fix(llm): move qwen-classifier to GPU, right-size context

CPU-only was too slow in practice: real classification calls blew past
OmniRoute's 60s timeout and retry-looped (504/499). Moved to GPU.

Also traced qwen-code's actual classifier transcript cap in its source
(MAX_TRANSCRIPT_MESSAGES=40, MAX_HISTORICAL_ACTION_CHARS=4000/message) —
worst case is ~40-50K tokens, not the 131072 originally set in
settings.json (copied from the main model's entry, not a real qwen-code
requirement). Dropped ctx-size to 65536 (~1.5x margin) so Q4_K_XL
weights + q4_0/q4_0 KV fit fully on GPU (~4.9GiB) inside the ~6.1GiB
free on the R9700, instead of needing partial CPU/GPU offload.

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

docker-compose.yml

@@ -50,31 +50,50 @@ services:
   # (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. Reuses a Q8_K_XL GGUF already sitting in the models volume from
-  # something earlier (better weight quality than the Q4_K_M-class file
-  # originally specced here, no download needed). KV cache dropped to
-  # q4_0/q4_0 to compensate: full-context q8_0/q8_0 (~9.6GiB) on top of the
-  # larger Q8 weights (~4.7GiB) left too little slack against the rest of
-  # the stack (omniroute's 10g mem_limit, qdrant, neo4j) on gameserver's
-  # 31GiB total RAM; q4_0/q4_0 (~5.1GiB) + weights (~4.7GiB) ≈ 9.8GiB leaves
-  # comfortable headroom instead. Weight precision matters more than KV
-  # precision for a classification task, so this trade favors the weights.
+  # tracker: OmniRoute semaphore/pr-agent investigation).
+  #
+  # Tried CPU-only first (own process avoids the GPU queue entirely) — too
+  # slow in practice: real classification calls blew past OmniRoute's 60s
+  # timeout and retry-looped (504→499→504...). Moved to GPU instead.
+  #
+  # Context sizing: qwen-code's classifier transcript is hard-capped in its
+  # own source (MAX_TRANSCRIPT_MESSAGES=40, MAX_HISTORICAL_ACTION_CHARS=4000
+  # per message, packages/core/src/permissions/classifier-transcript.ts) —
+  # worst case is ~40-50K tokens, nowhere near the 131072 originally set in
+  # settings.json (that number was copied from the main model's entry, not
+  # a real qwen-code requirement). 65536 ctx gives ~1.5x margin over that
+  # worst case. Qwen3-4B-Instruct-2507 is still the model choice — smallest
+  # Qwen3 with long native context (262144) without RoPE-scaling, in case
+  # that margin ever needs to grow.
+  #
+  # VRAM: Q4_K_XL weights (~2.37GiB) + q4_0/q4_0 KV at 65536 ctx (~2.53GiB)
+  # ≈ 4.9GiB, fits fully on GPU (--n-gpu-layers 999) inside the ~6.1GiB free
+  # on the R9700 after llama-server's 27B model, with ~1.2GiB headroom.
+  # GPU_MAX_HW_QUEUES=1 carried over from llama-server's comment above —
+  # same ROCm/ROCm#5706 clock-pinning bug applies now that two HIP contexts
+  # (this + llama-server) share the card.
   qwen-classifier:
-    image: ghcr.io/ggml-org/llama.cpp:server
+    image: ghcr.io/ggml-org/llama.cpp:server-rocm
     container_name: qwen-classifier
+    devices:
+      - /dev/kfd
+      - /dev/dri
+    group_add:
+      - "${HOST_VIDEO_GID:?run scripts/update.sh first to resolve this}"
+      - "${HOST_RENDER_GID:?run scripts/update.sh first to resolve this}"
+    security_opt:
+      - seccomp=unconfined
+    ipc: host
+    environment:
+      - GPU_MAX_HW_QUEUES=1
     volumes:
       - models:/models
     command: >
-      -m /models/${LLAMA_CLASSIFIER_MODEL_FILE:-Qwen3-4B-Instruct-2507-UD-Q8_K_XL.gguf}
+      -m /models/${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
+      --n-gpu-layers 999
+      --ctx-size 65536
       --parallel 1
       --cache-type-k q4_0
       --cache-type-v q4_0