diff --git a/ollama/docker-compose.yml b/ollama/docker-compose.yml index a4d25017bc..4b74f47e6b 100644 --- a/ollama/docker-compose.yml +++ b/ollama/docker-compose.yml @@ -8,7 +8,7 @@ services: PROXY_AUTH_ADD: "false" ollama: - image: ollama/ollama:0.17.4@sha256:b165fa2700dc374f8d4f9e8314d81c7be75487c76eee2b46ef4b511a496b736c + image: ollama/ollama:0.17.5@sha256:719122581b6932e1240ae70d788859089cb80d17e23cd4f98ba960b0290f70cb environment: OLLAMA_ORIGINS: "*" OLLAMA_CONTEXT_LENGTH: 8192 diff --git a/ollama/umbrel-app.yml b/ollama/umbrel-app.yml index b018446f38..2d1a553316 100644 --- a/ollama/umbrel-app.yml +++ b/ollama/umbrel-app.yml @@ -3,7 +3,7 @@ id: ollama name: Ollama tagline: Self-host open source AI models like DeepSeek-R1, Llama, and more category: ai -version: "0.17.4" +version: "0.17.5" port: 11434 description: >- Ollama allows you to download and run advanced AI models directly on your own hardware. Self-hosting AI models ensures full control over your data and protects your privacy. @@ -38,16 +38,15 @@ defaultUsername: "" defaultPassword: "" dependencies: [] releaseNotes: >- - This release adds new models and improvements to tool call handling. + This release fixes crashes and memory issues in Qwen 3.5 models. Key highlights in this release: - - New Qwen 3.5 multimodal model family is now available - - New LFM2 hybrid model family optimized for on-device deployment is now available - - Tool call indices are now included in parallel tool calls - - Fixed tool calls in Qwen 3 and Qwen 3.5 not being parsed correctly during thinking - - Added Nemotron architecture support - - Web search capabilities added for models that support tools + - Fixed crash in Qwen 3.5 models when split over GPU and CPU + - Fixed Qwen 3.5 models repeating themselves due to missing presence penalty (you may need to re-download affected models, e.g. `ollama pull qwen3.5:35b`) + - Fixed memory issues and crashes in the MLX runner + - Fixed inability to run models imported from Qwen 3.5 GGUF files + - `ollama run --verbose` now shows peak memory usage when using the MLX engine Full release notes are available at https://github.com/ollama/ollama/releases