Codellama-2-7b-Miniguanaco-Mistral-AR-Q5_K_M.gguf$:
  loader: llama.cpp
  cpu: false
  threads: 0
  threads_batch: 0
  n_batch: 512
  no_mmap: true
  mlock: true
  mul_mat_q: true
  n_gpu_layers: 28
  tensor_split: ''
  n_ctx: 16384
  llama_cpp_seed: 0.0
  compress_pos_emb: 1
  alpha_value: 1
  rope_freq_base: 1000000
  numa: true