28 lines
711 B
Bash
28 lines
711 B
Bash
DUCK_LLAMA_SERVER_BIN=./vendor/llama.cpp/build/bin/llama-server
|
|
DUCK_MAIN_MODEL_PATH=./models/Qwen3.6/nonMTP/Qwen3.6-35B-A3B-UD-Q4_K_M.gguf
|
|
DUCK_MTP_MODEL_PATH=./models/Qwen3.6/MTP/Qwen3.6-35B-A3B-MTP-UD-Q4_K_M.gguf
|
|
|
|
DUCK_MAIN_PORT=8081
|
|
DUCK_MAIN_MTP_PORT=8085
|
|
DUCK_CTX_SIZE=4096
|
|
DUCK_N_GPU_LAYERS=20
|
|
DUCK_PARALLEL=1
|
|
DUCK_LLAMA_DEVICE=Vulkan0
|
|
DUCK_LLAMA_EXTRA_ARGS="--reasoning-budget 512 --cache-ram 0"
|
|
DUCK_HOST=127.0.0.1
|
|
|
|
DUCK_API_HOST=127.0.0.1
|
|
DUCK_API_PORT=8000
|
|
|
|
DUCK_WORKSPACE=./workspace
|
|
DUCK_DB_PATH=./data/duck.sqlite3
|
|
|
|
DUCK_MAX_INPUT_TOKENS=49152
|
|
DUCK_MAX_RECENT_EVENTS_TOKENS=12000
|
|
DUCK_MAX_MEMORY_TOKENS=8000
|
|
DUCK_MAX_SKILL_TOKENS=6000
|
|
|
|
QDRANT_URL=http://127.0.0.1:6333
|
|
|
|
DUCK_SKIP_LIVE_LLM_TESTS=0
|