OPEA
/

Safetensors
llama
4-bit precision
awq
Falcon3-3B-Base-int4-sym-awq-inc / quantization_config.json
n1ck-guo's picture
upload awq format
277ace4
raw
history blame
615 Bytes
{
"bits": 4,
"group_size": 128,
"sym": true,
"data_type": "int",
"enable_quanted_input": true,
"enable_minmax_tuning": true,
"seqlen": 2048,
"batch_size": 8,
"scale_dtype": "torch.float16",
"lr": 0.0005,
"minmax_lr": 0.0005,
"gradient_accumulate_steps": 1,
"iters": 2000,
"amp": true,
"nsamples": 512,
"low_gpu_mem_usage": false,
"to_quant_block_names": null,
"enable_norm_bias_tuning": false,
"dataset": "NeelNanda/pile-10k",
"autoround_version": "0.4.2",
"quant_method": "awq",
"zero_point": false,
"version": "gemm",
"modules_to_not_convert": [
"lm_head"
]
}