement learning or content moderation tasks.
```{important}
They are executed with `--is-embedding` and some may require `--trust-remote-code`.
```
## Example launch Command
```shell
python3 -m sglang.launch_server \
--model-path Qwen/Qwen2.5-Math-RM-72B \ # example HF/local path
--is-embedding \
--host 0.0.0.0 \
--tp-size=4 \ # set for tensor parallelism
--port 30000 \
```
## Supported models
| Model Family (Reward) | Example HuggingFace Identifier | Description |
|---------------------------------------------------------------------------|-----------------------------------------------------|---------------------------------------------------------------------------------|
| **Llama (3.1 Reward / `LlamaForSequenceClassification`)** | `Skywork/Skywork-Reward-Llama-3.1-8B-v0.2` | Reward model (preference classifier) based on Llama 3.1 (8B) for scoring and ranking responses for RLHF. |
| **Gemma 2 (27B Reward / `Gemma2ForSequenceClassification`)** | `Skywork/Skywork-Reward-Gemma-2-27B-v0.2` | Derived from Gemma‑2 (27B), this model provides human preference scoring for RLHF and multilingual tasks. |
| **InternLM 2 (Reward / `InternLM2ForRewardMode`)** | `internlm/internlm2-7b-reward` | InternLM 2 (7B)–based reward model used in alignment pipelines to guide outputs toward preferred behavior. |
| **Qwen2.5 (Reward - Math / `Qwen2ForRewardModel`)** | `Qwen/Qwen2.5-Math-RM-72B` | A 72B math-specialized RLHF reward model from the Qwen2.5 series, tuned for evaluating and refining responses. |
| **Qwen2.5 (Reward - Sequence / `Qwen2ForSequenceClassification`)** | `jason9693/Qwen2.5-1.5B-apeach` | A smaller Qwen2.5 variant used for sequence classification, offering an alternative RLHF scoring mechanism. |

7
docs/supported_models/diffusion_language_models.md → docs/supported_models/text_generation/diffusion_language_models.md
View File

@@ -1,5 +1,7 @@
# Diffusion Language Models
> This page covers **text generation** using diffusion-based LLMs. For **image and video generation**, see [Diffusion Models](../image_generation/diffusion_models.md).
Diffusion language models have shown promise for non-autoregressive text generation with parallel decoding capabilities. Unlike auto-regressive language models, different diffusion language models require different decoding strategies.
## Example Launch Command
@@ -25,6 +27,7 @@ threshold: 0.95
# Default: 32, for LLaDA2MoeModelLM
block_size: 32
```
## Example Client Code Snippet
Just like other supported models, diffusion language models can be used via the REST API or Python client.
@@ -78,6 +81,6 @@ curl -X POST "http://127.0.0.1:30000/generate" \
Below the supported models are summarized in a table.
| Model Family | Example Model | Description |
| ------------------------------------------ | -------------------------------------- | --------------------------------------------------------------------------- |
| Model Family | Example Model | Description |
| -------------------------- | ---------------------------- | ---------------------------------------------------------------------------------------------------- |
| **LLaDA2.0 (mini, flash)** | `inclusionAI/LLaDA2.0-flash` | LLaDA2.0-flash is a diffusion language model featuring a 100B Mixture-of-Experts (MoE) architecture. |

138
docs/supported_models/generative_models.md → docs/supported_models/text_generation/generative_models.md
View File

@@ -1,68 +1,70 @@
# Large Language Models
These models accept text input and produce text output (e.g., chat completions). They are primarily large language models (LLMs), some with mixture-of-experts (MoE) architectures for scaling.
## Example launch Command
```shell
python3 -m sglang.launch_server \
--model-path meta-llama/Llama-3.2-1B-Instruct \ # example HF/local path
--host 0.0.0.0 \
--port 30000 \
```
## Supported models
Below the supported models are summarized in a table.
