Hugging Face's logo

HelmholtzAI-FZJ
/
GSQ-F8-D64-G16-V8k

Model card Files Community
uwesis commited on
Commit
be041a3
•
1 Parent(s): 3e2bd6b

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -20,7 +20,7 @@ The group scaling experiment of GSQ:
20
  | | \( 2 $\times$ 32 \) | 0.32 | 220 | 0.05 | 25.42 | 0.76 | 100% | 8,157 |
21
  | | \( 4 $\times$ 16 \) | 0.18 | 226 | 0.03 | 28.02 | 0.08 | 100% | 8,143 |
22
  | | \( 16 $\times$ 4 \) | **0.03** | **233** | **0.004** | **34.61** | **0.91** | **99.98%** | **6,775** |
23
- | **GSQ F16-D16** \( V=256K \) | \( 1 $\times$ 16 \) | 1.42 | 179 | 0.13 | 20.70 | 0.56 | 100% | 254,044 |
24
  | | \( 2 $\times$ 8 \) | 0.82 | 199 | 0.09 | 22.20 | 0.63 | 100% | 257,273 |
25
  | | \( 4 $\times$ 4 \) | 0.74 | 202 | 0.08 | 22.75 | 0.63 | 62.46% | 43,767 |
26
  | | \( 8 $\times$ 2 \) | 0.50 | 211 | 0.06 | 23.62 | 0.66 | 46.83% | 22,181 |
 
20
  | | \( 2 $\times$ 32 \) | 0.32 | 220 | 0.05 | 25.42 | 0.76 | 100% | 8,157 |
21
  | | \( 4 $\times$ 16 \) | 0.18 | 226 | 0.03 | 28.02 | 0.08 | 100% | 8,143 |
22
  | | \( 16 $\times$ 4 \) | **0.03** | **233** | **0.004** | **34.61** | **0.91** | **99.98%** | **6,775** |
23
+ | **GSQ F16-D16** \( V=256K \) | \( 1 $\times$ 16 \) | 1.63 | 179 | 0.13 | 20.70 | 0.56 | 100% | 254,044 |
24
  | | \( 2 $\times$ 8 \) | 0.82 | 199 | 0.09 | 22.20 | 0.63 | 100% | 257,273 |
25
  | | \( 4 $\times$ 4 \) | 0.74 | 202 | 0.08 | 22.75 | 0.63 | 62.46% | 43,767 |
26
  | | \( 8 $\times$ 2 \) | 0.50 | 211 | 0.06 | 23.62 | 0.66 | 46.83% | 22,181 |