vllm serve llm-model-lab/thinking-v2-3epoch \
  --enable-lora \
  --lora-modules epoch1=llm-model-lab/gemma3-27b-orpo-thinking-v2-checkpoint-571 \
  --max-lora-rank 128 \
  --dtype bfloat16

Files

This checkpoint contains:

adapter_model.safetensors: LoRA weights
adapter_config.json: PEFT configuration
tokenizer.json, tokenizer_config.json: Tokenizer
chat_template.jinja: Gemma3 chat template
DeepSpeed training state files (for training resume only)