tools: default dash5 paths to home layout
This commit is contained in:
@@ -9,8 +9,8 @@ set -u
|
|||||||
cd "$(dirname "$0")/../.."
|
cd "$(dirname "$0")/../.."
|
||||||
export PATH=$HOME/.cargo/bin:/usr/local/cuda-12.9/bin:$PATH
|
export PATH=$HOME/.cargo/bin:/usr/local/cuda-12.9/bin:$PATH
|
||||||
export CUDA_HOME=${CUDA_HOME:-/usr/local/cuda-12.9}
|
export CUDA_HOME=${CUDA_HOME:-/usr/local/cuda-12.9}
|
||||||
MODEL=${MODEL:-/opt/wjh/models/qwen3-8b}
|
MODEL=${MODEL:-$HOME/models/qwen3-8b}
|
||||||
GGUF=${GGUF:-/opt/wjh/models/qwen3-8b/qwen3-8b-bf16.gguf}
|
GGUF=${GGUF:-$HOME/models/qwen3-8b/qwen3-8b-bf16.gguf}
|
||||||
LLAMA_BIN=${LLAMA_BIN:-third_party/llama.cpp/build/bin/llama-server}
|
LLAMA_BIN=${LLAMA_BIN:-third_party/llama.cpp/build/bin/llama-server}
|
||||||
XBIN=./target/release/xserv-server
|
XBIN=./target/release/xserv-server
|
||||||
PPS=${PPS:-1 2 4}
|
PPS=${PPS:-1 2 4}
|
||||||
|
|||||||
@@ -10,8 +10,8 @@
|
|||||||
# Run from the repo root on the GPU host. Produces bench-out/pp{1,2,4}-{xserv,llama}.
|
# Run from the repo root on the GPU host. Produces bench-out/pp{1,2,4}-{xserv,llama}.
|
||||||
|
|
||||||
set -u
|
set -u
|
||||||
MODEL="${MODEL:-/opt/wjh/models/qwen3-8b}"
|
MODEL="${MODEL:-$HOME/models/qwen3-8b}"
|
||||||
GGUF="${GGUF:-/opt/wjh/models/qwen3-8b/qwen3-8b-bf16.gguf}"
|
GGUF="${GGUF:-$HOME/models/qwen3-8b/qwen3-8b-bf16.gguf}"
|
||||||
LIMIT="${LIMIT:-20}"
|
LIMIT="${LIMIT:-20}"
|
||||||
MAXSEQ="${MAXSEQ:-2048}"
|
MAXSEQ="${MAXSEQ:-2048}"
|
||||||
PPS="${PPS:-1 2 4}"
|
PPS="${PPS:-1 2 4}"
|
||||||
|
|||||||
@@ -7,8 +7,8 @@
|
|||||||
# Run from the repo root on the GPU host. Produces bench-out/tp{1,2,4}-{xserv,llama}.
|
# Run from the repo root on the GPU host. Produces bench-out/tp{1,2,4}-{xserv,llama}.
|
||||||
|
|
||||||
set -u
|
set -u
|
||||||
MODEL="${MODEL:-/opt/wjh/models/qwen3-8b}"
|
MODEL="${MODEL:-$HOME/models/qwen3-8b}"
|
||||||
GGUF="${GGUF:-/opt/wjh/models/qwen3-8b/qwen3-8b-bf16.gguf}"
|
GGUF="${GGUF:-$HOME/models/qwen3-8b/qwen3-8b-bf16.gguf}"
|
||||||
LIMIT="${LIMIT:-30}"
|
LIMIT="${LIMIT:-30}"
|
||||||
MAXSEQ="${MAXSEQ:-2048}"
|
MAXSEQ="${MAXSEQ:-2048}"
|
||||||
TPS="${TPS:-1 2 4}"
|
TPS="${TPS:-1 2 4}"
|
||||||
|
|||||||
@@ -12,8 +12,8 @@ set -u
|
|||||||
cd "$(dirname "$0")/.."
|
cd "$(dirname "$0")/.."
|
||||||
export PATH=$HOME/.cargo/bin:/usr/local/cuda-12.9/bin:$PATH
|
export PATH=$HOME/.cargo/bin:/usr/local/cuda-12.9/bin:$PATH
|
||||||
export CUDA_HOME=${CUDA_HOME:-/usr/local/cuda-12.9}
|
export CUDA_HOME=${CUDA_HOME:-/usr/local/cuda-12.9}
|
||||||
MODEL=${MODEL:-/opt/wjh/models/qwen3-8b}
|
MODEL=${MODEL:-$HOME/models/qwen3-8b}
|
||||||
GGUF=${GGUF:-/opt/wjh/models/qwen3-8b/qwen3-8b-bf16.gguf}
|
GGUF=${GGUF:-$HOME/models/qwen3-8b/qwen3-8b-bf16.gguf}
|
||||||
LIMIT=${LIMIT:-20}
|
LIMIT=${LIMIT:-20}
|
||||||
PPS=${PPS:-1 2 4}
|
PPS=${PPS:-1 2 4}
|
||||||
BIN=./target/release/xserv-server
|
BIN=./target/release/xserv-server
|
||||||
|
|||||||
@@ -16,8 +16,8 @@
|
|||||||
set -e
|
set -e
|
||||||
|
|
||||||
REMOTE="dash5"
|
REMOTE="dash5"
|
||||||
REMOTE_DIR="/opt/wjh/projects/xserv"
|
REMOTE_DIR="${REMOTE_DIR:-~/projects/xserv}"
|
||||||
REMOTE_MODEL_DIR="${REMOTE_MODEL_DIR:-/opt/wjh/models/qwen3-8b}"
|
REMOTE_MODEL_DIR="${REMOTE_MODEL_DIR:-~/models/qwen3-8b}"
|
||||||
LOCAL_DIR="$(cd "$(dirname "$0")/.." && pwd)"
|
LOCAL_DIR="$(cd "$(dirname "$0")/.." && pwd)"
|
||||||
|
|
||||||
ACTION="${1:-build}"
|
ACTION="${1:-build}"
|
||||||
|
|||||||
Reference in New Issue
Block a user