feat: add llama-swap configuration to .env and update llama-swap.sh for new variables
Browse files- .env.example +6 -1
- llama-swap.sh +9 -9
.env.example
CHANGED
|
@@ -19,4 +19,9 @@ SERVER_INTERNAL_PORT="8080"
|
|
| 19 |
RPC_HOSTS="HOST:PORT,HOST:PORT"
|
| 20 |
HF_TOKEN=""
|
| 21 |
LLAMA_API_KEY=""
|
| 22 |
-
LLAMA_CACHE="/tmp/empty-llama-cache"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
RPC_HOSTS="HOST:PORT,HOST:PORT"
|
| 20 |
HF_TOKEN=""
|
| 21 |
LLAMA_API_KEY=""
|
| 22 |
+
LLAMA_CACHE="/tmp/empty-llama-cache"
|
| 23 |
+
LLAMA_SWAP_VERSION="cuda-unified"
|
| 24 |
+
LLAMA_SWAP_IMAGE="ghcr.io/mostlygeek/llama-swap:${LLAMA_SWAP_VERSION}"
|
| 25 |
+
LLAMA_SWAP_CONTAINER_NAME="llama-swap"
|
| 26 |
+
LLAMA_SWAP_PORT="8080"
|
| 27 |
+
LLAMA_SWAP_API_KEY="${LLAMA_API_KEY}"
|
llama-swap.sh
CHANGED
|
@@ -35,15 +35,15 @@ fi
|
|
| 35 |
mkdir -p "${PRESETS_DIR}/llama-swap-logs"
|
| 36 |
mkdir -p "${PRESETS_DIR}/llama-swap-state"
|
| 37 |
|
| 38 |
-
DOCKER=(sudo docker)
|
| 39 |
-
"${DOCKER[@]}" image inspect "${BUILDER_IMAGE}" >/dev/null
|
| 40 |
-
"${DOCKER[@]}" build \
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
|
| 48 |
"${DOCKER[@]}" rm --force "${LLAMA_SWAP_CONTAINER_NAME}" >/dev/null 2>&1 || true
|
| 49 |
"${DOCKER[@]}" run --detach --rm \
|
|
|
|
| 35 |
mkdir -p "${PRESETS_DIR}/llama-swap-logs"
|
| 36 |
mkdir -p "${PRESETS_DIR}/llama-swap-state"
|
| 37 |
|
| 38 |
+
# DOCKER=(sudo docker)
|
| 39 |
+
# "${DOCKER[@]}" image inspect "${BUILDER_IMAGE}" >/dev/null
|
| 40 |
+
# "${DOCKER[@]}" build \
|
| 41 |
+
# --build-arg "BUILDER_IMAGE=${BUILDER_IMAGE}" \
|
| 42 |
+
# --build-arg "LLAMA_SWAP_VERSION=${LLAMA_SWAP_VERSION}" \
|
| 43 |
+
# --build-arg "LLAMA_SWAP_SHA256=${LLAMA_SWAP_SHA256}" \
|
| 44 |
+
# --tag "${LLAMA_SWAP_IMAGE}" \
|
| 45 |
+
# --file "${SCRIPT_DIR}/Dockerfile.llama-swap" \
|
| 46 |
+
# "${SCRIPT_DIR}"
|
| 47 |
|
| 48 |
"${DOCKER[@]}" rm --force "${LLAMA_SWAP_CONTAINER_NAME}" >/dev/null 2>&1 || true
|
| 49 |
"${DOCKER[@]}" run --detach --rm \
|