mavis-ai commited on
Commit
de631d7
·
verified ·
1 Parent(s): 31bcaf3

Upload Multilingual-e5-large-Q8

Browse files
Files changed (2) hide show
  1. NOTICE +1 -1
  2. README.md +4 -4
NOTICE CHANGED
@@ -22,7 +22,7 @@ Embedding dimension:
22
 
23
  Modification notice:
24
 
25
- This repository contains a R.E.V.I.S. q8 quantized distribution.
26
  Selected 2D weight tensors were converted to symmetric per-row int8
27
  representation with per-row scale tensors.
28
 
 
22
 
23
  Modification notice:
24
 
25
+ This repository contains a R.E.V.I.S. Q8 quantized distribution.
26
  Selected 2D weight tensors were converted to symmetric per-row int8
27
  representation with per-row scale tensors.
28
 
README.md CHANGED
@@ -15,7 +15,7 @@ tags:
15
  - revis
16
  ---
17
 
18
- # mavis-ai/multilingual-e5-large-q8
19
 
20
  This repository contains an 8-bit quantized MLX-compatible distribution of `intfloat/multilingual-e5-large`, prepared for use with **R.E.V.I.S.** as its local semantic embedding model.
21
 
@@ -35,7 +35,7 @@ For the original model card, training details, intended usage, and evaluation in
35
 
36
  ## Quantization
37
 
38
- This package stores selected 2D weight tensors using a R.E.V.I.S. q8 format:
39
 
40
  - Quantization type: symmetric per-row int8
41
  - Scale format: per-row scale tensor
@@ -110,11 +110,11 @@ Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, Furu Wei
110
  Multilingual E5 Text Embeddings: A Technical Report
111
  ```
112
 
113
- R.E.V.I.S. q8 package:
114
 
115
  ```text
116
  Prepared and redistributed by MAVIS / R.E.V.I.S.
117
- Quantization: symmetric per-row int8 q8 package for local MLX embedding runtime
118
  ```
119
 
120
  ## Modification Notice
 
15
  - revis
16
  ---
17
 
18
+ # mavis-ai/Multilingual-e5-large-Q8
19
 
20
  This repository contains an 8-bit quantized MLX-compatible distribution of `intfloat/multilingual-e5-large`, prepared for use with **R.E.V.I.S.** as its local semantic embedding model.
21
 
 
35
 
36
  ## Quantization
37
 
38
+ This package stores selected 2D weight tensors using a R.E.V.I.S. Q8 format:
39
 
40
  - Quantization type: symmetric per-row int8
41
  - Scale format: per-row scale tensor
 
110
  Multilingual E5 Text Embeddings: A Technical Report
111
  ```
112
 
113
+ R.E.V.I.S. Q8 package:
114
 
115
  ```text
116
  Prepared and redistributed by MAVIS / R.E.V.I.S.
117
+ Quantization: symmetric per-row int8 Q8 package for local MLX embedding runtime
118
  ```
119
 
120
  ## Modification Notice