llama.cpp
ggml : implement REGLU/GEGLU/SWIGLU ops
#14158
Merged

Commits
  • implement unary REGLU/GEGLU/SWIGLU cpu ops
    qnixsynapse committed 354 days ago
  • relax constraints
    qnixsynapse committed 354 days ago
  • duplicate shape of source
    qnixsynapse committed 354 days ago
  • fix ggml_vec_geglu_f16
    qnixsynapse committed 354 days ago
  • special case gated ops
    qnixsynapse committed 354 days ago
  • implement unary REGLU/GEGLU/SWIGLU cuda ops
    qnixsynapse committed 354 days ago
  • tighten constraints again
    qnixsynapse committed 354 days ago
  • refactor into GGML_GLU_OP
    qnixsynapse committed 354 days ago
  • metal : add glu kernels
    qnixsynapse committed 354 days ago
  • add CUDA_GLU_BLOCK_SIZE [no ci]
    qnixsynapse committed 354 days ago
  • more constraints and use 64bit ints
    qnixsynapse committed 354 days ago
  • 64bit multiplication [no ci]
    qnixsynapse committed 354 days ago
  • implement swapped variants (cpu/cuda)
    qnixsynapse committed 354 days ago
  • update comment [no ci]
    qnixsynapse committed 354 days ago
  • Vulkan: Add GLU ops and shaders
    qnixsynapse committed 354 days ago
  • SYCL: Implement fused kernel GEGLU, SWIGLU and REGLU for single up+gate
    qnixsynapse committed 354 days ago
  • ggml : implement GLU for split up/gate (#14181)
    qnixsynapse committed 354 days ago
  • GGML: increase OP count in assertion
    qnixsynapse committed 354 days ago
  • Refactor: Optimize SYCL element-wise operations with unary function inlining
    qnixsynapse committed 354 days ago
  • vulkan: Increase workgroup size for GLU, for performance (#14345)
    jeffbolznv committed 348 days ago
  • merge fix
    CISC committed 348 days ago
  • Merge branch 'master' into cisc/unary-reglu-geglu-swiglu
    CISC committed 348 days ago
  • metal : add support for split and swap
    ggerganov committed 347 days ago
Loading