ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

[pre-commit.ci] auto fixes from pre-commit.com hooks

Browse source 
for more information, see https://pre-commit.ci
This commit is contained in:
pre-commit-ci[bot] 2026-06-29 08:06:01 +00:00
commit a3c9c60168
1 changed files with 3 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

4
studio/backend/core/inference/llama_cpp.py
View file

@ -5765,7 +5765,9 @@ class LlamaCppBackend:
# mtp_engaged alone is a no-op once budget_frac is set;
# pass the byte-accurate overhead so MTP KV is reserved.
mtp_engaged = _mtp_will_engage_cpu,
mtp_overhead_fn = (_mtp_bytes if _mtp_will_engage_cpu else None),
mtp_overhead_fn = (
_mtp_bytes if _mtp_will_engage_cpu else None
),
budget_frac = _CPU_RAM_BUDGET_FRAC,
)
_cpu_cap = max(4096, min(_ctx_ceiling, _fit))