gpt-sw3-6.7b-v2-instruct-4bit-gptq / quantize_config.json
Ekgren's picture
Create quantize_config.json
6f72481
{
"bits": 4,
"block_name_to_quantize": "transformer.h",
"damp_percent": 0.1,
"dataset": "c4",
"desc_act": false,
"disable_exllama": true,
"group_size": 128,
"max_input_length": null,
"model_seqlen": 2048,
"module_name_preceding_first_block": [
"transformer.wte",
"transformer.wpe",
"transformer.drop"
],
"pad_token_id": null,
"quant_method": "gptq",
"sym": true,
"tokenizer": null,
"true_sequential": true,
"use_cuda_fp16": true
}