AkaiNa

3 ms·

1. 1.▲
   Show HN: NeuralScript – A pure-Rust AOT compiler(github.com)

   5 points by AkaiNa 6mo ago|0 comments
2. 2.▲
   by AkaiNa 8mo ago

   You can contact me on discord at brandon3183 or use the email registered with this account. Its less meant for data center scalability and more so meant for the everyday person who cant afford h100s and other data center scale gpu/npu
3. 3.▲
   by AkaiNa 8mo ago

   Yes. But standard block floating point uses a linear grid scaled by a shared exponent. Whereas AXS-6 uses a NormalFloat grid scaled by a shared exponent to maximize information density for bell-curve distributed weights. Essentially a Block
4. 4.▲
   AXS-6: Adaptive EXponent Sharing – 6-Bit Training Format(github.com)

   2 points by AkaiNa 8mo ago|5 comments