If you are unsure if a specific architecture is implemented, you can search for it via GitHub. For example, to search for `Qwen3ForCausalLM`, use the expression:
```
repo:sgl-project/sglang path:/^python\/sglang\/srt\/models\// Qwen3ForCausalLM
```
in the GitHub search bar.
| Model Family (Variants) | Example HuggingFace Identifier | Description |
|-------------------------------------|--------------------------------------------------|----------------------------------------------------------------------------------------|
| **DeepSeek** (v1, v2, v3/R1) | `deepseek-ai/DeepSeek-R1` | Series of advanced reasoning-optimized models (including a 671B MoE) trained with reinforcement learning; top performance on complex reasoning, math, and code tasks. [SGLang provides Deepseek v3/R1 model-specific optimizations](../basic_usage/deepseek.md) and [Reasoning Parser](../advanced_features/separate_reasoning.ipynb)|
| **Kimi K2** (Thinking, Instruct) | `moonshotai/Kimi-K2-Instruct` | Moonshot AI's 1 trillion parameter MoE model (32B active) with 128K–256K context; state-of-the-art agentic intelligence with stable long-horizon agency across 200–300 sequential tool calls. Features MLA attention and native INT4 quantization. [See Reasoning Parser docs](../advanced_features/separate_reasoning.ipynb)|
| **Kimi Linear** (48B-A3B) | `moonshotai/Kimi-Linear-48B-A3B-Instruct` | Moonshot AI's hybrid linear attention model (48B total, 3B active) with 1M token context; features Kimi Delta Attention (KDA) for up to 6× faster decoding and 75% KV cache reduction vs full attention. |
| **GPT-OSS** | `openai/gpt-oss-20b`, `openai/gpt-oss-120b` | OpenAI’s latest GPT-OSS series for complex reasoning, agentic tasks, and versatile developer use cases.|
| **Qwen** (3, 3MoE, 3Next, 2.5, 2 series) | `Qwen/Qwen3-0.6B`, `Qwen/Qwen3-30B-A3B` `Qwen/Qwen3-Next-80B-A3B-Instruct ` | Alibaba’s latest Qwen3 series for complex reasoning, language understanding, and generation tasks; Support for MoE variants along with previous generation 2.5, 2, etc. [SGLang provides Qwen3 specific reasoning parser](../advanced_features/separate_reasoning.ipynb)|
| **Llama** (2, 3.x, 4 series) | `meta-llama/Llama-4-Scout-17B-16E-Instruct` | Meta's open LLM series, spanning 7B to 400B parameters (Llama 2, 3, and new Llama 4) with well-recognized performance. [SGLang provides Llama-4 model-specific optimizations](../basic_usage/llama4.md) |
| **Mistral** (Mixtral, NeMo, Small3) | `mistralai/Mistral-7B-Instruct-v0.2` | Open 7B LLM by Mistral AI with strong performance; extended into MoE (“Mixtral”) and NeMo Megatron variants for larger scale. |
| **Gemma** (v1, v2, v3) | `google/gemma-3-1b-it` | Google’s family of efficient multilingual models (1B–27B); Gemma 3 offers a 128K context window, and its larger (4B+) variants support vision input. |
| **Phi** (Phi-1.5, Phi-2, Phi-3, Phi-4, Phi-MoE series) | `microsoft/Phi-4-multimodal-instruct`, `microsoft/Phi-3.5-MoE-instruct` | Microsoft’s Phi family of small models (1.3B–5.6B); Phi-4-multimodal (5.6B) processes text, images, and speech, Phi-4-mini is a high-accuracy text model and Phi-3.5-MoE is a mixture-of-experts model. |
| **MiniCPM** (v3, 4B) | `openbmb/MiniCPM3-4B` | OpenBMB’s series of compact LLMs for edge devices; MiniCPM 3 (4B) achieves GPT-3.5-level results in text tasks. |
| **OLMo** (2, 3) | `allenai/OLMo-3-1125-32B`, `allenai/OLMo-3-32B-Think`, `allenai/OLMo-2-1124-7B-Instruct` | Allen AI’s series of Open Language Models designed to enable the science of language models. |
| **OLMoE** (Open MoE) | `allenai/OLMoE-1B-7B-0924` | Allen AI’s open Mixture-of-Experts model (7B total, 1B active parameters) delivering state-of-the-art results with sparse expert activation. |
