Skip to content

Hardcode the conditioning in the mlx weights. #203

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

Draft
LaurentMazare wants to merge 2 commits into
base: main
Choose a base branch
from
Draft

Hardcode the conditioning in the mlx weights. #203

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
40a180b
Hardcode the conditioning in the mlx weights.
LaurentMazare Feb 4, 2025
f221467
Do not include the conditioning.
LaurentMazare Feb 4, 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
10 changes: 7 additions & 3 deletions scripts/import_mlx.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,9 +40,6 @@ def import_model(
model = {}
for name in ["text_emb.weight", "text_linear.weight"]:
model[name] = tch_model[name]
for name in tch_model.keys():
if name.startswith("condition_provider.conditioners"):
model[name] = tch_model[name]
model["out_norm.weight"] = tch_model["out_norm.alpha"][0, 0]
for idx in range(in_n_q):
src_name = f"emb.{idx}.weight"
Expand Down Expand Up @@ -99,6 +96,13 @@ def import_model(
model[layer + "gating.linear_out.weight"] = tch_model[
f"depformer.layers.{layer_idx}.gating.{idx}.linear_out.weight"
]
if "condition_provider.conditioners.description.embed.weight" in tch_model:
e = tch_model["condition_provider.conditioners.description.embed.weight"]
w = tch_model["condition_provider.conditioners.description.output_proj.weight"]
# 4 is very_good
e = e[4:5] @ w.T
print(f"adding the very_good conditioning {e.shape} to {model['text_emb.weight'].shape}")
model["text_emb.weight"] += e

save_file(model, out_path)

Expand Down
Loading