Frank
f93096f2bf
llama.cpp service replacing retired vllm-awq-qwen36 on port 7770 with alias frank-default-small. Wrapper sources frank-default-small.env for tunables: 3 parallel slots at 262144 ctx, draft-dflash n-max 7, KV K=q8_0 V=f16 on GPU, all MoE experts on CPU (--n-cpu-moe 40), non-expert tensors on GPU, mmproj vision + jinja. Unit: frank-default-small.service (ExecStart in dankserver repo clone /home/dank/vLLM/scripts).
18 lines
539 B
Bash
18 lines
539 B
Bash
# Tunables for frank-default-small (HauhauCS Q6_K_P + DFlash draft, dankserver)
|
|
# Runtime: llama-server b10108; RTX 3090 24GB; Ryzen 7 7700X; 62GB RAM.
|
|
MODEL_DIR=/home/dank/models
|
|
ALIAS=frank-default-small
|
|
HOST=0.0.0.0
|
|
PORT=7770
|
|
CTX_SIZE=262144
|
|
PARALLEL=3
|
|
SPEC_TYPE=draft-dflash
|
|
SPEC_N_MAX=7
|
|
CACHE_TYPE_K=q8_0
|
|
CACHE_TYPE_V=f16
|
|
# All 40 layers keep their MoE expert weights on CPU (targeted expert offload).
|
|
N_CPU_MOE=40
|
|
# Every non-expert tensor (attention, norms, embedding, LM head) goes to GPU.
|
|
N_GPU_LAYERS=all
|
|
THREADS=8
|
|
THREADS_HTTP=8 |