diff --git a/README.md b/README.md index 80748c4..057cd96 100644 --- a/README.md +++ b/README.md @@ -21,7 +21,7 @@ This is **not** a new engine. You already run the public [MiaAI / Anemll DeepSee | Piece | Where | URL | Window (live) | |---|---|---|---| | **DeepSeek-V4-Flash-0731** TP=2 | both Sparks | `http://127.0.0.1:8888/v1` | **347392** | -| **Qwen3.8-27B** Unsloth UD-Q4_K_XL + **mmproj** + **MTP4** + ngram | **node2 only** | `http://192.168.100.11:8100/v1` | **116224** engine. Client prompt budget **~95k** if you also want 20k out (in+out share the slot). Native VL. | +| **Qwen3.8-27B** Unsloth UD-Q4_K_XL + **mmproj** + **MTP4** + ngram | **node2 only** | `http://$N2_IP:8100/v1` | **116224** engine. Client prompt budget **~95k** if you also want 20k out (in+out share the slot). Native VL. | | **Qwen3-VL-4B sidecar** | **gone in Dream** | — | Prime-only. Dream eyes = the 27B roommate. | | **Baton** (this repo) | head node | `http://127.0.0.1:8877/v1` | advertises **347392** | diff --git a/dream-baton.py b/dream-baton.py index 796c05e..32da05d 100755 --- a/dream-baton.py +++ b/dream-baton.py @@ -22,7 +22,8 @@ HOST = os.environ.get("BATON_HOST", "127.0.0.1") PORT = int(os.environ.get("BATON_PORT", "8877")) -QWEN = os.environ.get("BATON_QWEN", "http://192.168.100.11:8100/v1").rstrip("/") +# Default is RFC 5737 TEST-NET-1; override with BATON_QWEN (or N2_IP via start-baton.sh). +QWEN = os.environ.get("BATON_QWEN", "http://192.0.2.10:8100/v1").rstrip("/") DS4F = os.environ.get("BATON_0731", "http://127.0.0.1:8888/v1").rstrip("/") QWEN_MODEL = os.environ.get("BATON_QWEN_MODEL", "Qwen3.8-27B") DS4F_MODEL = os.environ.get("BATON_0731_MODEL", "deepseek-v4-flash-0731") diff --git a/env.example b/env.example index 6453fca..d97d795 100644 --- a/env.example +++ b/env.example @@ -2,7 +2,8 @@ # cp env.example .env # # Qwen must be reached at N2_IP, never 127.0.0.1, from the head node. -N2_IP=192.168.100.11 +# 192.0.2.10 is RFC 5737 TEST-NET-1 — replace with your node2 address. +N2_IP=192.0.2.10 SPARK2=spark2 QWEN_PORT=8100 QWEN_CTX=116224 @@ -16,7 +17,7 @@ LLAMA_SERVER=${HOME}/llama.cpp-v9/build/bin/llama-server BATON_HOST=127.0.0.1 BATON_PORT=8877 BATON_0731=http://127.0.0.1:8888/v1 -BATON_QWEN=http://192.168.100.11:8100/v1 +BATON_QWEN=http://192.0.2.10:8100/v1 BATON_MAX_LEN=347392 BATON_QWEN_CTX=116224 BATON_QWEN_SAFE=100000 diff --git a/run-prime-then-solo-qwen-69.sh b/run-prime-then-solo-qwen-69.sh index 020a4ea..96a79ea 100755 --- a/run-prime-then-solo-qwen-69.sh +++ b/run-prime-then-solo-qwen-69.sh @@ -10,7 +10,8 @@ ROOT="$(cd "$(dirname "$0")" && pwd)" OUT="$ROOT/runs" LOG="$HOME/logs/prime-solo-qwen-69.$(date +%Y%m%d-%H%M%S).log" ALERT="$HOME/.hermes/scripts/alertbot-send.sh" -N2_IP="${N2_IP:-192.168.100.11}" +# RFC 5737 TEST-NET-1 fallback — override via N2_IP +N2_IP="${N2_IP:-192.0.2.10}" KW='{"chat_template_kwargs":{"thinking":false,"enable_thinking":false},"enable_thinking":false}' mkdir -p "$OUT" "$(dirname "$LOG")" exec > >(tee -a "$LOG") 2>&1 @@ -45,7 +46,7 @@ systemctl --user restart hermes-gateway-orchestrator hermes-gateway-dobby \ hermes-gateway-light hermes-gateway-smeagle || true if [[ -x "$HOME/.hermes/scripts/ensure-qwen-vision.sh" ]]; then QWEN_VISION_WAIT_SECS=300 QWEN_VISION_FORCE=1 \ - QWEN_VISION_BACKEND_HOST=192.168.100.11 \ + QWEN_VISION_BACKEND_HOST="${N2_IP}" \ bash "$HOME/.hermes/scripts/ensure-qwen-vision.sh" || echo "WARN vision n2" fi python3 - <<'PY' diff --git a/run-tool-eval-69.sh b/run-tool-eval-69.sh index 9bf1f5e..ce04719 100755 --- a/run-tool-eval-69.sh +++ b/run-tool-eval-69.sh @@ -35,12 +35,15 @@ run_one() { date -Iseconds } +# RFC 5737 TEST-NET-1 fallback — override via N2_IP +N2_IP="${N2_IP:-192.0.2.10}" + # keep the trio alive — fail loud if someone parked an engine curl -sf --max-time 5 http://127.0.0.1:8888/v1/models | grep -q deepseek -curl -sf --max-time 5 http://192.168.100.11:8100/v1/models | grep -qi qwen +curl -sf --max-time 5 "http://${N2_IP}:8100/v1/models" | grep -qi qwen run_one "dream-0731-348k" "http://127.0.0.1:8888" "deepseek-v4-flash-0731" -run_one "dream-qwen38-gguf-88k" "http://192.168.100.11:8100" "Qwen3.8-27B" +run_one "dream-qwen38-gguf-88k" "http://${N2_IP}:8100" "Qwen3.8-27B" echo "DONE both 69s" ls -lt "$OUT" | head diff --git a/scripts/run-teb.sh b/scripts/run-teb.sh index a0f45de..9370b06 100755 --- a/scripts/run-teb.sh +++ b/scripts/run-teb.sh @@ -7,7 +7,8 @@ ROOT="$(cd "$(dirname "$0")/.." && pwd)" # shellcheck disable=SC1091 [[ -f "$ROOT/.env" ]] && set -a && source "$ROOT/.env" && set +a export PATH="${HOME}/.local/bin:${PATH}" -N2_IP="${N2_IP:-192.168.100.11}" +# RFC 5737 TEST-NET-1 fallback — override via N2_IP in .env +N2_IP="${N2_IP:-192.0.2.10}" QWEN_PORT="${QWEN_PORT:-8100}" BATON_PORT="${BATON_PORT:-8877}" OUT="${ROOT}/runs" diff --git a/scripts/smoke.sh b/scripts/smoke.sh index 6a63aa4..78a2bfb 100755 --- a/scripts/smoke.sh +++ b/scripts/smoke.sh @@ -4,7 +4,8 @@ set -euo pipefail ROOT="$(cd "$(dirname "$0")/.." && pwd)" # shellcheck disable=SC1091 [[ -f "$ROOT/.env" ]] && set -a && source "$ROOT/.env" && set +a -N2_IP="${N2_IP:-192.168.100.11}" +# RFC 5737 TEST-NET-1 fallback — override via N2_IP in .env +N2_IP="${N2_IP:-192.0.2.10}" QWEN_PORT="${QWEN_PORT:-8100}" chat() { diff --git a/scripts/start-baton.sh b/scripts/start-baton.sh index 7a8d0eb..a99a79d 100755 --- a/scripts/start-baton.sh +++ b/scripts/start-baton.sh @@ -7,7 +7,8 @@ ROOT="$(cd "$(dirname "$0")/.." && pwd)" export BATON_HOST="${BATON_HOST:-127.0.0.1}" export BATON_PORT="${BATON_PORT:-8877}" export BATON_0731="${BATON_0731:-http://127.0.0.1:8888/v1}" -export BATON_QWEN="${BATON_QWEN:-http://192.168.100.11:8100/v1}" +# Prefer BATON_QWEN, else compose from N2_IP (RFC 5737 TEST-NET-1 if unset). +export BATON_QWEN="${BATON_QWEN:-http://${N2_IP:-192.0.2.10}:8100/v1}" export BATON_MAX_LEN="${BATON_MAX_LEN:-347392}" export BATON_QWEN_SAFE="${BATON_QWEN_SAFE:-75000}" diff --git a/scripts/start-roommate.sh b/scripts/start-roommate.sh index 69453b2..44f06cf 100755 --- a/scripts/start-roommate.sh +++ b/scripts/start-roommate.sh @@ -6,7 +6,8 @@ ROOT="$(cd "$(dirname "$0")/.." && pwd)" # shellcheck disable=SC1091 [[ -f "$ROOT/.env" ]] && set -a && source "$ROOT/.env" && set +a -N2_IP="${N2_IP:-192.168.100.11}" +# RFC 5737 TEST-NET-1 fallback — override via N2_IP in .env +N2_IP="${N2_IP:-192.0.2.10}" SPARK2="${SPARK2:-spark2}" PORT="${QWEN_PORT:-8100}" CTX="${QWEN_CTX:-124000}" diff --git a/scripts/status.sh b/scripts/status.sh index a98a75b..322f189 100755 --- a/scripts/status.sh +++ b/scripts/status.sh @@ -4,7 +4,8 @@ set -euo pipefail ROOT="$(cd "$(dirname "$0")/.." && pwd)" # shellcheck disable=SC1091 [[ -f "$ROOT/.env" ]] && set -a && source "$ROOT/.env" && set +a -N2_IP="${N2_IP:-192.168.100.11}" +# RFC 5737 TEST-NET-1 fallback — override via N2_IP in .env +N2_IP="${N2_IP:-192.0.2.10}" probe() { local name="$1" url="$2"