@@ -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%):<br>
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)
