Remix.run Logo
mh- 21 hours ago

To my understanding, with batched inference and other "optimizations" you wouldn't get the exact same token predictions even with temp=0.0.