Update ancient Docker instructions

This commit is contained in:
oobabooga 2023-11-17 19:52:30 -08:00
parent e0ca49ed9c
commit d1a58da52f
6 changed files with 16 additions and 22 deletions

View file

@ -5,5 +5,4 @@ Dockerfile
/models
/presets
/prompts
/softprompts
/training

View file

@ -3,13 +3,8 @@
# https://developer.nvidia.com/cuda-gpus you can find the version for your card here
TORCH_CUDA_ARCH_LIST=7.5
# these commands worked for me with roughly 4.5GB of vram
CLI_ARGS=--model llama-7b-4bit --wbits 4 --listen --auto-devices
# the following examples have been tested with the files linked in docs/README_docker.md:
# example running 13b with 4bit/128 groupsize : CLI_ARGS=--model llama-13b-4bit-128g --wbits 4 --listen --groupsize 128 --pre_layer 25
# example with loading api extension and public share: CLI_ARGS=--model llama-7b-4bit --wbits 4 --listen --auto-devices --no-stream --extensions api --share
# example running 7b with 8bit groupsize : CLI_ARGS=--model llama-7b --load-in-8bit --listen --auto-devices
# your command-line flags go here:
CLI_ARGS=
# the port the webui binds to on the host
HOST_PORT=7860
@ -21,10 +16,5 @@ HOST_API_PORT=5000
# the port the api binds to inside the container
CONTAINER_API_PORT=5000
# the port the api stream endpoint binds to on the host
HOST_API_STREAM_PORT=5005
# the port the api stream endpoint binds to inside the container
CONTAINER_API_STREAM_PORT=5005
# the version used to install text-generation-webui from
WEBUI_VERSION=HEAD

View file

@ -73,5 +73,5 @@ RUN --mount=type=cache,target=/root/.cache/pip,rw \
ENV CLI_ARGS=""
EXPOSE ${CONTAINER_PORT:-7860} ${CONTAINER_API_PORT:-5000} ${CONTAINER_API_STREAM_PORT:-5005}
EXPOSE ${CONTAINER_PORT:-7860} ${CONTAINER_API_PORT:-5000}
CMD . /app/venv/bin/activate && python3 server.py ${CLI_ARGS}

View file

@ -11,7 +11,6 @@ services:
ports:
- "${HOST_PORT:-7860}:${CONTAINER_PORT:-7860}"
- "${HOST_API_PORT:-5000}:${CONTAINER_API_PORT:-5000}"
- "${HOST_API_STREAM_PORT:-5005}:${CONTAINER_API_STREAM_PORT:-5005}"
stdin_open: true
tty: true
volumes: