Skip to content

Fix llama tokenizer padding_side when using model.generate in inference mode #3644

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 2 commits into from
Nov 26, 2025
+15 −5
Merged

Fix llama tokenizer padding_side when using model.generate in inference mode #3644

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
078e5a7
Only restore training mode after generation, if the model started out…
dmsuehir Nov 25, 2025
8fd2c13
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Nov 26, 2025
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: 5 additions & 1 deletion unsloth/models/llama.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1988,6 +1988,9 @@ def unsloth_fast_generate(
*args,
**kwargs,
):
# If the model starts out in training mode, restore training mode after generation
restore_training_mode = self.training

FastLlamaModel.for_inference(self)

dtype = _get_dtype(dtype_from_config(self.config))
Expand Down Expand Up @@ -2043,7 +2046,8 @@ def unsloth_fast_generate(
# accelerate.utils.operations.send_to_device = accelerate_old_send_to_device
# pass

FastLlamaModel.for_training(self)
if restore_training_mode:
FastLlamaModel.for_training(self)

return output

Expand Down
14 changes: 10 additions & 4 deletions unsloth/models/loader.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -258,7 +258,7 @@ def from_pretrained(
if model_name.lower().endswith("-bf16"):
load_in_4bit = False
load_in_8bit = False
load_in_fp8 = False
load_in_fp8 = False
load_in_16bit = True

if USE_MODELSCOPE and not os.path.exists(model_name):
Expand Down Expand Up @@ -387,7 +387,7 @@ def from_pretrained(
if model_name.lower().endswith("-bf16"):
load_in_4bit = False
load_in_8bit = False
load_in_fp8 = False
load_in_fp8 = False
load_in_16bit = True

model_config = AutoConfig.from_pretrained(
Expand Down Expand Up @@ -711,10 +711,16 @@ def from_pretrained(
)
load_in_4bit = False
load_in_8bit = False
load_in_fp8 = False
load_in_fp8 = False
load_in_16bit = False

if int(load_in_4bit) + int(load_in_8bit) + int(load_in_16bit) + int(load_in_fp8 != False) >= 2:
if (
int(load_in_4bit)
+ int(load_in_8bit)
+ int(load_in_16bit)
+ int(load_in_fp8 != False)
>= 2
):
raise RuntimeError(
"Unsloth: Can only load in 4bit or 8bit or 16bit, not a combination!\n"
"Also, we by default set `load_in_4bit = True`.\n"
Expand Down