Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Apr 8, 2026

• Merge branch 'abetlen:main' into main

  adromir

  authored
  4fbfc01

  Copy full SHA for 4fbfc01

• feat: Update llama.cpp to ggerganov/llama.cpp@3bd9aa1f9 (#2176)

  Show description for 1bcc5bc
  abetlen

  authored
  1bcc5bc

  Copy full SHA for 1bcc5bc

• chore: Update .gitignore to exclude project memory files

  

  Adromir

  committed
  b709d5f

  Copy full SHA for b709d5f

Commits on Apr 6, 2026

• fix(ci): Switch Windows to ROCm Clang and Ninja for robust HIP builds

  

  Adromir

  committed
  508a377

  Copy full SHA for 508a377

• fix(ci): Resolve Linux __AMDGCN_WAVEFRONT_SIZE error via repo pinning and compat macros

  

  Adromir

  committed
  b793ac9

  Copy full SHA for b793ac9

• fix(ci): Resolve ROCm build errors for v0.3.20

  Show description for 59ed42f
  

  Adromir

  committed
  59ed42f

  Copy full SHA for 59ed42f

• Merge branch 'abetlen:main' into main

  adromir

  authored
  da0ed5e

  Copy full SHA for da0ed5e

Commits on Apr 3, 2026

• chore: bump version to 0.3.20 (#2171)

  abetlen

  authored
  02d6bee

  Copy full SHA for 02d6bee

• fix(misc): replace deprecated llama.cpp references (#2170)

  Show description for 08e088c
  abetlen

  authored
  08e088c

  Copy full SHA for 08e088c

• feat: Update llama.cpp to ggerganov/llama.cpp@f49e9178767d557a522618b16ce8694f9ddac628 (#2169)

  abetlen

  authored
  100b275

  Copy full SHA for 100b275

Commits on Apr 2, 2026

• fix(ci): Apply learnings from working ROCm reference workflow

  Show description for 5486454
  

  Adromir

  committed
  5486454

  Copy full SHA for 5486454

• fix(ci): Use GGML_HIP=ON instead of GGML_HIPBLAS=ON

  Show description for 6f7990b
  

  Adromir

  committed
  6f7990b

  Copy full SHA for 6f7990b

• fix(ci): Explicit hip_DIR cmake vars + non-fatal ROCm verification

  Show description for 324c444
  

  Adromir

  committed
  324c444

  Copy full SHA for 324c444

• fix(ci): Fix HIP compilation \u2013 add dev packages, compiler flags, diagnostics

  Show description for caacba0
  

  Adromir

  committed
  caacba0

  Copy full SHA for caacba0

• fix(ci): Correct ROCm build verification logic

  Show description for 6663078
  

  Adromir

  committed
  6663078

  Copy full SHA for 6663078

Commits on Apr 1, 2026

• feat(ci): Add RDNA4 GPU targets gfx1200/gfx1201 (RX 9060/9070)

  Show description for 859b5ee
  

  Adromir

  committed
  859b5ee

  Copy full SHA for 859b5ee

• fix(ci): Add amdclang compiler flags and ROCm linkage verification

  Show description for 93beedb
  

  Adromir

  committed
  93beedb

  Copy full SHA for 93beedb

• fix(ci): Use amdgpu-install script for ROCm 7.2.1 on Ubuntu 24.04

  Show description for 1b80b75
  

  Adromir

  committed
  1b80b75

  Copy full SHA for 1b80b75

• ci: Replace all workflows with ROCm-only build & release workflow

  Show description for b2cda07
  

  Adromir

  committed
  b2cda07

  Copy full SHA for b2cda07

Commits on Mar 30, 2026

• feat(server): add model-load chat_template_kwargs (#2168)

  abetlen

  authored
  7257ba9

  Copy full SHA for 7257ba9

Commits on Mar 29, 2026

• ci: publish release wheels as py3-none (#2166)

  Show description for 7613aca
  abetlen

  authored
  7613aca

  Copy full SHA for 7613aca

• fix(ci): publish distinct manylinux and musllinux cpu wheels (#2165)

  Show description for fcd932a
  abetlen

  authored
  fcd932a

  Copy full SHA for fcd932a

Commits on Mar 25, 2026

• Bump version to 0.3.19 (#2162)

  abetlen

  authored
  f54421b

  Copy full SHA for f54421b

• feat: Update llama.cpp to ggerganov/llama.cpp@c0159f9c1f874da15e94f371d136f5920b4b5335 (#2161)

  Show description for c670222
  abetlen

  authored
  c670222

  Copy full SHA for c670222

• fix: handle embedding models without KV memory (#2160)

  Show description for ac59e5a
  abetlen

  authored
  ac59e5a

  Copy full SHA for ac59e5a

• fix(ci): reduce CUDA binary wheel size only including cubins for current arches and one PTX target for forward compatibility (#2158)

  Show description for 5f9c231
  abetlen

  authored
  5f9c231

  Copy full SHA for 5f9c231

Commits on Mar 24, 2026

• chore: bump version (#2157)

  abetlen

  authored
  d6f46a5

  Copy full SHA for d6f46a5

• feat: expose attention_type parameter in Llama.__init__ (#2143)

  Show description for 7b38c31
  

  3 peopleauthored
  7b38c31

  Copy full SHA for 7b38c31

• fix(ci): docker build workflow (#2156)

  Show description for ccc6bc0
  abetlen

  authored
  ccc6bc0

  Copy full SHA for ccc6bc0

• fix(ci): cuda wheel workflow (#2155)

  Show description for 909ebf1
  abetlen

  authored
  909ebf1

  Copy full SHA for 909ebf1

• fix(ci): release wheel workflow (#2154)

  Show description for f0391c5
  abetlen

  authored
  f0391c5

  Copy full SHA for f0391c5

Commits on Mar 23, 2026

• chore: Bump version (#2153)

  abetlen

  authored
  a6b1807

  Copy full SHA for a6b1807

• fix: Qwen 3.5 support (#2152)

  Show description for 11e7a55
  abetlen

  authored
  11e7a55

  Copy full SHA for 11e7a55

• ci: add riscv64 wheel builds to release workflow (#2139)

  Show description for e1f8ac0
  

  gounthar

  and

  abetlen

  authored
  e1f8ac0

  Copy full SHA for e1f8ac0

• feat: Update llama.cpp to ggerganov/llama.cpp@49bfddeca18e62fa3d39114a23e9fcbdf8a22388 (#2151)

  Show description for 18aa31e
  abetlen

  authored
  18aa31e

  Copy full SHA for 18aa31e