Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
buildbot
3 months ago
|
parent
|
context
|
favorite
| on:
Open models by OpenAI
One thing to note is that MXFP4 is a block scaled format, with 4.25 bits per weight. This lets it represent a lot more numbers than just raw FP4 would with say 1 mantissa and 2 exponent bits.
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search: