# Inference config for CALM-70B.
#
# Usage:
#   oumi infer -i -c ./oumi_infer.yaml
#

model:
  model_name: "uiuc-convai/CALM-70B"
  adapter_model: null  # Update for LoRA-tuned models.
  model_max_length: 2048  # Update based on your hardware constraints
  torch_dtype_str: "bfloat16"
  attn_implementation: "sdpa"
  load_pretrained_weights: True
  trust_remote_code: True

generation:
  max_new_tokens: 2048

engine: NATIVE