lvkaokao's picture
Upload quantized model FLUX.2-dev-MXFP4-RTN-AutoRound
59fb46e verified
Raw History Blame Contribute Delete
1.12 kB
{
"_class_name": "Flux2Transformer2DModel",
"_diffusers_version": "0.36.0.dev0",
"_name_or_path": "/models/FLUX.2-dev/transformer",
"_use_default_values": [
"guidance_embeds"
],
"attention_head_dim": 128,
"axes_dims_rope": [
32,
32,
32,
32
],
"eps": 1e-06,
"guidance_embeds": true,
"in_channels": 128,
"joint_attention_dim": 15360,
"mlp_ratio": 3.0,
"num_attention_heads": 48,
"num_layers": 8,
"num_single_layers": 48,
"out_channels": null,
"patch_size": 1,
"quantization_config": {
"act_bits": 4,
"act_data_type": "mx_fp",
"act_dynamic": true,
"act_group_size": 32,
"act_sym": true,
"autoround_version": "0.15.0",
"bits": 4,
"block_name_to_quantize": "transformer_blocks,single_transformer_blocks",
"data_type": "mx_fp",
"enable_quanted_input": false,
"group_size": 32,
"packing_format": "auto_round:llm_compressor",
"quant_method": "auto-round",
"static_attention_granularity": "tensor",
"static_kv_granularity": "tensor",
"sym": true
},
"rope_theta": 2000,
"timestep_guidance_channels": 256
}