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

add chunked context/prefill runtime option to trtllm-serve #2731

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
+16 −5
Open

add chunked context/prefill runtime option to trtllm-serve #2731

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
21 changes: 16 additions & 5 deletions tensorrt_llm/commands/serve.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,10 @@
default="localhost",
help="Hostname of the server.")
@click.option("--port", type=int, default=8000, help="Port of the server.")
@click.option("--backend",
type=click.Choice(["pytorch"]),
default=None,
help="Set to 'pytorch' for pytorch path. Default is cpp path.")
@click.option("--max_beam_width",
type=int,
default=BuildConfig.max_beam_width,
Expand All @@ -35,6 +39,11 @@
help=
"Maximum number of batched input tokens after padding is removed in each batch."
)
@click.option("--chunked_context",
is_flag=True,
default=False,
help="Enable chunked context (chunked prefill)."
)
@click.option(
"--max_seq_len",
type=int,
Expand All @@ -55,10 +64,11 @@
is_flag=True,
default=False,
help="Flag for HF transformers.")
def main(model: str, tokenizer: str, host: str, port: int, max_beam_width: int,
max_batch_size: int, max_num_tokens: int, max_seq_len: int,
tp_size: int, pp_size: int, kv_cache_free_gpu_memory_fraction: float,
trust_remote_code: bool):
def main(model: str, tokenizer: str, host: str, port: int, backend: str,
max_beam_width: int, max_batch_size: int, max_num_tokens: int,
max_seq_len: int, tp_size: int, pp_size: int,
kv_cache_free_gpu_memory_fraction: float,
trust_remote_code: bool, chunked_context: bool):
"""Running an OpenAI API compatible server

MODEL: model name | HF checkpoint path | TensorRT engine path
Expand All @@ -79,7 +89,8 @@ def main(model: str, tokenizer: str, host: str, port: int, max_beam_width: int,
trust_remote_code=trust_remote_code,
build_config=build_config,
kv_cache_config=kv_cache_config,
)
enable_chunked_prefill=chunked_context,
backend=backend if backend == "pytorch" else None)

llm = LLM(**llm_args.to_dict())

Expand Down