Code

Introduce bfloat16 support#6412

Code

Merged

ggerganov merged 8 commits intoggerganov:masterggerganov/llama.cpp:masterfrom jart:bf16jart/llama.cpp:bf16Copy head branch name to clipboardMay 8, 2024

+1,228-102

Commits

Commits on May 8, 2024

• Introduce bfloat16 support

  Show description for 55e962a
  jart

  committed
  55e962a

  Copy full SHA for 55e962a

• Remove GGML code that's not needed

  jart

  committed
  823d45a

  Copy full SHA for 823d45a

• Minimize the GGML API surface area for BF16

  jart

  committed
  180bfcd

  Copy full SHA for 180bfcd

• Remove bf16 luts

  jart

  committed
  d6892c4

  Copy full SHA for d6892c4

• Make the GGML header look nicer

  jart

  committed
  ce0442d

  Copy full SHA for ce0442d

• Fix documentation

  jart

  committed
  bc278c8

  Copy full SHA for bc278c8

• Apply ggerganov's fixes for test-backend-ops

  jart

  committed
  2741a99

  Copy full SHA for 2741a99

• Add BF16 code for new ggml_validate_row_data() function

  jart

  committed
  632624e

  Copy full SHA for 632624e