# A100 Configuration
gpu_type: a100
gpu_model: "NVIDIA A100 80GB"

# Default attention implementation
default_attention: flash_attention_2

# Pretraining defaults
pretrain:
  batch_size: 8
  sequence_length: 8192
  num_steps: 10
  warmup_steps: 3

# Inference defaults
inference:
  num_requests: 10
  prompt_length: 512
  generation_length: 100
  warmup_requests: 2

# Hardware specs (for reference)
hardware:
  memory_gb: 80
  tdp_watts: 400
  compute_capability: "8.0"