Will Mitchell PRO

WillMitchell's activity

• All
• Models
• Datasets
• Spaces
• Buckets
• Papers
• Collections
• Community
• Posts
• Upvotes
• Likes
• Articles

New activity in Jundot/GLM-5.2-oQ4e-mtp 2 months ago

Benchmarks on M3 Ultra 512GB (oMLX v0.5.1): MTP +33% decode, quality vs imatrix GGUF

👍 1

1

#1 opened 2 months ago by

WillMitchell

New activity in mlx-community/GLM-5.2-DQ4plus-q8 2 months ago

config declares num_nextn_predict_layers=1 but MTP draft tensors were dropped (disables speculative decoding)

1

#4 opened 2 months ago by

WillMitchell