Skip to content
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

Jump to bottom

fix(export): GPT models w/ bias=False convert properly #11255

Merged
merged 2 commits into from
Nov 12, 2024
Merged

fix(export): GPT models w/ bias=False convert properly #11255

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
80ff72c
fix(export): GPT models w/ bias=False convert properly
terrykong Nov 12, 2024
58a0068
Merge branch 'main' into terryk/export-gpt-bias-false
terrykong Nov 12, 2024
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
8 changes: 8 additions & 0 deletions nemo/export/trt_llm/tensorrt_llm_build.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -118,6 +118,14 @@ def build_and_save_engine(
build_config.lora_config = lora_config

model = model_cls.from_config(model_config)
if not model_config.bias and model_config.architecture == 'GPTForCausalLM':
# NOTE: GPT models in megatron-core that set bias=False sets the bias false globally
# whereas bias=False in TRTLLM GPT models sets it false everywhere except
# LayerNorm. This change makes TRTLLM's implementation match megatron-core.
for name, module in model.named_modules():
if isinstance(module, tensorrt_llm.layers.normalization.LayerNorm):
module.bias = None
module.register_parameter('bias', None)
model = optimize_model(
model,
use_parallel_embedding=model_config.use_parallel_embedding,
Expand Down
Loading