qwen3-6-lora/docker-compose.eval.yml

services:
  vllm-eval:
    image: vllm/vllm-openai:cu130-nightly-aarch64
    container_name: vllm-eval
    restart: "no"
    ipc: host
    ports:
      - "8001:8000"
    volumes:
      # Fase 6.4.25b: apunta al candidato v2b-bf16 (base + LoRA #1 + LoRA #2
      # corregido con los 13 seeds del diagnostico de causa raiz), no al v2-bf16
      # original (que fallo la puerta 5 modo full por el bug de formato de
      # tool-call). Es el checkpoint contra el que se remide la puerta 5 tras la
      # iteracion correctiva, antes de decidir NVFP4 vs fallback.
      - /home/aleleba/ft-models/Qwen3.6-35B-A3B-mcp-v2b-bf16:/model:ro
    command:
      - "--model=/model"
      - "--served-model-name=qwen3.6-35b-a3b-mcp-v2b-bf16"
      - "--tensor-parallel-size=1"
      - "--max-model-len=32768"
      - "--enable-auto-tool-choice"
      - "--tool-call-parser=qwen3_coder"
      - "--reasoning-parser=qwen3"
      - "--default-chat-template-kwargs={\"preserve_thinking\": true}"
      - "--trust-remote-code"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

  # Fase 5: clona 1:1 el docker-compose.yml real de produccion (vllm-qwen36, ver
  # PLAN.md) -- mismos flags de vLLM (incluido --speculative-config real, la
  # primera vez que se prueba en este proyecto), cambiando solo container_name,
  # puerto (8002, produccion usa 8000 y el vllm-eval de Fase 4 usa 8001), volumen
  # (checkpoint NVFP4 de Fase 5 en vez de RedHatAI--Qwen3.6-35B-A3B-NVFP4),
  # --model/--served-model-name, y restart: "no". Nunca se toca vllm-qwen36 ni su
  # compose real de Portainer -- este es un contenedor nuevo y propio del repo.
  vllm-eval-nvfp4:
    image: vllm/vllm-openai:cu130-nightly-aarch64
    container_name: vllm-eval-nvfp4
    restart: "no"
    runtime: nvidia
    environment:
      NVIDIA_VISIBLE_DEVICES: all
      NVIDIA_DRIVER_CAPABILITIES: compute,utility
    ports:
      - "8002:8000"
    ipc: host
    ulimits:
      memlock: -1
      stack: 67108864
    volumes:
      - /home/aleleba/ft-models/Qwen3.6-35B-A3B-mcp-NVFP4:/models/Qwen3.6-35B-A3B-mcp-NVFP4:ro
    command: >
      --model /models/Qwen3.6-35B-A3B-mcp-NVFP4
      --served-model-name qwen3.6-35b-a3b-mcp-nvfp4
      --host 0.0.0.0
      --port 8000
      --tensor-parallel-size 1
      --trust-remote-code
      --quantization compressed-tensors
      --moe-backend flashinfer_cutlass
      --kv-cache-dtype fp8_e4m3
      --gpu-memory-utilization 0.45
      --max-model-len 524288
      --max-num-seqs 8
      --max-num-batched-tokens 32768
      --enable-chunked-prefill
      --enable-prefix-caching
      --speculative-config '{"method":"mtp","num_speculative_tokens":1}'
      --reasoning-parser qwen3
      --tool-call-parser qwen3_coder
      --enable-auto-tool-choice
      --default-chat-template-kwargs '{"preserve_thinking":true}'
      --limit-mm-per-prompt '{"image":4,"video":0,"audio":0}'
      --generation-config vllm
      --override-generation-config '{"temperature":0.6,"top_p":0.80,"top_k":20,"presence_penalty":0.0,"repetition_penalty":1.0}'
      --hf-overrides '{"text_config":{"rope_scaling":{"rope_type":"yarn","factor":2.0,"original_max_position_embeddings":262144}}}'
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 600s

