diff options
| author | Felix Morgner <felix.morgner@gmail.com> | 2026-10-09 14:54:06 +0200 |
|---|---|---|
| committer | Felix Morgner <felix.morgner@gmail.com> | 2026-10-09 14:54:21 +0200 |
| commit | 9622cfbae5b5d004a5901c81dd48ec04ee0fdea6 (patch) | |
| tree | ce9e8c8447fadd2eed2f4ef65b51aaf62864636f | |
| parent | d814715d028f45b048f833ed8e2c5ae83dafe0f5 (diff) | |
| download | local-ai-9622cfbae5b5d004a5901c81dd48ec04ee0fdea6.tar.xz local-ai-9622cfbae5b5d004a5901c81dd48ec04ee0fdea6.zip | |
chore: pre-refactor commit
| -rw-r--r-- | .gitignore | 3 | ||||
| -rw-r--r-- | comfyui/.gitkeep (renamed from inference/.gitkeep) | 0 | ||||
| -rw-r--r-- | docker-compose.yml | 47 | ||||
| -rw-r--r-- | inference/modelfiles/qwen2.5-coder-128k.Modelfile | 3 | ||||
| -rw-r--r-- | inference/modelfiles/qwen2.5-coder-64k.Modelfile | 6 | ||||
| -rw-r--r-- | ollama/.gitkeep | 0 |
6 files changed, 41 insertions, 18 deletions
@@ -1 +1,2 @@ -/inference/ollama +/comfyui +/ollama diff --git a/inference/.gitkeep b/comfyui/.gitkeep index e69de29..e69de29 100644 --- a/inference/.gitkeep +++ b/comfyui/.gitkeep diff --git a/docker-compose.yml b/docker-compose.yml index bb5046b..8c1fa10 100644 --- a/docker-compose.yml +++ b/docker-compose.yml @@ -1,29 +1,60 @@ services: - inference: + ollama: image: ollama/ollama:rocm restart: always environment: - OLLAMA_MAX_QUEUE: 512 + OLLAMA_MAX_QUEUE: 8192 OLLAMA_MAX_LOADED_MODELS: 1 - OLLAMA_NUM_PARALLEL: 1 - OLLAMA_KEEP_ALIVE: 15m + OLLAMA_NUM_PARALLEL: 5 + OLLAMA_KEEP_ALIVE: 0 OLLAMA_FLASH_ATTENTION: 1 OLLAMA_GPU_OVERHEAD: 2147483648 OLLAMA_NO_CLOUD: 1 LLAMA_ARG_IMAGE_MAX_TOKENS: 1120 volumes: - type: bind - source: ./inference/ollama + source: ./ollama/data target: /root/.ollama read_only: false - type: bind - source: ./inference/modelfiles + source: ./ollama/modelfiles target: /root/modelfiles - read_only: false + read_only: true ports: - 11434:11434 devices: - /dev/kfd:/dev/kfd - /dev/dri:/dev/dri - shm_size: 4gb + + comfy-ui: + image: corundex/comfyui-rocm:latest + restart: always + environment: + CLI_ARGS: --listen 0.0.0.0 + volumes: + - type: bind + source: ./comfyui/custom_nodes + target: /workspace/ComfyUI/custom_nodes + read_only: false + - type: bind + source: ./comfyui/input + target: /workspace/ComfyUI/input + read_only: false + - type: bind + source: ./comfyui/models + target: /workspace/ComfyUI/models + read_only: false + - type: bind + source: ./comfyui/output/ + target: /workspace/ComfyUI/output/ + read_only: false + - type: bind + source: ./comfyui/user + target: /workspace/ComfyUI/user + read_only: false + ports: + - 8188:8188 + devices: + - /dev/kfd:/dev/kfd + - /dev/dri:/dev/dri diff --git a/inference/modelfiles/qwen2.5-coder-128k.Modelfile b/inference/modelfiles/qwen2.5-coder-128k.Modelfile deleted file mode 100644 index ac6d19c..0000000 --- a/inference/modelfiles/qwen2.5-coder-128k.Modelfile +++ /dev/null @@ -1,3 +0,0 @@ -FROM qwen2.5-coder:14b - -PARAMETER num_ctx 131072 diff --git a/inference/modelfiles/qwen2.5-coder-64k.Modelfile b/inference/modelfiles/qwen2.5-coder-64k.Modelfile deleted file mode 100644 index 09395dd..0000000 --- a/inference/modelfiles/qwen2.5-coder-64k.Modelfile +++ /dev/null @@ -1,6 +0,0 @@ -FROM qwen2.5-coder:14b - -PARAMETER num_ctx 65536 - -SYSTEM """You are a helpful assistant with access to tools. When you decide to call a tool, you MUST output ONLY the raw JSON object matching the tool call schema. Do NOT wrap the JSON block in markdown code blocks like ```json or prefix it with any text. Output the JSON object directly.""" - diff --git a/ollama/.gitkeep b/ollama/.gitkeep new file mode 100644 index 0000000..e69de29 --- /dev/null +++ b/ollama/.gitkeep |