minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Config] Retire get_global_server_args, and clear the deprecated flags that have a replacement (#38375)

Browse Source
This commit is contained in:
Cheng Wan
2026-09-08 16:42:12 -07:00
committed by GitHub
parent 5177a3ec08
commit db272201a2
213 changed files with 625 additions and 892 deletions
Download Patch File Download Diff File Expand all files Collapse all files
benchmark/bench_linear_attention/bench_int8_checkpoint_reuse.py
+2 -2
View File
@@ -20,7 +20,7 @@ Method, per K in ``--num-prefixes``:
Run the server twice and compare (same flags, toggle int8):
python -m sglang.launch_server --model-path <gdn-or-kda-hybrid> --tp 4 \
--trust-remote-code --mamba-scheduler-strategy extra_buffer \
--trust-remote-code --mamba-radix-cache-strategy extra_buffer \
--max-mamba-cache-size 256 [--enable-int8-mamba-checkpoint] --port 30000
python benchmark/bench_linear_attention/bench_int8_checkpoint_reuse.py \
@@ -28,7 +28,7 @@ Run the server twice and compare (same flags, toggle int8):
NOTE: prefix-tokens must cross the mamba cache chunk granularity (typically ~512),
otherwise nothing is cacheable and reuse is 0 by construction (not a regression).
Use ``--mamba-scheduler-strategy extra_buffer`` on the server: ``no_buffer`` only
Use ``--mamba-radix-cache-strategy extra_buffer`` on the server: ``no_buffer`` only
snapshots state at the full-sequence leaf, so a divergent-suffix probe never reuses.
"""