  # Fase 6 -- candidato v2 (base + LoRA #1 + LoRA #2 de diseno en Penpot, mergeado
  # y recuantizado a NVFP4). Clon 1:1 de vllm-eval-nvfp4, que a su vez clona el
  # compose real de produccion, incluido --speculative-config: es el contenedor
  # contra el que se corren las puertas 2/3/4/5 del candidato, y solo sirve si
  # replica exactamente los flags con los que se va a servir. Respecto de
  # vllm-eval-nvfp4 cambian SOLO container_name, puerto (8004; 8000 produccion,
  # 8001 Fase 4, 8002/8003 Fase 5), volumen, --model y --served-model-name.
  # Nunca se toca vllm-qwen36 ni su compose real de Portainer.
  vllm-eval-nvfp4-v2:
    image: vllm/vllm-openai:cu130-nightly-aarch64
    container_name: vllm-eval-nvfp4-v2
    restart: "no"
    runtime: nvidia
    environment:
      NVIDIA_VISIBLE_DEVICES: all
      NVIDIA_DRIVER_CAPABILITIES: compute,utility
    ports:
      - "8004:8000"
    ipc: host
    ulimits:
      memlock: -1
      stack: 67108864
    volumes:
      - /home/aleleba/ft-models/Qwen3.6-35B-A3B-mcp-v2-NVFP4:/models/Qwen3.6-35B-A3B-mcp-v2-NVFP4:ro
    command: >
      --model /models/Qwen3.6-35B-A3B-mcp-v2-NVFP4
      --served-model-name qwen3.6-35b-a3b-mcp-v2-nvfp4
      --host 0.0.0.0
      --port 8000
      --tensor-parallel-size 1
      --trust-remote-code
      --quantization compressed-tensors
      --moe-backend flashinfer_cutlass
      --kv-cache-dtype fp8_e4m3
      --gpu-memory-utilization 0.45
      --max-model-len 524288
      --max-num-seqs 8
      --max-num-batched-tokens 32768
      --enable-chunked-prefill
      --enable-prefix-caching
      --speculative-config '{"method":"mtp","num_speculative_tokens":1}'
      --reasoning-parser qwen3
      --tool-call-parser qwen3_coder
      --enable-auto-tool-choice
      --default-chat-template-kwargs '{"preserve_thinking":true}'
      --limit-mm-per-prompt '{"image":4,"video":0,"audio":0}'
      --generation-config vllm
      --override-generation-config '{"temperature":0.6,"top_p":0.80,"top_k":20,"presence_penalty":0.0,"repetition_penalty":1.0}'
      --hf-overrides '{"text_config":{"rope_scaling":{"rope_type":"yarn","factor":2.0,"original_max_position_embeddings":262144}}}'
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 600s

