EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[Revision] Replace enable_flashinfer_mla argument with attention_backend (#5052)

Browse Source
This commit is contained in:
Baizhou Zhang
2025-04-05 01:23:02 -07:00
committed by GitHub
parent ca8d02abd5
commit efbae697b3
9 changed files with 92 additions and 82 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
docs/backend/server_arguments.md
View File

@@ -138,7 +138,7 @@ Please consult the documentation below to learn more about the parameters you ma
## Kernel backend
* `attention_backend`: The backend for attention computation and KV cache management.
* `attention_backend`: This argument specifies the backend for attention computation and KV cache management, which can be `fa3`, `flashinfer`, `triton`, or `torch_native`. When deploying DeepSeek models, use this argument to specify the MLA backend.
* `sampling_backend`: The backend for sampling.
## Constrained Decoding
@@ -192,5 +192,5 @@ Please consult the documentation below to learn more about the parameters you ma
* `cuda_graph_bs`: The batch sizes to capture by `CudaGraphRunner`. By default this is done for you.
* `torchao_config`: Experimental feature that optimizes the model with [torchao](https://github.com/pytorch/ao). Possible choices are: int8dq, int8wo, int4wo-<group_size>, fp8wo, fp8dq-per_tensor, fp8dq-per_row.
* `triton_attention_num_kv_splits`: Use to adjust the number of KV splits in triton kernels. Default is 8.
* `enable_flashinfer_mla`: Use the attention backend with flashinfer MLA wrapper for deepseek models. When providing this argument, `attention_backend` argument is overridden.
* `flashinfer_mla_disable_ragged`: Disable usage of ragged prefill wrapper for flashinfer mla attention backend. Should be used when `enable_flashinfer_mla` is turned on.
* `enable_flashinfer_mla`: Use the attention backend with FlashInfer MLA wrapper for DeepSeek models. **This argument will be deprecated in the next release. Please use `--attention_backend flashinfer` instead to enable FlashfIner MLA.**
* `flashinfer_mla_disable_ragged`: Disable the use of the ragged prefill wrapper for the FlashInfer MLA attention backend. Only use it when FlashInfer is being used as the MLA backend.