| **MiniMax-M2** (M2, M2.1) | `minimax/MiniMax-M2`, `minimax/MiniMax-M2.1` | MiniMax’s SOTA LLM for coding & agentic workflows. |
| **StableLM** (3B, 7B) | `stabilityai/stablelm-tuned-alpha-7b` | StabilityAI’s early open-source LLM (3B & 7B) for general text generation; a demonstration model with basic instruction-following ability. |
| **Command-(R,A)** (Cohere) | `CohereLabs/c4ai-command-r-v01`, `CohereLabs/c4ai-command-r7b-12-2024`, `CohereLabs/c4ai-command-a-03-2025` | Cohere’s open conversational LLM (Command series) optimized for long context, retrieval-augmented generation, and tool use. |
| **DBRX** (Databricks) | `databricks/dbrx-instruct` | Databricks’ 132B-parameter MoE model (36B active) trained on 12T tokens; competes with GPT-3.5 quality as a fully open foundation model. |
| **Grok** (xAI) | `xai-org/grok-1` | xAI’s grok-1 model known for vast size(314B parameters) and high quality; integrated in SGLang for high-performance inference. |
| **ChatGLM** (GLM-130B family) | `THUDM/chatglm2-6b` | Zhipu AI’s bilingual chat model (6B) excelling at Chinese-English dialogue; fine-tuned for conversational quality and alignment. |
| **InternLM 2** (7B, 20B) | `internlm/internlm2-7b` | Next-gen InternLM (7B and 20B) from SenseTime, offering strong reasoning and ultra-long context support (up to 200K tokens). |
| **ExaONE 3** (Korean-English) | `LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct` | LG AI Research’s Korean-English model (7.8B) trained on 8T tokens; provides high-quality bilingual understanding and generation. |
| **Baichuan 2** (7B, 13B) | `baichuan-inc/Baichuan2-13B-Chat` | BaichuanAI’s second-generation Chinese-English LLM (7B/13B) with improved performance and an open commercial license. |
| **XVERSE** (MoE) | `xverse/XVERSE-MoE-A36B` | Yuanxiang’s open MoE LLM (XVERSE-MoE-A36B: 255B total, 36B active) supporting ~40 languages; delivers 100B+ dense-level performance via expert routing. |
| **SmolLM** (135M–1.7B) | `HuggingFaceTB/SmolLM-1.7B` | Hugging Face’s ultra-small LLM series (135M–1.7B params) offering surprisingly strong results, enabling advanced AI on mobile/edge devices. |
| **GLM-4** (Multilingual 9B) | `ZhipuAI/glm-4-9b-chat` | Zhipu’s GLM-4 series (up to 9B parameters) – open multilingual models with support for 1M-token context and even a 5.6B multimodal variant (Phi-4V). |
| **MiMo** (7B series) | `XiaomiMiMo/MiMo-7B-RL` | Xiaomi's reasoning-optimized model series, leverages Multiple-Token Prediction for faster inference. |
| **ERNIE-4.5** (4.5, 4.5MoE series) | `baidu/ERNIE-4.5-21B-A3B-PT` | Baidu's ERNIE-4.5 series which consists of MoE with 47B and 3B active parameters, with the largest model having 424B total parameters, as well as a 0.3B dense model. |
| **Arcee AFM-4.5B** | `arcee-ai/AFM-4.5B-Base` | Arcee's foundational model series for real world reliability and edge deployments. |
| **Persimmon** (8B) | `adept/persimmon-8b-chat` | Adept’s open 8B model with a 16K context window and fast inference; trained for broad usability and licensed under Apache 2.0. |
| **Solar** (10.7B) | `upstage/SOLAR-10.7B-Instruct-v1.0` | Upstage's 10.7B parameter model, optimized for instruction-following tasks. This architecture incorporates a depth-up scaling methodology, enhancing model performance. |
| **Tele FLM** (52B-1T) | `CofeAI/Tele-FLM` | BAAI & TeleAI's multilingual model, available in 52-billion and 1-trillion parameter variants. It is a decoder-only transformer trained on ~2T tokens |
