Qwen3-4B-AutoRound-FP8_BLOCK / quantization_config.json
fergusfinn's picture
AutoRound FP8_BLOCK of Qwen3-4B
09d57a9 verified
Raw
History Blame Contribute Delete
374 Bytes
{
"bits": 8,
"act_bits": 8,
"data_type": "fp",
"act_data_type": "fp",
"group_size": [
128,
128
],
"act_group_size": 128,
"sym": true,
"act_sym": true,
"act_dynamic": true,
"seqlen": 1024,
"autoround_version": "0.12.3",
"weight_block_size": [
128,
128
],
"quant_method": "fp8",
"fmt": "e4m3",
"activation_scheme": "dynamic"
}