fix(llm): shrink qwen-classifier's compute buffer for real VRAM headroom

Measured live: full-offload weights+KV (~4.9GiB estimate) actually used
~5.85GiB, leaving only ~700MB free on the R9700 — too tight, real OOM
risk for either GPU process. The gap was compute-buffer/graph overhead
the naive estimate didn't account for. Drop --batch-size/--ubatch-size
well below llama-server's defaults (2048/512) to shrink it — a
single-request classifier has no batching throughput to lose.

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

docker-compose.yml

@@ -66,12 +66,17 @@ services:
   # 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)
+  # VRAM: weights+KV math (~4.9GiB) predicted comfortable headroom in the
-  # ≈ 4.9GiB, fits fully on GPU (--n-gpu-layers 999) inside the ~6.1GiB free
+  # ~6.1GiB free on the R9700, but measured live it actually used ~5.85GiB —
-  # on the R9700 after llama-server's 27B model, with ~1.2GiB headroom.
+  # left only ~700MB free, too tight (compute-buffer/graph overhead the
-  # GPU_MAX_HW_QUEUES=1 carried over from llama-server's comment above —
+  # naive weights+KV estimate didn't account for). --batch-size/--ubatch-size
-  # same ROCm/ROCm#5706 clock-pinning bug applies now that two HIP contexts
+  # dropped well below llama-server's defaults (2048/512) to shrink that
-  # (this + llama-server) share the card.
+  # buffer — this classifier processes one request at a time, no batching
+  # throughput to lose. --n-gpu-layers 999 (full offload) kept since the
+  # ~1.5GB freed here should cover it; re-verify with rocm-smi after deploy
+  # before trusting this number again. 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-rocm
     container_name: qwen-classifier
@@ -95,6 +100,8 @@ services:
       --n-gpu-layers 999
       --ctx-size 65536
       --parallel 1
+      --batch-size 512
+      --ubatch-size 128
       --cache-type-k q4_0
       --cache-type-v q4_0
       --jinja