Skip to content

perf: refine context heuristics #256

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
mplatzer merged 2 commits into from
Dec 12, 2025
Merged

perf: refine context heuristics #256

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fd6d7b0
perf: refine context heuristics
mplatzer Dec 12, 2025
40539f6
fix
mplatzer Dec 12, 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
7 changes: 5 additions & 2 deletions mostlyai/engine/_tabular/argn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -164,10 +164,11 @@ def _regressor_heuristic(id: str, model_size: ModelSizeOrUnits, dim_input: int,
def _flat_context_heuristic(id: str, model_size: ModelSizeOrUnits, dim_input: int) -> list[int]:
if isinstance(model_size, dict):
return model_size[id]
model_size_layers = dict(S=[8], M=[64], L=[128])
model_size_layers = dict(S=[4], M=[16], L=[64])
layers = model_size_layers[model_size]
coefficient = round(np.log(max(dim_input, np.e)))
dims = [unit * coefficient for unit in layers]
_LOG.info(f"[ARGN] flat context heuristic: {dim_input=} -> {dims}")
return dims


Expand All @@ -176,10 +177,11 @@ def _sequential_context_heuristic(
) -> list[int]:
if isinstance(model_size, dict):
return model_size[id]
model_size_layers = dict(S=[8], M=[32], L=[64, 64])
model_size_layers = dict(S=[4], M=[16], L=[64, 64])
layers = model_size_layers[model_size]
coefficient = round(np.log(max(dim_input * seq_len_median, np.e)))
dims = [unit * coefficient for unit in layers]
_LOG.info(f"[ARGN] sequential context heuristic: {dim_input=} x {seq_len_median=} -> {dims}")
return dims


Expand All @@ -190,6 +192,7 @@ def _history_heuristic(id: str, model_size: ModelSizeOrUnits, dim_input: int, se
layers = model_size_layers[model_size]
coefficient = round(np.log(max(dim_input * seq_len_median, np.e)))
dims = [unit * coefficient for unit in layers]
_LOG.info(f"[ARGN] history heuristic: {dim_input=} x {seq_len_median=} -> {dims}")
return dims


Expand Down