  # Fase 6 -- diagnostico de aislamiento del drift del head de MTP (riesgo #10):
  # identico a vllm-eval-nvfp4-v2 pero SIN --speculative-config. El draft head de
  # MTP se copia del linaje base y nunca se fine-tunea, mientras que el target
  # model ya derivo DOS veces (LoRA #1 y LoRA #2), asi que la tasa de aceptacion
  # del speculative decoding puede caer y degradar la calidad servida sin que la
  # cuantizacion ni el dataset tengan nada que ver. Correr las mismas puertas en
  # 8004 (spec) y 8005 (nospec) separa las dos causas. Es un servicio de
  # diagnostico: no clona produccion 1:1 a proposito (esa es justo la variable
  # que se esta aislando) y los flags de produccion NUNCA se cambian por esto,
  # solo se reporta el hallazgo.
  vllm-eval-nvfp4-v2-nospec:
    image: vllm/vllm-openai:cu130-nightly-aarch64
    container_name: vllm-eval-nvfp4-v2-nospec
    restart: "no"
    runtime: nvidia
    environment:
      NVIDIA_VISIBLE_DEVICES: all
      NVIDIA_DRIVER_CAPABILITIES: compute,utility
    ports:
      - "8005:8000"
    ipc: host
    ulimits:
      memlock: -1
      stack: 67108864
    volumes:
      - /home/aleleba/ft-models/Qwen3.6-35B-A3B-mcp-v2-NVFP4:/models/Qwen3.6-35B-A3B-mcp-v2-NVFP4:ro
    command: >
      --model /models/Qwen3.6-35B-A3B-mcp-v2-NVFP4
      --served-model-name qwen3.6-35b-a3b-mcp-v2-nvfp4-nospec
      --host 0.0.0.0
      --port 8000
      --tensor-parallel-size 1
      --trust-remote-code
      --quantization compressed-tensors
      --moe-backend flashinfer_cutlass
      --kv-cache-dtype fp8_e4m3
      --gpu-memory-utilization 0.45
      --max-model-len 524288
      --max-num-seqs 8
      --max-num-batched-tokens 32768
      --enable-chunked-prefill
      --enable-prefix-caching
      --reasoning-parser qwen3
      --tool-call-parser qwen3_coder
      --enable-auto-tool-choice
      --default-chat-template-kwargs '{"preserve_thinking":true}'
      --limit-mm-per-prompt '{"image":4,"video":0,"audio":0}'
      --generation-config vllm
      --override-generation-config '{"temperature":0.6,"top_p":0.80,"top_k":20,"presence_penalty":0.0,"repetition_penalty":1.0}'
      --hf-overrides '{"text_config":{"rope_scaling":{"rope_type":"yarn","factor":2.0,"original_max_position_embeddings":262144}}}'
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 600s

  # Fase 5 -- diagnostico de aislamiento: identico a vllm-eval-nvfp4 pero SIN
  # --speculative-config, para determinar si la regresion de calidad observada
  # en las puertas 2-3 (vs. Fase 4) viene del speculative decoding (MTP) o de
  # la cuantizacion NVFP4 en si. Puerto 8003 (distinto de 8000 produccion, 8001
  # Fase 4, 8002 Fase 5 con speculative real). Servicio temporal de diagnostico,
  # no clona produccion 1:1 a proposito (esa es la variable que se esta aislando).
  vllm-eval-nvfp4-nospec:
    image: vllm/vllm-openai:cu130-nightly-aarch64
    container_name: vllm-eval-nvfp4-nospec
    restart: "no"
    runtime: nvidia
    environment:
      NVIDIA_VISIBLE_DEVICES: all
      NVIDIA_DRIVER_CAPABILITIES: compute,utility
    ports:
      - "8003:8000"
    ipc: host
    ulimits:
      memlock: -1
      stack: 67108864
    volumes:
      - /home/aleleba/ft-models/Qwen3.6-35B-A3B-mcp-NVFP4:/models/Qwen3.6-35B-A3B-mcp-NVFP4:ro
    command: >
      --model /models/Qwen3.6-35B-A3B-mcp-NVFP4
      --served-model-name qwen3.6-35b-a3b-mcp-nvfp4-nospec
      --host 0.0.0.0
      --port 8000
      --tensor-parallel-size 1
      --trust-remote-code
      --quantization compressed-tensors
      --moe-backend flashinfer_cutlass
      --kv-cache-dtype fp8_e4m3
      --gpu-memory-utilization 0.45
      --max-model-len 524288
      --max-num-seqs 8
      --max-num-batched-tokens 32768
      --enable-chunked-prefill
      --enable-prefix-caching
      --reasoning-parser qwen3
      --tool-call-parser qwen3_coder
      --enable-auto-tool-choice
      --default-chat-template-kwargs '{"preserve_thinking":true}'
      --limit-mm-per-prompt '{"image":4,"video":0,"audio":0}'
      --generation-config vllm
      --override-generation-config '{"temperature":0.6,"top_p":0.80,"top_k":20,"presence_penalty":0.0,"repetition_penalty":1.0}'
      --hf-overrides '{"text_config":{"rope_scaling":{"rope_type":"yarn","factor":2.0,"original_max_position_embeddings":262144}}}'
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 600s