3 ms·
The same idea is in both the R1 and S1 papers (<think> tokens are used similarly). Basically they're using special tokens to mark in the prompt where the LLM sh
by jebarker 2y ago
The same idea is in both the R1 and S1 papers (<think> tokens are used similarly). Basically they're using special tokens to mark in the prompt where the LLM should think more/revise the previous response. This can be repeated many times until some stop criteria occurs. S1 manually inserts these with heuristics, R1 learns the placement through RL I think.
- whimsicalism 2y ago? theyre not special tokens really
- jebarker 2y agoi'm not actually sure whether they're special tokens in the sense of being in the vocabulary
- whimsicalism 2y ago<think> might be i think "wait" is tokenized like any other in the pretraining