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.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
docker publish: drop the build caches (#10959)
Daniel Handff8cb32761f
diff --git a/.github/workflows/docker-publish-rocm.yml b/.github/workflows/docker-publish-rocm.ymlindex 883f0600ddd..1aaea50a733 100644--- a/.github/workflows/docker-publish-rocm.yml+++ b/.github/workflows/docker-publish-rocm.yml@@ -251,8 +251,13 @@ jobs:file: ./docker/Dockerfile.rocmplatforms: linux/amd64labels: ${{ steps.meta.outputs.labels }}- cache-from: type=gha,scope=build-rocm-amd64- cache-to: type=gha,scope=build-rocm-amd64,mode=max+ # No cache-from / cache-to. mode=max on the GHA backend would export a+ # 21.4GB image's intermediate layers into the repo-wide Actions cache,+ # which already holds 40GB across 45 entries serving the notebooks pip+ # and GGUF fixtures, and would evict them by LRU. This job also runs+ # weekly while GHA drops entries unused for 7 days, so the cache would+ # usually be cold on arrival. Measured cold: 8 min of layer work plus+ # 5.6 min exporting layers.outputs: type=image,name=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }},push-by-digest=true,name-canonical=true,push=truebuild-args: |ROCM_VERSION=${{ needs.prepare.outputs.rocm_version }}diff --git a/.github/workflows/docker-publish.yml b/.github/workflows/docker-publish.ymlindex dab5a487b6a..61f0dde1152 100644--- a/.github/workflows/docker-publish.yml+++ b/.github/workflows/docker-publish.yml@@ -265,14 +265,12 @@ jobs:file: ./docker/Dockerfileplatforms: ${{ matrix.platform }}labels: ${{ steps.meta.outputs.labels }}- # Registry cache, not type=gha. mode=max exports every intermediate layer,- # which on the GHA backend put 127 blob entries and 32GB into a 50GB repo- # cache in a single morning and starved the GGUF and pip caches sharing it.- # A cache tag on the image's own repo has no such budget, so mode=max keeps- # its rebuild speed without the eviction pressure. The Docker Hub login- # above covers both reading and writing this ref.- cache-from: type=registry,ref=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:buildcache-core-${{ matrix.platform == 'linux/amd64' && 'amd64' || 'arm64' }}- cache-to: type=registry,ref=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:buildcache-core-${{ matrix.platform == 'linux/amd64' && 'amd64' || 'arm64' }},mode=max+ # No cache-from / cache-to. The buildcache-core-* refs this used to write+ # held 21.29GB (amd64) and 20.26GB (arm64) on the image's own Docker Hub+ # repo, and prepare freezes UNSLOTH_REF / UNSLOTH_ZOO_REF to SHAs that move+ # most nights, so the install layer they were meant to save is invalidated+ # anyway. What is left is the base and apt layers, against a build whose+ # wall time is dominated by exporting and pushing 27GB.outputs: type=image,name=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }},push-by-digest=true,name-canonical=true,push=true# Keep prose OUT of build-args: build-push-action forwards every non-empty# line verbatim, so a #-line becomes a bogus --build-arg.@@ -459,12 +457,9 @@ jobs:file: ./docker/Dockerfile.studioplatforms: ${{ matrix.platform }}labels: ${{ steps.meta.outputs.labels }}- # Kept at mode=min. The original reason was the GHA cache quota, which no- # longer applies now that this is a registry ref on the image's own repo,- # but the remaining cost is Docker Hub storage rather than a hard budget,- # so raising it is a separate decision from moving the transport.- cache-from: type=registry,ref=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:buildcache-studio-${{ matrix.platform == 'linux/amd64' && 'amd64' || 'arm64' }}- cache-to: type=registry,ref=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:buildcache-studio-${{ matrix.platform == 'linux/amd64' && 'amd64' || 'arm64' }},mode=min+ # No cache-from / cache-to, for the same reason as the core build above.+ # The buildcache-studio-* refs held 13.90GB (amd64) and 13.48GB (arm64),+ # and at mode=min they only carried the final layers in the first place.outputs: type=image,name=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }},push-by-digest=true,name-canonical=true,push=true# the same values the base build baked, so Studio matches it even if upstream# moved mid-run (build-args must be KEY=VALUE only)