Commits

Breadcrumbs

History for

onmain

User selector

All users

Datepicker

All time

Commit History

Commits on Feb 12, 2024

• Detect AMD GPU info via sysfs and block old cards

  Show description for 6d84f07
  dhiltgen

  committed
  6d84f07

  Copy full SHA for 6d84f07

Commits on Jan 29, 2024

• Merge pull request ollama#2246 from dhiltgen/reject_cuda_without_avx

  Show description for 4072b58
  dhiltgen

  authored
  4072b58

  Copy full SHA for 4072b58

Commits on Jan 28, 2024

• Don't disable GPUs on arm without AVX

  Show description for 15562e8
  dhiltgen

  committed
  15562e8

  Copy full SHA for 15562e8

• Harden for zero detected GPUs

  Show description for f07f8b7
  dhiltgen

  committed
  f07f8b7

  Copy full SHA for f07f8b7

Commits on Jan 27, 2024

• Merge pull request ollama#2116 from dhiltgen/cc_50_80

  Show description for e02ecfb
  dhiltgen

  authored
  e02ecfb

  Copy full SHA for e02ecfb

• Update gpu_info_rocm.c

  jaglinux

  authored
  59d8712

  Copy full SHA for 59d8712

Commits on Jan 26, 2024

• Detect lack of AVX and fallback to CPU mode

  Show description for 667a2ba
  dhiltgen

  committed
  667a2ba

  Copy full SHA for 667a2ba

• Merge pull request ollama#2195 from dhiltgen/rocm_real_gpus

  Show description for 30c43c2
  dhiltgen

  authored
  30c43c2

  Copy full SHA for 30c43c2

• Ignore AMD integrated GPUs

  Show description for 9d7b5d6
  dhiltgen

  committed
  9d7b5d6

  Copy full SHA for 9d7b5d6

• Fix crash on cuda ml init failure

  Show description for 5d9c4a5
  dhiltgen

  committed
  5d9c4a5

  Copy full SHA for 5d9c4a5

Commits on Jan 24, 2024

• More logging for gpu management

  Show description for 013fd07
  dhiltgen

  committed
  013fd07

  Copy full SHA for 013fd07

Commits on Jan 23, 2024

• Report more information about GPUs in verbose mode

  Show description for 987c16b
  dhiltgen

  committed
  987c16b

  Copy full SHA for 987c16b

Commits on Jan 20, 2024

• Add compute capability 5.0, 7.5, and 8.0

  dhiltgen

  committed
  a447a08

  Copy full SHA for a447a08

• increase minimum overhead to 1024MiB (ollama#2114)

  jmorganca

  authored
  f32ea81

  Copy full SHA for f32ea81

• Add support for CUDA 5.2 cards

  dhiltgen

  committed
  681a914

  Copy full SHA for 681a914

Commits on Jan 19, 2024

• More WSL paths

  dhiltgen

  committed
  552db98

  Copy full SHA for 552db98

• Fix CPU-only build under Android Termux enviornment.

  Show description for eb76f3e
  lainedfles

  committed
  eb76f3e

  Copy full SHA for eb76f3e

Commits on Jan 18, 2024

• Merge pull request ollama#2056 from dhiltgen/slog

  Show description for abec7f0
  dhiltgen

  authored
  abec7f0

  Copy full SHA for abec7f0

• Mechanical switch from log to slog

  Show description for fedd705
  dhiltgen

  committed
  fedd705

  Copy full SHA for fedd705

Commits on Jan 14, 2024

• Let gpu.go and gen_linux.sh also find CUDA on Arch Linux

  xyproto

  committed
  f4bf1d5

  Copy full SHA for f4bf1d5

Commits on Jan 11, 2024

• Fix up the CPU fallback selection

  Show description for 7427fa1
  dhiltgen

  committed
  7427fa1

  Copy full SHA for 7427fa1

• Merge pull request ollama#1819 from dhiltgen/multi_variant

  Show description for de2fbde
  dhiltgen

  authored
  de2fbde

  Copy full SHA for de2fbde

• Always dynamically load the llm server library

  Show description for 39928a4
  dhiltgen

  committed
  39928a4

  Copy full SHA for 39928a4

• Build multiple CPU variants and pick the best

  Show description for d88c527
  dhiltgen

  committed
  d88c527

  Copy full SHA for d88c527

• fix gpu_test.go Error (same type) uint64->uint32 (ollama#1921)

  fpreiss

  authored
  3bc8b98

  Copy full SHA for 3bc8b98

• Support multiple variants for a given llm lib type

  Show description for 8da7bef
  dhiltgen

  committed
  8da7bef

  Copy full SHA for 8da7bef

• Increase minimum CUDA memory allocation overhead and fix minimum overhead for multi-gpu (ollama#1896)

  Show description for b24e8d1
  jmorganca

  authored
  b24e8d1

  Copy full SHA for b24e8d1

Commits on Jan 10, 2024

• Harden GPU mgmt library lookup

  Show description for 3c49c3a
  dhiltgen

  committed
  3c49c3a

  Copy full SHA for 3c49c3a

Commits on Jan 9, 2024

• calculate overhead based number of gpu devices (ollama#1875)

  jmorganca

  authored
  c336693

  Copy full SHA for c336693

• Set corret CUDA minimum compute capability version

  Show description for 1961a81
  dhiltgen

  committed
  1961a81

  Copy full SHA for 1961a81

• update rough cuda overhead estimate to 15% + 384MiB

  jmorganca

  committed
  6df83e6

  Copy full SHA for 6df83e6

• revert cuda overhead to 20%

  jmorganca

  committed
  6164f37

  Copy full SHA for 6164f37

• add TODO for cuda overhead

  jmorganca

  committed
  6566387

  Copy full SHA for 6566387

• update cuda overhead to 20% to fix crashes when switching between models and large context sizes

  jmorganca

  committed
  3770893

  Copy full SHA for 3770893

• update cuda overhead to 15% or 400MiB

  jmorganca

  committed
  f6cb0a5

  Copy full SHA for f6cb0a5

Pagination

Previous

Next