Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

CI: hold the unpinned version-compat lanes below torch 2.15 until torchvision 0.30 is published (#15774)

Daniel Han6a7c2c6bb048
diff --git a/.github/workflows/version-compat-ci.yml b/.github/workflows/version-compat-ci.yml
index 101ba3294cb..4d0774c0a78 100644
--- a/.github/workflows/version-compat-ci.yml
+++ b/.github/workflows/version-compat-ci.yml
@@ -140,8 +140,9 @@ jobs:
# 3.11, not 3.12: lanes sharing an interpreter share one pip cache key.
# Both pins must equal pyproject's ceilings (test_a_ceiling_lane_pins_the_declared_ceiling).
python: '3.11'
+ # torch 2.15 reached the CPU index before torchvision 0.30, so the pair failed the import guard.
torch_pins: >-
- torch torchvision 'torchcodec>=0.12'
+ 'torch<2.15' torchvision 'torchcodec>=0.12'
pkg_pins: >-
'transformers==5.17.0' 'trl==1.15.0'
peft sentence-transformers accelerate 'datasets>=3.4.1,!=4.0.*,!=4.1.0,!=4.4.*,!=4.5.0,<5.0.0'
@@ -149,8 +150,9 @@ jobs:
- slug: latest
continue_on_error: true
python: '3.12'
+ # torch 2.15 reached the CPU index before torchvision 0.30, so the pair failed the import guard.
torch_pins: >-
- torch torchvision 'torchcodec>=0.12'
+ 'torch<2.15' torchvision 'torchcodec>=0.12'
pkg_pins: >-
transformers trl peft sentence-transformers
accelerate datasets bitsandbytes triton hf_transfer
@@ -259,8 +261,9 @@ jobs:
trl_pin: 'trl==0.18.2'
- slug: latest
python: '3.12'
+ # torch 2.15 reached the CPU index before torchvision 0.30, so the pair failed the import guard.
torch_pins: >-
- torch torchvision 'torchcodec>=0.12'
+ 'torch<2.15' torchvision 'torchcodec>=0.12'
pkg_pins: >-
transformers peft accelerate datasets
bitsandbytes triton hf_transfer