Top-$nσ$: Not All Logits Are You Need
Paper
•
2411.07641
•
Published
•
19
•
4
Tang Chenxia
tomorrowdawn
·
AI & ML interests
None yet
Recent Activity
upvoted
a
paper
about 1 month ago
Low-Bit Quantization Favors Undertrained LLMs: Scaling Laws for
Quantized LLMs with 100T Training Tokens
upvoted
a
paper
about 2 months ago
Top-nσ: Not All Logits Are You Need
commented
a paper
about 2 months ago
Top-$nσ$: Not All Logits Are You Need
Organizations
tomorrowdawn's activity
commented
2
papers
about 2 months ago