n0p/mlx-video
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Refactor weight loading and sanitization processes for audio models

Browse Source
This commit is contained in:
Prince Canuma
2026-01-23 17:31:25 +01:00
parent 2681f75d2f
commit 02bfa228d9
18 changed files with 510 additions and 498 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
mlx_video/convert.py
View File

@@ -355,6 +355,9 @@ def sanitize_audio_vae_weights(weights: Dict[str, mx.array]) -> Dict[str, mx.arr
"""
sanitized = {}
if "audio_vae." in weights:
return weights
for key, value in weights.items():
new_key = key
@@ -364,9 +367,9 @@ def sanitize_audio_vae_weights(weights: Dict[str, mx.array]) -> Dict[str, mx.arr
elif key.startswith("audio_vae.per_channel_statistics."):
# Map per-channel statistics
if "mean-of-means" in key:
new_key = "per_channel_statistics._mean_of_means"
new_key = "per_channel_statistics.mean_of_means"
elif "std-of-means" in key:
new_key = "per_channel_statistics._std_of_means"
new_key = "per_channel_statistics.std_of_means"
else:
continue # Skip other statistics keys
else: