Remove obsolete v3 root Compose

docker-compose.yml

@@ -1,87 +0,0 @@
-# GLM-5.2-MXFP8-NVFP4-NF3-Hybrid v3.6 — daily-driver serving config.
-# 2026-07-09: damage-optimal expert allocation + BF16-stored non-expert tier
-# (single-hop online-MXFP8). Checkpoint re-sharded to 184x <=2GB files ->
-# fastsafetensors is reliable again (boot ~2 min).
-# 4x 96GB sm120 GPUs. Native mxfp8 tier + squeezed envelope -> 422,656-token
-# KV pool @ 262k max context (measured 2026-07-09 on 4x RTX PRO 6000 Max-Q).
-# If warmup OOMs on your box, drop --gpu-memory-utilization to 0.96 (~-20k pool).
-# mxfp8_tier_nokvb.json mount = the measured overlay convention (kv_b_proj +
-# shared_experts stay true BF16; this is the config every published number used).
-#   hf download madeby561/GLM-5.2-MXFP8-NVFP4-NF3-Hybrid --local-dir ./glm52-mxfp8
-#   MODEL_DIR=./glm52-mxfp8 docker compose up
-services:
-  glm52:
-    image: madeby561/vllm-glm52-nvfp4-nf3-hybrid:v3
-    network_mode: host
-    ipc: host
-    shm_size: 32gb
-    init: true
-    ulimits:
-      memlock: -1
-      stack: 67108864
-    deploy:
-      resources:
-        reservations:
-          devices:
-            - driver: nvidia
-              count: all
-              capabilities: [gpu]
-    volumes:
-      - ${MODEL_DIR:?set MODEL_DIR to the downloaded checkpoint dir}:/model:ro
-      - ${MODEL_DIR}/mxfp8_tier_nokvb.json:/opt/venv/lib/python3.12/site-packages/mxfp8_tier.json:ro
-      - vllm-cache:/cache
-    environment:
-      CUDA_VISIBLE_DEVICES: "0,1,2,3"
-      CUDA_DEVICE_ORDER: PCI_BUS_ID
-      CUDA_DEVICE_MAX_CONNECTIONS: "32"
-      CUTE_DSL_ARCH: sm_120a
-      OMP_NUM_THREADS: "16"
-      PYTORCH_CUDA_ALLOC_CONF: expandable_segments:True
-      SAFETENSORS_FAST_GPU: "1"
-      NCCL_IB_DISABLE: "1"
-      NCCL_P2P_LEVEL: SYS
-      NCCL_PROTO: LL,LL128,Simple
-      VLLM_USE_B12X_FP8_GEMM: "1"
-      VLLM_USE_B12X_MOE: "1"
-      VLLM_USE_B12X_SPARSE_INDEXER: "1"
-      VLLM_USE_V2_MODEL_RUNNER: "1"
-      B12X_W4A16_TC_DECODE: "1"
-      B12X_MOE_FORCE_A16: "1"
-      VLLM_DCP_GLOBAL_TOPK: "1"
-      VLLM_DCP_SHARD_DRAFT: "1"
-      VLLM_ENABLE_PCIE_ALLREDUCE: "1"
-      VLLM_PCIE_ALLREDUCE_BACKEND: b12x
-      VLLM_PCIE_ONESHOT_ALLREDUCE_MAX_SIZE: 64KB
-      # v3 image bakes env-tunable AR crossovers (CARPATCH): this routing
-      # puts decode-step all-reduces on the fast paths -> ~95-101 t/s
-      # single-stream decode (vs 67-74 on stock crossovers).
-      VLLM_PCIE_ONESHOT_MAX_BYTES: "65536"
-      B12X_DENSE_SPLITK_TURBO: "1"
-      XDG_CACHE_HOME: /cache/jit
-      TRITON_CACHE_DIR: /cache/jit/triton
-      HYBRID_TIER: both
-      HYBRID_KEPT: b12x_nf3
-      HYBRID_NF3: b12x_nf3
-      HYBRID_B12X_MAX_TOKENS: "2048"
-      HYBRID_MXFP8_NATIVE: "1"
-    command: >
-      vllm serve /model
-      --served-model-name GLM-5.2 --host 0.0.0.0 --port 5001
-      --trust-remote-code --tensor-parallel-size 4
-      --decode-context-parallel-size 4 --dcp-comm-backend ag_rs --dcp-kv-cache-interleave-size 1
-      --kv-cache-dtype fp8
-      --attention-backend B12X_MLA_SPARSE
-      --moe-backend b12x
-      --load-format fastsafetensors
-      -cc.pass_config.fuse_allreduce_rms=True
-      --gpu-memory-utilization 0.968
-      --max-model-len 262144 --max-num-seqs 8 --max-num-batched-tokens 2048
-      --max-cudagraph-capture-size 64
-      --async-scheduling --enable-chunked-prefill --enable-prefix-caching
-      --enable-auto-tool-choice --tool-call-parser glm47
-      --reasoning-parser glm45
-      --default-chat-template-kwargs '{"reasoning_effort":"high"}'
-      --hf-overrides '{"use_index_cache":true,"index_topk_pattern":"FFFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSSFSSS"}'
-      --speculative-config '{"method":"mtp","num_speculative_tokens":5,"moe_backend":"b12x","draft_sample_method":"probabilistic"}'
-volumes:
-  vllm-cache: