FP16 vs. BF16: Same Size, Different Trade
Why training moved from FP16 to BF16, and why a BF16 model may still run fine in FP16: the same 16 bits split differently, measured on Qwen3-8B.
bf16
1 post
1 post
Why training moved from FP16 to BF16, and why a BF16 model may still run fine in FP16: the same 16 bits split differently, measured on Qwen3-8B.