Thinking Machines 4-bit replication: I trained a GPT end to end with a 4-bit forward pass on one B200
Thinking Machines wrote last week that Inkling 2 Small was trained with a native 4-bit forward pass. I wanted to check this for myself.
Thinking Machines wrote last week that Inkling 2 Small was trained with a native 4-bit forward pass. I wanted to check this for myself.