vh
355a2407a2
ana-ml2 was upgraded 2026-06 from dual RTX 6000 Ada (46GB, cc 8.9) to
dual RTX PRO 6000 Blackwell Max-Q (96GB, cc 12.0 / sm_120). Update the
stale hardware facts across the workspace:
- CLAUDE.md servers table row
- servers/ana-ml2/README.md hardware spec (+ refreshed system-details.txt)
- stacks/vllm compose + .env.example FP8/KV comments (Ada cc 8.9 -> Blackwell cc 12.0)
- stacks/llama-swap config VRAM-budget comment (48GB -> 96GB, GPU-0 pin)
Also corrects the adjacent stale 'Phi-4-mini' comment in the granite
service block (the service has been Granite 4.1 8B since 34a43a0).
Doc/comment-only; no runtime change.
100 lines
5.0 KiB
Bash
100 lines
5.0 KiB
Bash
# vllm stack tunables. Copy this to `.env` on the server before deploying.
|
|
#
|
|
# cp .env.example .env
|
|
# # edit .env with real values
|
|
# docker compose up -d
|
|
|
|
# Image version — pin for reproducibility (`latest` for edge)
|
|
VLLM_VERSION=latest
|
|
|
|
# Host ports (container always listens on 8000 internally)
|
|
EMBED_PORT=8001
|
|
RERANK_PORT=8002
|
|
REWARD_PORT=8003
|
|
|
|
# GPU assignment — all services share this GPU
|
|
# (ana-ml2 has 0 and 1; default 1 keeps 0 free for heavy LLM work in llama-swap)
|
|
GPU_ID=1
|
|
|
|
# Models — reference by full repo name in API requests
|
|
EMBED_MODEL=Qwen/Qwen3-Embedding-0.6B
|
|
RERANK_MODEL=Qwen/Qwen3-Reranker-0.6B
|
|
# Skywork is a local-path AWQ output, not from HF Hub. Bind-mounted into the
|
|
# reward container at /local-models — see compose.yaml. No env var here for
|
|
# the model path itself since it's hard-coded in the compose command.
|
|
|
|
# GPU memory split — fractions are of TOTAL GPU memory, not free memory.
|
|
# vLLM profiles each service independently, so each slice must be large enough
|
|
# to fit that service's model + KV cache with no awareness of the others.
|
|
# Setting any one too low causes that container to OOM on KV cache allocation
|
|
# with `Available KV cache memory: -X.XX GiB`.
|
|
#
|
|
# Layout on a 48 GB Ada, RIGHT-SIZED 2026-06-05 to free room for the granite
|
|
# summarizer's CUDA graphs + a LoRA hedge. The trio was wildly over-provisioned:
|
|
# embed ran a 5.89x KV pool, reward 3.90x — pure waste for utility models that
|
|
# need ~1.5x. Trimmed to free ~10 GB. (Utilization = fraction of TOTAL GPU mem;
|
|
# each service profiles independently. Too low → that container OOMs on KV with
|
|
# `Available KV cache memory: -X.XX GiB`.)
|
|
# EMBED 0.07 (~3.4 GB) — 0.6B Qwen3 embed @8k; 1.1GB weights + ~2GB KV (~3x)
|
|
# RERANK 0.07 (~3.4 GB) — 0.6B Qwen3 rerank @8k; comfortable
|
|
# REWARD 0.18 (~8.6 GB) — 8B Skywork AWQ @16k; 4.4GB weights + ~3.5GB KV (~1.5x)
|
|
#
|
|
# RETUNED 2026-06-12 for the Ada->Blackwell swap (cards are now 96 GB, so the
|
|
# Ada-era fractions reserved ~2x the bytes for the same models). Empirically
|
|
# re-floored: util climbed in 0.01 steps until each service was stable under
|
|
# 20x parallel inference.
|
|
# embed/rerank 0.05 (~4.8 GB ea) — the load-floor for the 0.6B models (0.04
|
