Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
coder543
11 months ago
|
parent
|
context
|
favorite
| on:
An analysis of DeepSeek's R1-Zero and R1
Sure, but that default batch size would only matter if the person in question was actually generating and measuring parallel requests, not just measuring the straight line performance of sequential requests... and I have no confidence they were.
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search: