Skip to content

llama-tornado script curation #112

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
mikepapadim merged 2 commits into from
May 4, 2026
Merged

llama-tornado script curation #112

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4dba662
Drop redundant option and restructure argument groups to separate Adv…
orionpapadakis May 4, 2026
e0414a3
Simplify `prefill-decode` argument help text and improve naming consi…
orionpapadakis May 4, 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
27 changes: 9 additions & 18 deletions llama-tornado
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -499,17 +499,13 @@ def create_parser() -> argparse.ArgumentParser:
help="Execute the command after showing it (use with --show-command)",
)

# Prefill/Decode optimization
prefill_group = parser.add_argument_group("Prefill/Decode Optimization")
# Prefill-Decode optimizations
prefill_group = parser.add_argument_group("Prefill-Decode Optimizations")
prefill_group.add_argument(
"--with-prefill-decode",
dest="with_prefill_decode",
action="store_true",
help=(
"Enable prefill/decode separation. "
"Alone: sequential prefill (skip logits) + standard decode. "
"With --batch-prefill-size N (N>1): batched GPU prefill via TornadoVMMasterPlanWithBatchPrefillDecode."
),
help="Enable single-token prefill decode",
)
prefill_group.add_argument(
"--batch-prefill-size",
Expand All @@ -518,22 +514,17 @@ def create_parser() -> argparse.ArgumentParser:
default=None,
metavar="N",
help=(
"Prefill chunk size (requires --with-prefill-decode). "
"N=1: sequential prefill (same as --with-prefill-decode alone). "
"N>1: batched prefill processing N tokens per chunk (llama.prefillBatchSize=N)."
"Enable batching in prefill when --with-prefill-decode is active and N>1. "
),
)
prefill_group.add_argument(

# Advanced CUDA features
advanced_cuda_features_group = parser.add_argument_group("Advanced CUDA Features")
advanced_cuda_features_group.add_argument(
"--cuda-graphs",
dest="cuda_graphs",
action="store_true",
help="Enable CUDA graph capture/replay (llama.cudaGraphs=true); PTX backend only",
)
prefill_group.add_argument(
"--no-cuda-graphs",
dest="no_cuda_graphs",
action="store_true",
help="Disable CUDA graph capture/replay (llama.cudaGraphs=false); no-op, disabled by default",
help="Enable CUDA graph capture/replay (llama.cudaGraphs=true); PTX backend only.",
)

# Advanced options
Expand Down
Loading