feat(llm): dedicate a CPU-only backend for the qwen-code tool-call classifier

fastModel in ~/.qwen/settings.json (permissions.autoMode.classifier) was
aliased onto llama-server's own 27B connection, so every tool-call safety
check queued behind whatever heavy generation was already running on that
model's 2 GPU slots.

Add qwen-classifier: a separate llama.cpp instance, CPU-only, running
Qwen3-4B-Instruct-2507 (the smallest Qwen3 with native >=131072 context,
qwen-code's requirement, without lossy RoPE scaling). Structurally isolated
from llama-server's queue instead of sharing it. Sized for gameserver's
~17GiB free system RAM: q8_0/q8_0 KV at full 131072 ctx (~9.8GiB) + Q4_K_M-
class weights (~2.3GiB) fits comfortably, with better KV quality than the
q4_0 that would've been needed to fit this on the GPU's ~6GiB free VRAM.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

.env.example

@@ -29,18 +29,27 @@ LLAMA_GPU_LAYERS=999
 # this size) — total ~25.6GB, ~6GB headroom, the same footprint the old
 # 131072 fp16 setting used. See docs/research/qwen3.8-27b-quant.md.
 LLAMA_CTX_SIZE=262144
-# Concurrent request slots. Was implicitly 4 (llama.cpp's compiled-in
-# default) with no flag set — under concurrent subagent fan-out, 4 requests
-# split the same GPU compute, so a large-context prefill can queue behind
-# others long enough to blow past OmniRoute's stream-idle timeout, which then
-# cancels the request (see issue-tracker notes on the timeout/cancel loop).
-# Dropped to 2 so each slot gets more compute and finishes prefill sooner;
-# raise back toward 4 if throughput (not latency) becomes the bottleneck
-# instead. Each slot gets LLAMA_CTX_SIZE / LLAMA_PARALLEL tokens of context —
-# real sessions have hit ~66K tokens, so don't drop LLAMA_CTX_SIZE without
-# checking that per-slot number stays comfortably above observed usage.
+# Concurrent request slots — the real hardware ceiling for this GPU, not a
+# tunable to raise for throughput (was implicitly 4, llama.cpp's compiled-in
+# default; dropped to 2 because more contended prefill was blowing requests
+# past OmniRoute's idle timeout — see OMNIROUTE_STREAM_IDLE_TIMEOUT_MS below).
+# The 3rd+ request now queues on llama.cpp itself instead — its own queue has
+# no timeout (tools/server/server-queue.cpp), it just waits for a slot — so
+# the timeout that matters moved to OmniRoute's per-connection
+# providerSpecificData.timeoutMs (dashboard/API only, not in this file; see
+# handoff notes in the issue tracker). Each slot gets LLAMA_CTX_SIZE /
+# LLAMA_PARALLEL tokens of context — real sessions have hit ~66K tokens, so
+# don't drop LLAMA_CTX_SIZE without checking that per-slot number stays
+# comfortably above observed usage.
 LLAMA_PARALLEL=2
 
+# Dedicated CPU-only backend for qwen-code's tool-call harmfulness classifier
+# (fastModel in ~/.qwen/settings.json) — see docker-compose.yml's
+# qwen-classifier service comment for the why. 131072 ctx / q8_0 KV / Q4_K_M-
+# class weights ≈ 12GiB, fits gameserver's ~17GiB free system RAM with room
+# to spare.
+LLAMA_CLASSIFIER_MODEL_FILE=Qwen3-4B-Instruct-2507-UD-Q4_K_XL.gguf
+
 # --- Lazytainer ---
 # Seconds of inactivity before llama-server is stopped. 900 = 15 min.
 LAZYTAINER_INACTIVE_TIMEOUT=900

docker-compose.yml

@@ -46,6 +46,39 @@ services:
       - "lazytainer.group.llamaserver.inactiveTimeout=${LAZYTAINER_INACTIVE_TIMEOUT:-900}"
       - "lazytainer.group.llamaserver.minPacketThreshold=2"
 
