Skip to content

[https://nvbugs/5521949][fix] Fix head_size handling in ModelConfig.get_bindings_model_config #8100

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

Closed
wants to merge 1 commit into from
Closed

[https://nvbugs/5521949][fix] Fix head_size handling in ModelConfig.get_bindings_model_config #8100

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
3 changes: 2 additions & 1 deletion tensorrt_llm/_torch/model_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -524,12 +524,13 @@ def get_bindings_model_config(self,
)

# For kv cache size calculation: set size_per_head
head_size = None
head_dim_names = ["head_size", "head_dim"]
for head_dim_name in head_dim_names:
if head_dim_name in self.pretrained_config:
head_size = getattr(self.pretrained_config, head_dim_name)
break
else:
if head_size is None:
logger.warning(
f"head_size/head_dim is not set, using default value {hidden_size // num_heads}"
)
Expand Down
11 changes: 7 additions & 4 deletions tests/unittest/llmapi/test_llm_pytorch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -557,7 +557,6 @@ def test_codellama_fp8_with_bf16_lora() -> None:


@skip_gpu_memory_less_than_80gb
@pytest.mark.skip(reason="https://nvbugs/5521949")
def test_bielik_11b_v2_2_instruct_multi_lora() -> None:
model_dir = f"{llm_models_root()}/Bielik-11B-v2.2-Instruct"

Expand All @@ -584,12 +583,16 @@ def test_bielik_11b_v2_2_instruct_multi_lora() -> None:
lora_model.save_pretrained(lora_path)
lora_paths.append(lora_path)

trtllm_lora_config = LoraConfig(lora_dir=lora_paths,
lora_target_modules=target_modules,
trtllm_lora_config = LoraConfig(lora_target_modules=target_modules,
max_lora_rank=8,
max_loras=2,
max_cpu_loras=2)
llm = LLM(model_dir, lora_config=trtllm_lora_config)
llm = LLM(
model_dir,
lora_config=trtllm_lora_config,
# Disable CUDA graph
# TODO: remove this once we have a proper fix for CUDA graph in LoRA
cuda_graph_config=None)

prompts = [
"Kim był Mikołaj Kopernik i z czego zasłynął?",
Expand Down
Loading