quantize_config.json

{
  "bits": 4,
  "group_size": 128,
  "sym": true,
  "data_type": "fp8_gaudi3_to_int_sym",
  "enable_quanted_input": true,
  "enable_minmax_tuning": true,
  "seqlen": 512,
  "batch_size": 4,
  "scale_dtype": "torch.bfloat16",
  "lr": 0.005,
  "minmax_lr": 0.005,
  "gradient_accumulate_steps": 1,
  "iters": 50,
  "amp": true,
  "nsamples": 512,
  "low_gpu_mem_usage": true,
  "to_quant_block_names": null,
  "enable_norm_bias_tuning": false,
  "autoround_version": "0.4.5",
  "quant_method": "gptq",
  "desc_act": false,
  "true_sequential": false,
  "damp_percent": 0.01
}