+  # Dedicated backend for qwen-code's tool-call harmfulness classifier
+  # (fastModel in ~/.qwen/settings.json). Was aliased onto llama-server's own
+  # 27B connection — every classification call then queued behind whatever
+  # heavy generation was already running on that model's 2 GPU slots (issue
+  # tracker: OmniRoute semaphore/pr-agent investigation). CPU-only, own
+  # process, own queue: structurally can't contend with llama-server for a
+  # GPU slot. Needs >=131072 ctx (qwen-code requirement); Qwen3-4B-Instruct-2507
+  # is the smallest Qwen3 that supports that natively (262144) without
+  # RoPE-scaling — the smaller 0.6B/1.7B/4B (non-2507) models only go to
+  # 40960. Sized for gameserver's ~17GiB free RAM: q8_0/q8_0 KV at full
+  # 131072 ctx is ~9.8GiB + ~2.3GiB Q4_K_M-class weights ≈ 12GiB, comfortable
+  # headroom, and better KV quality than the q4_0 that would've been needed
+  # to squeeze this onto the GPU's ~6GiB free VRAM instead.
+  qwen-classifier:
+    image: ghcr.io/ggml-org/llama.cpp:server
+    container_name: qwen-classifier
+    volumes:
+      - models:/models
+    command: >
+      -m /models/${LLAMA_CLASSIFIER_MODEL_FILE:-Qwen3-4B-Instruct-2507-UD-Q4_K_XL.gguf}
+      --host 0.0.0.0
+      --port 8080
+      --n-gpu-layers 0
+      --ctx-size 131072
+      --parallel 1
+      --cache-type-k q8_0
+      --cache-type-v q8_0
+      --jinja
+    expose:
+      - "8080"
+    restart: unless-stopped
+    networks: [ai-stack]
+
   # ponytail: one-off downloader, not a standing service — run via
   # `docker compose --profile tools run --rm downloader`. Folded into
   # scripts/update.sh, which runs this every time; the `test -f` guard is
@@ -67,6 +100,22 @@ services:
         curl -L --fail --create-dirs -o /models/${LLAMA_MODEL_FILE:-Qwen3.8-27B-UD-Q4_K_XL.gguf}
         https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/resolve/main/${LLAMA_MODEL_FILE:-Qwen3.8-27B-UD-Q4_K_XL.gguf}
 
+  # Same pattern as downloader above, separate service so this one small
+  # file doesn't get re-checked/re-pulled by the big model's job.
+  downloader-classifier:
+    image: curlimages/curl:latest
+    profiles: ["tools"]
+    user: root
+    volumes:
+      - models:/models
+    entrypoint: ["sh", "-c"]
+    command:
+      - >
+        test -f /models/${LLAMA_CLASSIFIER_MODEL_FILE:-Qwen3-4B-Instruct-2507-UD-Q4_K_XL.gguf} &&
+        echo "already downloaded, skipping" ||
+        curl -L --fail --create-dirs -o /models/${LLAMA_CLASSIFIER_MODEL_FILE:-Qwen3-4B-Instruct-2507-UD-Q4_K_XL.gguf}
+        https://huggingface.co/unsloth/Qwen3-4B-Instruct-2507-GGUF/resolve/main/${LLAMA_CLASSIFIER_MODEL_FILE:-Qwen3-4B-Instruct-2507-UD-Q4_K_XL.gguf}
+
   # Fetches the three Qwen-Image FP8 files ComfyUI needs (diffusion model,
   # text encoder, VAE) — same test -f guard pattern as downloader above.
   # See docs/research/image-generation-model-choice.md and issue #42.

scripts/update.sh

@@ -233,6 +233,7 @@ docker compose build --pull
 echo "==> ensuring models are downloaded (skips already-present files)"
 docker compose --profile tools run --rm downloader
 docker compose --profile tools run --rm downloader-fast
+docker compose --profile tools run --rm downloader-classifier
 docker compose --profile tools run --rm downloader-comfyui
 
 echo "==> bringing up omniroute"