MOVZXandHinata
d9ca3bc4f2
MAX_SEQS and the prefill chunk were fixed in the start profile, so any change to concurrency needed an edit to the script. Read both from .env, and validate them before the engine starts. The chunk must be at least the sequence count, the rule vLLM applies at scheduler.py:309. A MAX_SEQS whose full context exceeds the 924,993-token pool prints a note, because that shape preempts and recomputes instead of failing. The captured graph list was a fixed [1,2,4,8,12,16], which covers a decode batch of 4 requests and nothing wider. One decode step is K+1 = 4 tokens per request, so the list now follows MAX_SEQS: 1, 2, 4, then every 4 tokens up to MAX_SEQS x 4. At the shipped 4 it produces the old list exactly. Above 8 graphs a note prints, since each size costs capture time and graph VRAM. Drop --mm-processor-cache-gb. This workload sends each image once, so the cache never had a hit to keep, and vLLM's own 4 GiB default is the better base. Co-Authored-By: Hinata <hinata@idihore.id> Signed-off-by: MOVZX <movzx@yahoo.com>
65 lines
2.2 KiB
YAML
65 lines
2.2 KiB
YAML
services:
|
|
vllm:
|
|
image: ${IMAGE:-vllm}:${IMAGE_TAG:-latest}
|
|
container_name: ${CONTAINER_NAME:-vllm}
|
|
build:
|
|
context: .
|
|
args:
|
|
MAX_JOBS: ${MAX_JOBS:-32}
|
|
ARCH: ${ARCH:-8.6}
|
|
additional_contexts:
|
|
ops: ./artifacts
|
|
uvcache: ${UV_CACHE_DIR:-${HOME}/.cache/uv}
|
|
pipcache: ${PIP_CACHE_DIR:-${HOME}/.cache/pip}
|
|
deploy:
|
|
resources:
|
|
reservations:
|
|
devices:
|
|
- driver: nvidia
|
|
count: all
|
|
capabilities: [gpu]
|
|
command: ["${MODEL_MOUNT_PATH:-/models/Qwen3.8-Flash-Next}"]
|
|
shm_size: 16g
|
|
ulimits:
|
|
memlock: -1
|
|
stack: 67108864
|
|
ports:
|
|
- "${HOST_BIND_ADDR:-0.0.0.0}:${HOST_PORT:-8080}:8000"
|
|
environment:
|
|
MODEL_NAME: ${MODEL_NAME:-Hinata}
|
|
PORT: 8000
|
|
SCALES: /opt/vllm/qsa_kv_scales.json
|
|
VLLM_API_KEY: ${VLLM_API_KEY:-}
|
|
PLE_HOME: ${PLE_HOME:-0}
|
|
CUDA_VISIBLE_DEVICES: ${CUDA_VISIBLE_DEVICES:-0,1,2,3}
|
|
AUTO_TOPO: ${AUTO_TOPO:-0}
|
|
AUTO_TOPO_NCCL: ${AUTO_TOPO_NCCL:-legacy}
|
|
AUTO_TOPO_BIND: ${AUTO_TOPO_BIND:-0}
|
|
LANE_GPUS: ${LANE_GPUS:-}
|
|
LANE_CPUS: ${LANE_CPUS:-}
|
|
TOPO_TIMEOUT: ${TOPO_TIMEOUT:-20}
|
|
GUARD: ${GUARD:-warn}
|
|
COUNTERS: ${COUNTERS:-0}
|
|
VLLM_LOG_STATS_INTERVAL: ${VLLM_LOG_STATS_INTERVAL:-1}
|
|
MAX_SEQS: ${MAX_SEQS:-4}
|
|
MAX_BATCHED_TOKENS: ${MAX_BATCHED_TOKENS:-2048}
|
|
QWEN38_ABLITERATION: ${QWEN38_ABLITERATION:-Qwen3.8-Flash-Next-Uncensored}
|
|
volumes:
|
|
- "${MODEL_DIR:?Set MODEL_DIR in .env}:${MODEL_MOUNT_PATH:-/models/Qwen3.8-Flash-Next}:ro"
|
|
- ${ABLITERATION_DIR:-./abliteration}:/opt/qwen38/abliteration:ro
|
|
- ${CACHE_HOST_DIR:-./.vllm-cache}:/cache
|
|
- ${SCALES:-${MODEL_DIR}/qsa_kv_scales.json}:/opt/vllm/qsa_kv_scales.json:ro
|
|
- ./start.sh:/opt/vllm/start.sh:ro
|
|
restart: unless-stopped
|
|
stop_grace_period: 20s
|
|
logging:
|
|
driver: json-file
|
|
options:
|
|
max-size: ${LOG_MAX_SIZE:-100m}
|
|
max-file: ${LOG_MAX_FILE:-14}
|
|
healthcheck:
|
|
interval: ${HEALTHCHECK_INTERVAL:-60s}
|
|
timeout: ${HEALTHCHECK_TIMEOUT:-90s}
|
|
start_period: ${HEALTHCHECK_START_PERIOD:-30m}
|
|
retries: ${HEALTHCHECK_RETRIES:-3}
|