What's Changed
π Features
- Honor registry mirrors for model pulls and backend downloads (#931) @doringeman
- Add vLLM ROCm Docker variant for AMD GPUs (#917) @dmedovich
- Add MUSA and OpenVINO support using upstream llama.cpp images (#897)
- Run daemon and client from a single binary (#900) @areebahmeddd
- Use official llama.cpp images for Linux (#879)
- Add ContextSize path from FromDirectory (#895) @sathira10
- Extract context size (#866) @sathira10
π Bug Fixes
- Use rm -f to handle missing .py files in llamacpp install (#932)
- Fix gosec g602 error for make validate-all (#911) @areebahmeddd
- Use path.Join instead of filepath.Join for tar entry names (#898)
- Increase timeout for E2E tests to improve stability (#891)
- Update release workflow to prioritize GitHub Release tags (#890)
- Optimize latest release tag retrieval in release.yml (#896) @IgnasiBarrera
- Cap max_tokens in E2E chat helpers to prevent test timeouts (#893)
- Add path sanitization to prevent directory traversal in downloader (#889) @IgnasiBarrera
- Pin Python to 3.12 in vLLM and SGLang build stages (#883)
π¦ Dependencies
- Bump llama.cpp to b9209 (#919)
- Bump llama.cpp to b9174 (#918) @krissetto
- Bump llama.cpp to b9102 (#908)
- Bump llama.cpp to b9014 (#901)
- Bump llama.cpp to b8943 (#885)
- Bump docker/cagent-action from 1.5.1 to 1.5.3 (#920)
- Bump docker/cagent-action in the github-actions group (#909)
- Bump openssl (#904)
- Bump the go-modules-root group with 4 updates (#886)
- Bump rustls-webpki (#888)
- Bump the github-actions group with 2 updates (#887)
- Bump openssl (#882)
π§ Maintenance
- Remove fragile bump-pinata and update-docs jobs from release workflow (#933) @IgnasiBarrera
- Install vLLM ROCm from upstream wheels (#923) @dmedovich
- Update vllm install to use --torch-backend auto (#921)
- Pin golangci-lint version in .versions and read it in CI (#913) @doringeman
- Declare contents: read on ci/e2e-test/integration-test workflows (#914) @arpitjain799
- Bump model-runner Go dependency in pinata during release (#881) @IgnasiBarrera
Full Changelog: v1.1.38...v1.2.1