The @appliedcompute announcement calls Spork an efficient post-training recipe and reports a 100% rise in its chess-puzzle solve rate.
@appliedcompute introduced Spork as an efficient post-training recipe. It says Spork increased its solve rate on chess puzzles by 100%.
6.1K
7 Sources, first seen ago
—
Not ranked yet
—
Not ranked yet
@appliedcomputeIntroducing Spork, an efficient post-training recipe which increases our solve-rate on chess puzzles by 100% when compared to our best GRPO setup. It uses a GRPO warmup, splits task-specific datasets further into partitions, trains specialists, and then uses multi-teacher on-policy distillation. Like a Spork, it leverages gains by simply consolidating specialists.
@lindensliRT @appliedcompute: Introducing Spork, an efficient post-training recipe which increases our solve-rate on chess puzzles by 100% when compa…
@ypatil125Crazy cool how this works. Shows you that there are a lot more interesting and effective ways to train than just vanilla GRPO.@appliedcomputeIntroducing Spork, an efficient post-training recipe which increases our solve-rate on chess puzzles by 100% when compared to our best GRPO setup. It uses a GRPO warmup, splits task-specific datasets further into partitions, trains specialists, and then uses multi-teacher on-policy distillation. Like a Spork, it leverages gains by simply consolidating specialists.@lindensliRT @appliedcompute: Introducing Spork, an efficient post-training recipe which increases our solve-rate on chess puzzles by 100% when compa…@ypatil125Crazy cool how this works. Shows you that there are a lot more interesting and effective ways to train than just vanilla GRPO.@appliedcomputeIntroducing Spork, an efficient post-training recipe which increases our solve-rate on chess puzzles by 100% when compared to our best GRPO setup. It uses a GRPO warmup, splits task-specific datasets further into partitions, trains specialists, and then uses multi-teacher on-policy distillation. Like a Spork, it leverages gains by simply consolidating specialists.@lindensliRT @appliedcompute: Introducing Spork, an efficient post-training recipe which increases our solve-rate on chess puzzles by 100% when compa…@ypatil125Crazy cool how this works. Shows you that there are a lot more interesting and effective ways to train than just vanilla GRPO.@appliedcomputeIntroducing Spork, an efficient post-training recipe which increases our solve-rate on chess puzzles by 100% when compared to our best GRPO setup. It uses a GRPO warmup, splits task-specific datasets further into partitions, trains specialists, and then uses multi-teacher on-policy distillation. Like a Spork, it leverages gains by simply consolidating specialists.@lindensliRT @appliedcompute: Introducing Spork, an efficient post-training recipe which increases our solve-rate on chess puzzles by 100% when compa…@ypatil125Crazy cool how this works. Shows you that there are a lot more interesting and effective ways to train than just vanilla GRPO.
