diff --git a/doc/en/DeepseekR1_V3_tutorial.md b/doc/en/DeepseekR1_V3_tutorial.md index fb26b77..9c031dc 100644 --- a/doc/en/DeepseekR1_V3_tutorial.md +++ b/doc/en/DeepseekR1_V3_tutorial.md @@ -86,7 +86,7 @@ Memory: standard DDR5-4800 server DRAM (1 TB), each socket with 8×DDR5-4800 #### Change Log - Longer Context (from 4K to 8K for 24GB VRAM) and Slightly Faster Speed （+15%):
Integrated the highly efficient Triton MLA Kernel from the fantastic sglang project, enable much longer context length and slightly faster prefill/decode speed -- We suspect that some of the improvements come from the change of hardwre platform (4090D->4090) +- We suspect that some of the improvements come from the change of hardware platform (4090D->4090) #### Benchmark Results