hletrd commited on
Commit
c36d810
·
verified ·
1 Parent(s): 7b78772

Card: add Tencent's original description

Browse files
Files changed (1) hide show
  1. README.md +0 -2
README.md CHANGED
@@ -36,8 +36,6 @@ tags:
36
 
37
  [tencent/Hy4-flash](https://huggingface.co/tencent/Hy4-flash) (220B parameters, 12B active) converted for [MLX](https://github.com/ml-explore/mlx) on Apple silicon. The Engram tables sit in four separate files that are memory-mapped from disk, so only the 94 GB of text weights have to fit in memory.
38
 
39
- > This build has not been run on Apple silicon yet. This note comes out after the M5 test.
40
-
41
  - **Files:** the text weights in 4-bit and 8-bit, the Engram tables, the DFly2 drafter, the vision tower, and the `hy4_mlx` loader code.
42
  - **Speculative decoding:** the DFly2 drafter is implemented, and the chat command uses it by default.
43
  - **Run it:** see [How to run](#how-to-run).
 
36
 
37
  [tencent/Hy4-flash](https://huggingface.co/tencent/Hy4-flash) (220B parameters, 12B active) converted for [MLX](https://github.com/ml-explore/mlx) on Apple silicon. The Engram tables sit in four separate files that are memory-mapped from disk, so only the 94 GB of text weights have to fit in memory.
38
 
 
 
39
  - **Files:** the text weights in 4-bit and 8-bit, the Engram tables, the DFly2 drafter, the vision tower, and the `hy4_mlx` loader code.
40
  - **Speculative decoding:** the DFly2 drafter is implemented, and the chat command uses it by default.
41
  - **Run it:** see [How to run](#how-to-run).