| **Ling** (16.8B–290B) | `inclusionAI/Ling-lite`, `inclusionAI/Ling-plus` | InclusionAI’s open MoE models. Ling-Lite has 16.8B total / 2.75B active parameters, and Ling-Plus has 290B total / 28.8B active parameters. They are designed for high performance on NLP and complex reasoning tasks. |
| **Granite 3.0, 3.1** (IBM) | `ibm-granite/granite-3.1-8b-instruct` | IBM's open dense foundation models optimized for reasoning, code, and business AI use cases. Integrated with Red Hat and watsonx systems. |
| **Granite 3.0 MoE** (IBM) | `ibm-granite/granite-3.0-3b-a800m-instruct` | IBM’s Mixture-of-Experts models offering strong performance with cost-efficiency. MoE expert routing designed for enterprise deployment at scale. |
| **GPT-J** (6B) | `EleutherAI/gpt-j-6b` | EleutherAI's GPT-2-like causal language model (6B) trained on the [Pile](https://pile.eleuther.ai/) dataset. |
| **Orion** (14B) | `OrionStarAI/Orion-14B-Base` | A series of open-source multilingual large language models by OrionStarAI, pretrained on a 2.5T token multilingual corpus including Chinese, English, Japanese, Korean, etc, and it exhibits superior performance in these languages. |
| **Llama Nemotron Super** (v1, v1.5, NVIDIA) | `nvidia/Llama-3_3-Nemotron-Super-49B-v1`, `nvidia/Llama-3_3-Nemotron-Super-49B-v1_5` | The [NVIDIA Nemotron](https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/) family of multimodal models provides state-of-the-art reasoning models specifically designed for enterprise-ready AI agents. |
| **Llama Nemotron Ultra** (v1, NVIDIA) | `nvidia/Llama-3_1-Nemotron-Ultra-253B-v1` | The [NVIDIA Nemotron](https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/) family of multimodal models provides state-of-the-art reasoning models specifically designed for enterprise-ready AI agents. |
| **NVIDIA Nemotron Nano 2.0** | `nvidia/NVIDIA-Nemotron-Nano-9B-v2` | The [NVIDIA Nemotron](https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/) family of multimodal models provides state-of-the-art reasoning models specifically designed for enterprise-ready AI agents. `Nemotron-Nano-9B-v2` is a hybrid Mamba-Transformer language model designed to increase throughput for reasoning workloads while achieving state-of-the-art accuracy compared to similarly-sized models. |
| **StarCoder2** (3B-15B) | `bigcode/starcoder2-7b` | StarCoder2 is a family of open large language models (LLMs) specialized for code generation and understanding. It is the successor to StarCoder, jointly developed by the BigCode project (a collaboration between Hugging Face, ServiceNow Research, and other contributors). |
| **Jet-Nemotron** | `jet-ai/Jet-Nemotron-2B` | Jet-Nemotron is a new family of hybrid-architecture language models that surpass state-of-the-art open-source full-attention language models, while achieving significant efficiency gains. |
| **Trinity** (Nano, Mini) | `arcee-ai/Trinity-Mini` | Arcee's foundational MoE Trinity family of models, open weights under Apache 2.0. |
# Large Language Models
These models accept text input and produce text output (e.g., chat completions). They are primarily large language models (LLMs), some with mixture-of-experts (MoE) architectures for scaling.
## Example launch Command
```shell
python3 -m sglang.launch_server \
--model-path meta-llama/Llama-3.2-1B-Instruct \ # example HF/local path
--host 0.0.0.0 \
--port 30000 \
```
## Supported models
Below the supported models are summarized in a table.
If you are unsure if a specific architecture is implemented, you can search for it via GitHub. For example, to search for `Qwen3ForCausalLM`, use the expression:
```
repo:sgl-project/sglang path:/^python\/sglang\/srt\/models\// Qwen3ForCausalLM
```
in the GitHub search bar.