|
# OOMs on engine init; their footprint is fixed weights+context, not KV).
|
|
# reward 0.10 (~9.6 GB) — the real over-provision was here (0.18 -> 0.10).
|
|
# Net: ~11 GB freed on GPU 1. (Fractions are of TOTAL card VRAM — re-floor if
|
|
# the cards change again.)
|
|
EMBED_GPU_MEM_UTIL=0.05
|
|
RERANK_GPU_MEM_UTIL=0.05
|
|
REWARD_GPU_MEM_UTIL=0.10
|
|
|
|
# Context length caps — lower these if VRAM is tight.
|
|
# Qwen3-Embedding supports up to 32k; reranker up to 32k.
|
|
# Skywork capped at 16k server-side as defense-in-depth; JudgeClient also
|
|
# enforces the cap at dispatch time per spec.
|
|
EMBED_MAX_MODEL_LEN=8192
|
|
RERANK_MAX_MODEL_LEN=8192
|
|
REWARD_MAX_MODEL_LEN=16384
|
|
|
|
# Optional API key — leave blank for no auth (fine on the internal network).
|
|
# If set, all three services require `Authorization: Bearer <key>`.
|
|
API_KEY=
|
|
|
|
# HuggingFace token — only needed for gated models in the HF-Hub-loaded
|
|
# services (embed/rerank). Reward is local-path, ignores this.
|
|
HF_TOKEN=
|
|
|
|
# === granite-4.1-8b (production summarizer / dreaming agent) ===
|
|
# Replaced phi4-mini 2026-06-05 (Granite 4.1 8B beat phi4 on precision in
|
|
# brokkr's R15 P03 model-fitness eval). Same GPU-1 slot, reusing phi4's port.
|
|
GRANITE_PORT=8004
|
|
# GPU 1 — co-located with the embed/rerank/reward trio. With phi4 retired, GPU 1
|
|
# has ~19 GB free; granite at 56K + FP8 KV needs ~15-16 GB → ~2-3 GB margin.
|
|
# NOTE: llama-swap also uses GPU 1 dynamically — a large swap-in could contend;
|
|
# pin llama-swap to GPU 0 for clean separation (follow-up).
|
|
GRANITE_GPU_ID=1
|
|
# Official IBM pre-quantized FP8 (compressed-tensors) — calibrated, ~9.6 GB,
|
|
# loaded directly (FP8 native on Blackwell cc 12.0). Fallback to vLLM-native dynamic FP8
|
|
# from BF16: GRANITE_MODEL=ibm-granite/granite-4.1-8b + GRANITE_QUANT=fp8.
|
|
GRANITE_MODEL=ibm-granite/granite-4.1-8b-fp8
|
|
GRANITE_QUANT=compressed-tensors
|
|
GRANITE_SERVED_NAME=granite-4.1-8b
|
|
# 131072 ctx — MAXED 2026-06-13 (was 51200/50K). GPU-1 rebalance: granite shares
|
|
# the card with the trio + qwen35-vl (vision). Qwen was over-provisioned on KV
|
|
# (20x conc @ 32k), so trimming it freed room for granite's ~305k-token pool here.
|
|
# PagedAttention allocates KV per ACTUAL token, so 131072 is only a CEILING — a 1k
|
|
# summarize turn uses ~1k tokens, so the pool holds ~300 concurrently; the "2.33x"
|
|
# headline is worst-case (every request maxing 131k). Granite 4.1 supports 131072.
|
|
GRANITE_MAX_MODEL_LEN=131072
|
|
# FP8 KV cache (native on Blackwell cc 12.0). At 50K ≈ ~4.2 GB (vs ~8.4 GB at fp16).
|
|
GRANITE_KV_CACHE_DTYPE=fp8
|
|
# util 0.35 (~33.6 GB) — tuned 2026-06-13 to leave ~3.5 GB free on GPU 1 alongside
|
|
# the trio + qwen co-tenants. On this shared card vLLM needs free >= util*total at
|
|
# startup, and START ORDER matters: trim qwen FIRST, then grow granite, else granite
|
|
# OOMs against the full card. (0.37 overshot to 1.7 GB free; 0.35 lands ~3.7 GB.)
|
|
GRANITE_GPU_MEM_UTIL=0.35
|