SGLang provides several optimizations specifically designed for the DeepSeek model to boost its inference speed. This document outlines current optimizations for DeepSeek. Additionally, the SGLang team is actively developing enhancements for [DeepSeek-V3](https://github.com/sgl-project/sglang/issues/2591).
SGLang provides several optimizations specifically designed for the DeepSeek model to boost its inference speed. This document outlines current optimizations for DeepSeek. Additionally, the SGLang team is actively developing enhancements for [DeepSeekV3](https://github.com/sgl-project/sglang/issues/2591).
SGLang is recognized as one of the top engines for [DeepSeek model inference](https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3).
### Download Weights
If you encounter errors when starting the server, ensure the weights have finished downloading. It's recommended to download them beforehand or restart multiple times until all weights are downloaded. Please refer to [DeepSeek V3]([https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3#installation--launch](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base#61-inference-with-deepseek-infer-demo-example-only)) offical guide to download the weights.
### Launch with One node of 8 H200
Please refer to [the example](https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3#using-docker-recommended). **Note that Deepseek V3 is already in FP8. So we should not run it with any quantization arguments like `--quantization fp8 --kv-cache-dtype fp8_e5m2`.** Also, `--enable-dp-attention` can be useful to improve for Deepseek V3/R1's throughput. Please refer to [Data Parallelism Attention](https://docs.sglang.ai/references/deepseek.html#multi-head-latent-attention-mla-throughput-optimizations) for detail.
### Running examples on Multi-node
-[Serving with two H20*8 nodes](https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3#example-serving-with-two-h208-nodes).
-[Serving with two H200*8 nodes and docker](https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3#example-serving-with-two-h2008-nodes-and-docker).
**Description**: [MLA](https://arxiv.org/pdf/2405.04434) is an innovative attention mechanism introduced by the DeepSeek team, aimed at improving inference efficiency. SGLang has implemented specific optimizations for this, including:
**Description**: [MLA](https://arxiv.org/pdf/2405.04434) is an innovative attention mechanism introduced by the DeepSeek team, aimed at improving inference efficiency. SGLang has implemented specific optimizations for this, including:
-**Weight Absorption**: By applying the associative law of matrix multiplication to reorder computation steps, this method balances computation and memory access and improves efficiency in the decoding phase.
-**Weight Absorption**: By applying the associative law of matrix multiplication to reorder computation steps, this method balances computation and memory access and improves efficiency in the decoding phase.
-**Triton Decoding Kernel Optimization**: In the MLA decoding kernel, there is only one KV head. This optimization reduces memory access to the KV cache by processing multiple query heads within one block, accelerating the decoding process.
-**Triton Decoding Kernel Optimization**: In the MLA decoding kernel, there is only one KV head. This optimization reduces memory access to the KV cache by processing multiple query heads within one block, accelerating the decoding process.
-**FP8 Quantization**: W8A8 FP8 and KV Cache FP8 quantization enables efficient FP8 inference. Additionally, we have implemented Batched Matrix Multiplication (BMM) operator to facilitate FP8 inference in MLA with weight absorption.
-**FP8 Quantization**: W8A8 FP8 and KV Cache FP8 quantization enables efficient FP8 inference. Additionally, we have implemented Batched Matrix Multiplication (BMM) operator to facilitate FP8 inference in MLA with weight absorption.
...
@@ -24,7 +44,7 @@ Overall, with these optimizations, we have achieved up to a 7x acceleration in o
...
@@ -24,7 +44,7 @@ Overall, with these optimizations, we have achieved up to a 7x acceleration in o
**Reference**: Check [Blog](https://lmsys.org/blog/2024-09-04-sglang-v0-3/#deepseek-multi-head-latent-attention-mla-throughput-optimizations) and [Slides](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/lmsys_1st_meetup_deepseek_mla.pdf) for more details.
**Reference**: Check [Blog](https://lmsys.org/blog/2024-09-04-sglang-v0-3/#deepseek-multi-head-latent-attention-mla-throughput-optimizations) and [Slides](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/lmsys_1st_meetup_deepseek_mla.pdf) for more details.
## Data Parallelism Attention
### Data Parallelism Attention
**Description**: This optimization involves data parallelism (DP) for the MLA attention mechanism of DeepSeek Series Models, which allows for a significant reduction in the KV cache size, enabling larger batch sizes. Each DP worker independently handles different types of batches (prefill, decode, idle), which are then synchronized before and after processing through the Mixture-of-Experts (MoE) layer.
**Description**: This optimization involves data parallelism (DP) for the MLA attention mechanism of DeepSeek Series Models, which allows for a significant reduction in the KV cache size, enabling larger batch sizes. Each DP worker independently handles different types of batches (prefill, decode, idle), which are then synchronized before and after processing through the Mixture-of-Experts (MoE) layer.
...
@@ -40,17 +60,18 @@ Overall, with these optimizations, we have achieved up to a 7x acceleration in o
...
@@ -40,17 +60,18 @@ Overall, with these optimizations, we have achieved up to a 7x acceleration in o
**Description**: For users with limited memory on a single node, SGLang supports serving DeepSeek Series Models, including DeepSeek V3, across multiple nodes using tensor parallelism. This approach partitions the model parameters across multiple GPUs or nodes to handle models that are too large for one node's memory.
**Description**: For users with limited memory on a single node, SGLang supports serving DeepSeek Series Models, including DeepSeek V3, across multiple nodes using tensor parallelism. This approach partitions the model parameters across multiple GPUs or nodes to handle models that are too large for one node's memory.
**Usage**: Check [here](https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3#example-serving-with-2-h208) for usage examples.
**Usage**: Check [here](https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3#example-serving-with-2-h208) for usage examples.
## Block-wise FP8
### Block-wise FP8
**Description**: SGLang implements block-wise FP8 quantization with two key optimizations:
**Description**: SGLang implements block-wise FP8 quantization with two key optimizations:
-**Activation**: E4M3 format using per-token-per-128-channel sub-vector scales with online casting.
-**Activation**: E4M3 format using per-token-per-128-channel sub-vector scales with online casting.
-**Weight**: Per-128x128-block quantization for better numerical stability.
-**Weight**: Per-128x128-block quantization for better numerical stability.
**Usage**: turn on by default for DeepSeek V3 models.
**Usage**: turn on by default for DeepSeek V3 models.