| Model Family (Variants) | Example HuggingFace Identifier | Description |
|-------------------------------------|--------------------------------------------------|----------------------------------------------------------------------------------------|
| **DeepSeek** (v1, v2, v3/R1) | `deepseek-ai/DeepSeek-R1` | Series of advanced reasoning-optimized models (including a 671B MoE) trained with reinforcement learning; top performance on complex reasoning, math, and code tasks. [SGLang provides Deepseek v3/R1 model-specific optimizations](../basic_usage/deepseek.md) and [Reasoning Parser](../advanced_features/separate_reasoning.ipynb)|
| **Kimi K2** (Thinking, Instruct) | `moonshotai/Kimi-K2-Instruct` | Moonshot AI's 1 trillion parameter MoE model (32B active) with 128K–256K context; state-of-the-art agentic intelligence with stable long-horizon agency across 200–300 sequential tool calls. Features MLA attention and native INT4 quantization. [See Reasoning Parser docs](../advanced_features/separate_reasoning.ipynb)|
| **Kimi Linear** (48B-A3B) | `moonshotai/Kimi-Linear-48B-A3B-Instruct` | Moonshot AI's hybrid linear attention model (48B total, 3B active) with 1M token context; features Kimi Delta Attention (KDA) for up to 6× faster decoding and 75% KV cache reduction vs full attention. |
| **GPT-OSS** | `openai/gpt-oss-20b`, `openai/gpt-oss-120b` | OpenAI’s latest GPT-OSS series for complex reasoning, agentic tasks, and versatile developer use cases.|
| **Qwen** (3, 3MoE, 3Next, 2.5, 2 series) | `Qwen/Qwen3-0.6B`, `Qwen/Qwen3-30B-A3B` `Qwen/Qwen3-Next-80B-A3B-Instruct ` | Alibaba’s latest Qwen3 series for complex reasoning, language understanding, and generation tasks; Support for MoE variants along with previous generation 2.5, 2, etc. [SGLang provides Qwen3 specific reasoning parser](../advanced_features/separate_reasoning.ipynb)|
| **Llama** (2, 3.x, 4 series) | `meta-llama/Llama-4-Scout-17B-16E-Instruct` | Meta's open LLM series, spanning 7B to 400B parameters (Llama 2, 3, and new Llama 4) with well-recognized performance. [SGLang provides Llama-4 model-specific optimizations](../basic_usage/llama4.md) |
| **Mistral** (Mixtral, NeMo, Small3) | `mistralai/Mistral-7B-Instruct-v0.2` | Open 7B LLM by Mistral AI with strong performance; extended into MoE (“Mixtral”) and NeMo Megatron variants for larger scale. |
| **Gemma** (v1, v2, v3) | `google/gemma-3-1b-it` | Google’s family of efficient multilingual models (1B–27B); Gemma 3 offers a 128K context window, and its larger (4B+) variants support vision input. |
| **Phi** (Phi-1.5, Phi-2, Phi-3, Phi-4, Phi-MoE series) | `microsoft/Phi-4-multimodal-instruct`, `microsoft/Phi-3.5-MoE-instruct` | Microsoft’s Phi family of small models (1.3B–5.6B); Phi-4-multimodal (5.6B) processes text, images, and speech, Phi-4-mini is a high-accuracy text model and Phi-3.5-MoE is a mixture-of-experts model. |
| **MiniCPM** (v3, 4B) | `openbmb/MiniCPM3-4B` | OpenBMB’s series of compact LLMs for edge devices; MiniCPM 3 (4B) achieves GPT-3.5-level results in text tasks. |
| **OLMo** (2, 3) | `allenai/OLMo-3-1125-32B`, `allenai/OLMo-3-32B-Think`, `allenai/OLMo-2-1124-7B-Instruct` | Allen AI’s series of Open Language Models designed to enable the science of language models. |
| **OLMoE** (Open MoE) | `allenai/OLMoE-1B-7B-0924` | Allen AI’s open Mixture-of-Experts model (7B total, 1B active parameters) delivering state-of-the-art results with sparse expert activation. |
| **MiniMax-M2** (M2, M2.1) | `minimax/MiniMax-M2`, `minimax/MiniMax-M2.1` | MiniMax’s SOTA LLM for coding & agentic workflows. |
| **StableLM** (3B, 7B) | `stabilityai/stablelm-tuned-alpha-7b` | StabilityAI’s early open-source LLM (3B & 7B) for general text generation; a demonstration model with basic instruction-following ability. |
| **Command-(R,A)** (Cohere) | `CohereLabs/c4ai-command-r-v01`, `CohereLabs/c4ai-command-r7b-12-2024`, `CohereLabs/c4ai-command-a-03-2025` | Cohere’s open conversational LLM (Command series) optimized for long context, retrieval-augmented generation, and tool use. |
