Bypassing Transformer Softmax via Static Contraction
A JAX XLA-powered PoC that leverages branchless mathematical primitives to bypass the memory and execution bottlenecks of LLM softmax operations.
Open source
Reactions (1)
Similar products
by description similarity
UltraCompress
github.comfirst mathematically lossless 5-bit LLM compression
X
github.comA programming language with switchable memory modes
B-IR
github.comAn LLM-optimized programming language
A compiler and IDE for the Luon programming language - a liaison beween Lua and Oberon+ - targeting LuaJIT
Compiler Backend for LSTS (High-Level Typed Macro Assembler)