Fix bundles push and simplify Docker image (#935)

* docker fixes and simplifications

* v1.14.4

Author: pascal-pfeiffer

.dockerignore

@@ -2,8 +2,8 @@
 .github
 .mypy_cache
 .neptune
-bundles
 wandb
+llmstudio_mnt/
 .pytest_cache
 data
 documentation

.gitignore

@@ -14,6 +14,7 @@ subs/
 /datasets/
 .idea/
 .local/
+llmstudio_mnt/
 
 output
 

Dockerfile

@@ -46,17 +46,21 @@ RUN make setup && chmod -R 777 /workspace/.venv
 ENV PATH=/workspace/.venv/bin:$PATH
 
 # We need to create a mount point for the user to mount their volume
-# All persistent data lives in /home/llmstudio/mount
-RUN mkdir -p /home/llmstudio/mount
-ENV H2O_LLM_STUDIO_WORKDIR=/home/llmstudio/mount
+# All persistent data lives in /mount
+# RUN mkdir -p /mount
+ENV H2O_LLM_STUDIO_WORKDIR=/mount
 
 # Download the demo datasets and place in the /workspace/demo directory
 # Set the environment variable for the demo datasets
 ENV H2O_LLM_STUDIO_DEMO_DATASETS=/workspace/demo
 COPY llm_studio/download_default_datasets.py /workspace/
 RUN python download_default_datasets.py
 
-COPY . /workspace
+COPY ./llm_studio /workspace/llm_studio
+COPY ./prompts /workspace/prompts
+COPY ./model_cards /workspace/model_cards
+COPY ./LICENSE /workspace/LICENSE
+COPY ./entrypoint.sh /workspace/entrypoint.sh
 
 # Remove unnecessary packages remove build packages again
 # Prevent removal of cuda packages
@@ -67,22 +71,20 @@ RUN apt-get purge -y git curl python3.10-distutils software-properties-common \
 
 USER llmstudio
 
+ENV HF_HOME=/mount/huggingface
+ENV TRITON_CACHE_DIR=/mount/.triton/cache
+
 ENV HF_HUB_DISABLE_TELEMETRY=1
 ENV DO_NOT_TRACK=1
 
 # Set the environment variables for the wave server
 ENV H2O_WAVE_APP_ADDRESS=http://127.0.0.1:8756
 ENV H2O_WAVE_MAX_REQUEST_SIZE=25MB
 ENV H2O_WAVE_NO_LOG=true
-ENV H2O_WAVE_PRIVATE_DIR="/download/@/home/llmstudio/mount/output/download"
+ENV H2O_WAVE_PRIVATE_DIR="/download/@/mount/output/download"
 
 USER root
 
-# Make all of the files in the llmstudio directory read & writable for all users so that the
-# application can install other (non-persisted) new packages and other things
-# if it wants to. e.g. triton uses /home/llmstudio/.triton as a cache directory.
-RUN chmod -R 777 /home/llmstudio
-
 # Make the entrypoint.sh script executable by all users
 RUN chmod 755 /workspace/entrypoint.sh
 

Dockerfile.bundle

@@ -2,5 +2,5 @@ FROM cgr.dev/chainguard/bash:latest
 
 ARG VERSION
 
-COPY bundles/ai.h2o.llmstudio.${VERSION}.wave /app/ai.h2o.llmstudio.${VERSION}.wave
+COPY ./bundles/ai.h2o.llmstudio.${VERSION}.wave /app/ai.h2o.llmstudio.${VERSION}.wave
 ENV WAVE_BUNDLE_FILE /app/ai.h2o.llmstudio.${VERSION}.wave