| **DBRX** (Databricks) | `databricks/dbrx-instruct` | Databricks’ 132B-parameter MoE model (36B active) trained on 12T tokens; competes with GPT-3.5 quality as a fully open foundation model. |
| **Grok** (xAI) | `xai-org/grok-1` | xAI’s grok-1 model known for vast size(314B parameters) and high quality; integrated in SGLang for high-performance inference. |
| **ChatGLM** (GLM-130B family) | `THUDM/chatglm2-6b` | Zhipu AI’s bilingual chat model (6B) excelling at Chinese-English dialogue; fine-tuned for conversational quality and alignment. |
| **InternLM 2** (7B, 20B) | `internlm/internlm2-7b` | Next-gen InternLM (7B and 20B) from SenseTime, offering strong reasoning and ultra-long context support (up to 200K tokens). |
| **ExaONE 3** (Korean-English) | `LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct` | LG AI Research’s Korean-English model (7.8B) trained on 8T tokens; provides high-quality bilingual understanding and generation. |
| **Baichuan 2** (7B, 13B) | `baichuan-inc/Baichuan2-13B-Chat` | BaichuanAI’s second-generation Chinese-English LLM (7B/13B) with improved performance and an open commercial license. |
| **XVERSE** (MoE) | `xverse/XVERSE-MoE-A36B` | Yuanxiang’s open MoE LLM (XVERSE-MoE-A36B: 255B total, 36B active) supporting ~40 languages; delivers 100B+ dense-level performance via expert routing. |
| **SmolLM** (135M–1.7B) | `HuggingFaceTB/SmolLM-1.7B` | Hugging Face’s ultra-small LLM series (135M–1.7B params) offering surprisingly strong results, enabling advanced AI on mobile/edge devices. |
| **GLM-4** (Multilingual 9B) | `ZhipuAI/glm-4-9b-chat` | Zhipu’s GLM-4 series (up to 9B parameters) – open multilingual models with support for 1M-token context and even a 5.6B multimodal variant (Phi-4V). |
| **MiMo** (7B series) | `XiaomiMiMo/MiMo-7B-RL` | Xiaomi's reasoning-optimized model series, leverages Multiple-Token Prediction for faster inference. |
| **ERNIE-4.5** (4.5, 4.5MoE series) | `baidu/ERNIE-4.5-21B-A3B-PT` | Baidu's ERNIE-4.5 series which consists of MoE with 47B and 3B active parameters, with the largest model having 424B total parameters, as well as a 0.3B dense model. |
| **Arcee AFM-4.5B** | `arcee-ai/AFM-4.5B-Base` | Arcee's foundational model series for real world reliability and edge deployments. |
| **Persimmon** (8B) | `adept/persimmon-8b-chat` | Adept’s open 8B model with a 16K context window and fast inference; trained for broad usability and licensed under Apache 2.0. |
| **Solar** (10.7B) | `upstage/SOLAR-10.7B-Instruct-v1.0` | Upstage's 10.7B parameter model, optimized for instruction-following tasks. This architecture incorporates a depth-up scaling methodology, enhancing model performance. |
| **Tele FLM** (52B-1T) | `CofeAI/Tele-FLM` | BAAI & TeleAI's multilingual model, available in 52-billion and 1-trillion parameter variants. It is a decoder-only transformer trained on ~2T tokens |
| **Ling** (16.8B–290B) | `inclusionAI/Ling-lite`, `inclusionAI/Ling-plus` | InclusionAI’s open MoE models. Ling-Lite has 16.8B total / 2.75B active parameters, and Ling-Plus has 290B total / 28.8B active parameters. They are designed for high performance on NLP and complex reasoning tasks. |
| **Granite 3.0, 3.1** (IBM) | `ibm-granite/granite-3.1-8b-instruct` | IBM's open dense foundation models optimized for reasoning, code, and business AI use cases. Integrated with Red Hat and watsonx systems. |
| **Granite 3.0 MoE** (IBM) | `ibm-granite/granite-3.0-3b-a800m-instruct` | IBM’s Mixture-of-Experts models offering strong performance with cost-efficiency. MoE expert routing designed for enterprise deployment at scale. |
| **GPT-J** (6B) | `EleutherAI/gpt-j-6b` | EleutherAI's GPT-2-like causal language model (6B) trained on the [Pile](https://pile.eleuther.ai/) dataset. |
| **Orion** (14B) | `OrionStarAI/Orion-14B-Base` | A series of open-source multilingual large language models by OrionStarAI, pretrained on a 2.5T token multilingual corpus including Chinese, English, Japanese, Korean, etc, and it exhibits superior performance in these languages. |
