qev_config.json

{
  "format": "qev",
  "format_version": 3,
  "family": "lfm2",
  "runtime": "mlx",
  "backend": "mlx_vlm",
  "model_name": "qev-230m",
  "foundation_architecture": "Lfm2ForCausalLM",
  "base_model": "LiquidAI/LFM2.5-230M",
  "base_revision": "40cb2ad3b3044d5a41eee083a6103c8b523afa45",
  "modalities": [
    "text"
  ],
  "native_generation": "adapter_disabled",
  "question_isolation": "independent_rows",
  "hidden_size": 1024,
  "pointer_dim": 256,
  "pointer_bias": false,
  "lora_rank": 64,
  "lora_alpha": 128,
  "lora_targets": [
    "q_proj",
    "k_proj",
    "v_proj",
    "out_proj",
    "in_proj",
    "w1",
    "w2",
    "w3"
  ],
  "max_length": 1024,
  "max_state": 384,
  "multimodal_max_length": 8192,
  "temperature": 4.5947934199881395,
  "training_label": "t230-a",
  "mlx_dtype": "bfloat16"
}