Skip to content

Allow fp8 for non fast inference #3904

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
danielhanchen merged 5 commits into from
Feb 25, 2026
+8 −17
Merged

Allow fp8 for non fast inference #3904

Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
4b4c01a
Allow fp8 for non fast inference
Datta0 Jan 19, 2026
5ffa547
Extensive fp8 alow and quantizer patch
Datta0 Jan 19, 2026
a71f9d9
Merge branch 'main' into fp8_enable
Datta0 Feb 16, 2026
ed1dc56
Merge remote-tracking branch 'upstream/main' into fp8_enable
danielhanchen Feb 25, 2026
83082ac
Clean up commented-out code, duplicate import, and revert unnecessary…
danielhanchen Feb 25, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 0 additions & 6 deletions unsloth/models/loader.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -248,12 +248,6 @@ def from_pretrained(
fast_inference = False
break

# [TODO] For now fast_inference only works with fast_inference ie vLLM
if load_in_fp8 != False:
if not fast_inference:
raise NotImplementedError(
"Unsloth: set `fast_inference = True` when doing `load_in_fp8`."
)
# Check if 4bit is allowed specifically for AMD
if not ALLOW_BITSANDBYTES and not use_exact_model_name:
if load_in_4bit or load_in_8bit or model_name.lower().endswith("-bnb-4bit"):
Expand Down