qwen3.5-122b-awq.yaml

# Recipe: Qwen3.5-122B-A10B-AWQ
# Qwen3.5-122B model in AWQ quantization

recipe_version: "1"
name: Qwen3.5-122B-AWQ
description: vLLM serving Qwen3.5-122B-AWQ

# HuggingFace model to download (optional, for --download-model)
model: QuantTrio/Qwen3.5-122B-A10B-AWQ

# Only cluster is supported
cluster_only: false

# Container image to use
container: vllm-node

# No mods required
mods: {}
#  - mods/fix-qwen3.5-chat-template

# Default settings (can be overridden via CLI)
defaults:
  port: 8000
  host: 0.0.0.0
  tensor_parallel: 1
  gpu_memory_utilization: 0.9
  max_model_len: 150000
  max_num_batched_tokens: 32768

# Environment variables
env: {}

# The vLLM serve command template
command: |
  vllm serve QuantTrio/Qwen3.5-122B-A10B-AWQ \
    --max-model-len {max_model_len} \
    --gpu-memory-utilization {gpu_memory_utilization} \
    --port {port} \
    --host {host} \
    --load-format fastsafetensors \
    --kv-cache-dtype fp8_e4m3 \
    --trust-remote-code \
    --language-model-only \
    --enable-prefix-caching \
    --enable-auto-tool-choice \
    --tool-call-parser qwen3_coder \
    --reasoning-parser qwen3 \
    --speculative-config '{{"method":"mtp","num_speculative_tokens":3}}' \
    --max-num-batched-tokens {max_num_batched_tokens} \
    --max-num-seqs 8 \
    --stream-interval 5