llama-2-7b-rlmpq-balanced/rlmpq_policy.json

{
  "slug": "llama_2_7b",
  "hf_id": "meta-llama/Llama-2-7b-hf",
  "scenario": "Balanced",
  "params": {
    "lambda_mse": 1.0,
    "lambda_bit": 0.25,
    "max_mse_clip": 1.0,
    "target_bits": 4.5,
    "lambda_budget": 5.0
  },
  "policy": [
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    4,
    8,
    8,
    8
  ],
  "num_layers": 32,
  "avg_bits": 4.375,
  "total_reward": -4.832758,
  "compression_ratio": 3.6571,
  "approx_size_gb": 3.828,
  "bit_distribution": {
    "4": 29,
    "8": 3
  },
  "avg_mse_per_bit": {
    "4": 0.28526766,
    "8": 3.942e-05
  },
  "validation_elapsed_s": 0.0583
}