2.0.5

This is the last release before 3.0 and versioning changes

What's Changed

• LLM/VLM: Add API/generation support for seed, frequency_penalty, and presence_penalty by @droans in #76
• Fix: Return generation kwargs by @droans in #77
• Update Dockerfile to use libze-intel-gpu1 instead of level-zero for GPU dependencies. by @droans in #71
• perf(qwen3-tts): use LATENCY hint for OV compilations; fix streaming defaults to Qwen upstream by @Conradzz in #78
• support Qwen3vl by @SearchSavior in #80
• docs updates by @SearchSavior in #82
• • remove stupid documentation page by @SearchSavior in #83
• Docs (serve): Windows 127.0.0.1 requirement by @savvadesogle in #81
• make api keys optional by @SearchSavior in #85
• 2.0.5 support vlm bench by @SearchSavior in #86
• 2.0.5 docs update by @SearchSavior in #88
• Fix: Transformers v5 changed apply_chat_template output to BatchEncoding by @droans in #95
• Openarc studio endpoints and refactor of the endpoints by @mrelmida in #91
• Kokoro: weighted voicepack blending via voice_blend by @mistyuk in #99
• Add Qwen3.5 support by @droans in #103
• make gpu_metrics a soft dependency and refactor the openarc endpoints by @mrelmida in #100
• Avoid rewriting config file when nothing has changed by @solidDoWant in #115
• Get config file path from OPENARC_CONFIG_FILE if set by @solidDoWant in #114
• Make file logging path configurable by @solidDoWant in #119
• Allow loading models from paths relative to config file by @solidDoWant in #116
• fix: Extended wait time in Docker startup cmd for server readiness before autoloading model by @telnetdoogie in #112
• precompile python and dependencies in Docker image to improve cold start times by @telnetdoogie in #121
• Fix bugs with relative model path resolution by @solidDoWant in #125
• Fix transcribe Qwen3-ASR models blocking other API calls by @solidDoWant in #127
• Fix transcription endpoint requiring openarc_asr parameter when using Qwen3 ASR models by @solidDoWant in #117
• Download fix by @mrelmida in #122
• chore: Refactor some tests by @telnetdoogie in #129
• Implement minor Qwen3-ASR performance improvements by @solidDoWant in #128
• Support OpenAI-standard language key for Qwen3-ASR models by @solidDoWant in #131
• deprecate vlm-type argument from ModelLoadConfig by @SearchSavior in #123
• Fix wheel builds not containing any src/ files by @solidDoWant in #113
• Fix openarc serve start failing due to wrong dep version by @solidDoWant in #120
• Publish contianer image to GHCR by @solidDoWant in #124
• Fix: Don't use incomplete tokens by @droans in #136
• Enable OpenVINO model caching by @solidDoWant in #118
• Add support for Qwen3-ASR segments in verbose and diarize responses by @solidDoWant in #130
• Consolidate Dockerfiles into a single multi-target Dockerfile by @telnetdoogie in #138
• docs: Updated Docker documentation and compose example for deployed images by @telnetdoogie in #140
• Refactored tests to load all models from same ENV path, and additional tweaks by @telnetdoogie in #141
• Logging update by @mrelmida in #101
• Add model / test for Qwen3 ASR and example for more granular test outputs by @telnetdoogie in #142
• Add Unit Test run as a GH Action / PR check by @telnetdoogie in #145
• Add /readyz endponit for model readiness health checks by @solidDoWant in #143
• hotfix: patch qwen35/qwen36 to auto apply enable_reasoning by @SearchSavior in #148
• cleanup and refactor by @SearchSavior in #151
• fix issue with voice blending by @SearchSavior in #152
• docs: note OpenVINO nightly requirement for Qwen3 MoE models on GPU by @duncanmcqueen in #155
• Add scheduler config support by @droans in #156
• Add Qwen XML tool-call parser by @duncanmcqueen in #158
• Support vllm style --tool-call-parser for hermes, qwen35, gemma4 by @SearchSavior in #162
• support muse glimmer tool/reasoning parser by @SearchSavior in #163
• tool parser docs update by @SearchSavior in #165

New Contributors

• @Conradzz made their first contribution in #78
• @savvadesogle made their first contribution in #81
• @mrelmida made their first contribution in #91
• @mistyuk made their first contribution in #99
• @solidDoWant made their first contribution in #115
• @telnetdoogie made their first contribution in #112
• @duncanmcqueen made their first contribution in #155

Full Changelog: v2.0.4...v2.0.5