| **Llama Nemotron Super** (v1, v1.5, NVIDIA) | `nvidia/Llama-3_3-Nemotron-Super-49B-v1`, `nvidia/Llama-3_3-Nemotron-Super-49B-v1_5` | The [NVIDIA Nemotron](https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/) family of multimodal models provides state-of-the-art reasoning models specifically designed for enterprise-ready AI agents. |
| **Llama Nemotron Ultra** (v1, NVIDIA) | `nvidia/Llama-3_1-Nemotron-Ultra-253B-v1` | The [NVIDIA Nemotron](https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/) family of multimodal models provides state-of-the-art reasoning models specifically designed for enterprise-ready AI agents. |
| **NVIDIA Nemotron Nano 2.0** | `nvidia/NVIDIA-Nemotron-Nano-9B-v2` | The [NVIDIA Nemotron](https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/) family of multimodal models provides state-of-the-art reasoning models specifically designed for enterprise-ready AI agents. `Nemotron-Nano-9B-v2` is a hybrid Mamba-Transformer language model designed to increase throughput for reasoning workloads while achieving state-of-the-art accuracy compared to similarly-sized models. |
| **StarCoder2** (3B-15B) | `bigcode/starcoder2-7b` | StarCoder2 is a family of open large language models (LLMs) specialized for code generation and understanding. It is the successor to StarCoder, jointly developed by the BigCode project (a collaboration between Hugging Face, ServiceNow Research, and other contributors). |
| **Jet-Nemotron** | `jet-ai/Jet-Nemotron-2B` | Jet-Nemotron is a new family of hybrid-architecture language models that surpass state-of-the-art open-source full-attention language models, while achieving significant efficiency gains. |
| **Trinity** (Nano, Mini) | `arcee-ai/Trinity-Mini` | Arcee's foundational MoE Trinity family of models, open weights under Apache 2.0. |
| **Falcon-H1** (0.5B–34B) | `tiiuae/Falcon-H1-34B-Instruct` | TII's hybrid Mamba-Transformer architecture combining attention and state-space models for efficient long-context inference. |
| **Hunyuan-Large** (389B, MoE) | `tencent/Tencent-Hunyuan-Large` | Tencent's open-source MoE model with 389B total / 52B active parameters, featuring Cross-Layer Attention (CLA) for improved efficiency. |

11
docs/supported_models/text_generation/index.rst Normal file
View File

@@ -0,0 +1,11 @@
Text Generation
===============
Models for generating text from text or multimodal inputs.
.. toctree::
:maxdepth: 1
generative_models.md
multimodal_language_models.md
diffusion_language_models.md

2
docs/supported_models/multimodal_language_models.md → docs/supported_models/text_generation/multimodal_language_models.md
View File

@@ -50,6 +50,8 @@ in the GitHub search bar.
| **NVIDIA Nemotron Nano 2.0 VL** | `nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16` | NVIDIA Nemotron Nano v2 VL enables multi-image reasoning and video understanding, along with strong document intelligence, visual Q&A and summarization capabilities. It builds on Nemotron Nano V2, a hybrid Mamba-Transformer LLM, in order to achieve higher inference throughput in long document and video scenarios. | Use `--trust-remote-code`. You may need to adjust `--max-mamba-cache-size` [default is 512] to fit memory constraints. |
| **Ernie4.5-VL** | `baidu/ERNIE-4.5-VL-28B-A3B-PT` | Baidu's vision-language models(28B,424B). Support image and video comprehension, and also support thinking. | |
| **JetVLM** | | JetVLM is an vision-language model designed for high-performance multimodal understanding and generation tasks built upon Jet-Nemotron. | Coming soon |
| **Step3-VL** (10B) | `stepfun-ai/Step3-VL-10B` | StepFun's lightweight open-source 10B parameter VLM for multimodal intelligence, excelling in visual perception, complex reasoning, and human alignment. | |
| **Qwen3-Omni** | `Qwen/Qwen3-Omni-30B-A3B-Instruct` | Alibaba's omni-modal MoE model. Currently supports the **Thinker** component (multimodal understanding for text, images, audio, and video), while the **Talker** component (audio generation) is not yet supported. | |
## Video Input Support