Y
Hacker News
new
|
ask
|
show
|
jobs
by
iamnotagenius
311 days ago
Not quite true. Depends on number of KV heads. GLM4 32b at IQ4 quant and Q8 context can run full context with only 20GiB VRAM.