Makefile

@@ -3,7 +3,7 @@ SHELL := /bin/bash
 UV ?= uv
 RUN ?= $(UV) run
 PWD = $(shell pwd)
-DOCKER_IMAGE ?= gcr.io/vorvan/h2oai/h2o-llmstudio:nightly
+DOCKER_IMAGE ?= h2oairelease/h2oai-llmstudio-app:nightly
 APP_VERSION=$(shell sed -n 's/^version = //p' pyproject.toml | tr -d '"')
 
 ifeq ($(origin H2O_LLM_STUDIO_WORKDIR), environment)
@@ -188,7 +188,7 @@ endif
 		-it \
 		-u `id -u`:`id -g` \
 		-p 10101:10101 \
-		-v `pwd`/llmstudio_mnt:/home/llmstudio/mount \
+		-v `pwd`/llmstudio_mnt:/mount \
 		$(DOCKER_IMAGE)
 
 # Perform a local Trivy scan for CVEs

README.md

@@ -156,8 +156,7 @@ docker run \
     -it \
     -u `id -u`:`id -g` \
     -p 10101:10101 \
-    -v `pwd`/llmstudio_mnt:/home/llmstudio/mount \
-    -v ~/.cache:/home/llmstudio/.cache \
+    -v `pwd`/llmstudio_mnt:/mount \
     h2oairelease/h2oai-llmstudio-app:latest
 ```
 
@@ -180,8 +179,7 @@ docker run \
     -it \
     -u `id -u`:`id -g` \
     -p 10101:10101 \
-    -v `pwd`/llmstudio_mnt:/home/llmstudio/mount \
-    -v ~/.cache:/home/llmstudio/.cache \
+    -v `pwd`/llmstudio_mnt:/mount \
     h2o-llmstudio
 ```
 

app.toml.template

@@ -15,7 +15,7 @@ CPUReservation = "10"
 MemoryReservation = "118Gi"
 MemoryLimit = "118Gi"
 GPUCount = 1
-VolumeMount = "/home/llmstudio/mount"
+VolumeMount = "/mount"
 VolumeSize = "1Ti"
 ResourceVolumeSize = "1Ti"
 EnableSHM = true
@@ -24,7 +24,7 @@ RoutingMode = "BASE_URL"
 
 [[Env]]
 Name = "H2O_LLM_STUDIO_WORKDIR"
-Value = "/home/llmstudio/mount"
+Value = "/mount"
 
 [[Env]]
 Name = "HOME"
@@ -36,7 +36,7 @@ Value = "True"
 
 [[Env]]
 Name = "H2O_WAVE_PRIVATE_DIR"
-Value = "/download/@/home/llmstudio/mount/output/download"
+Value = "/download/@/mount/output/download"
 
 [[Env]]
 Name = "HF_HUB_ENABLE_HF_TRANSFER"

documentation/docs/get-started/set-up-llm-studio.md

@@ -252,8 +252,7 @@ docker run \
     --rm \
     -it \
     -p 10101:10101 \
-    -v `pwd`/llmstudio_mnt:/home/llmstudio/mount \
-    -v ~/.cache:/home/llmstudio/.cache \
+    -v `pwd`/llmstudio_mnt:/mount \
     h2oairelease/h2oai-llmstudio-app:latest
 ```
 
@@ -275,8 +274,7 @@ docker run \
     --rm \
     -it \
     -p 10101:10101 \
-    -v `pwd`/llmstudio_mnt:/home/llmstudio/mount \
-    -v ~/.cache:/home/llmstudio/.cache \
+    -v `pwd`/llmstudio_mnt:/mount \
     h2o-llmstudio
 ```
 

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "h2o-llmstudio"
-version = "1.14.3"
+version = "1.14.4"
 readme = "README.md"
 license = {file = "LICENSE"}
 requires-python = "==3.10.*"

tests/ui/llm_studio_page.py

@@ -3,7 +3,7 @@
 from hac_playwright.pages.base import BasePage
 from playwright.sync_api import expect
 
-CLOUD_FILESYSTEM_PATH = "/home/llmstudio/mount/data/user/oasst"
+CLOUD_FILESYSTEM_PATH = "/mount/data/user/oasst"
 LOCAL_FILESYSTEM_PATH = os.path.join(os.getcwd(), "data/user/oasst")