EdgeDiffusion / mp_quant_metadata.json
ChenHe727's picture
Upload per-layer bit-width assignment metadata
862bcab verified
{
"distill_step": 10000,
"distill_loss": 1.0463419775873422,
"is_ema": true,
"quantization": {
"method": "mixed_precision_gptq",
"group_size": 128,
"damping": 0.01,
"assignment": {
"down_blocks.0.attentions.0.proj_in": "int4",
"down_blocks.0.attentions.0.proj_out": "int4",
"down_blocks.0.attentions.0.transformer_blocks.0.attn1.to_k": "int4",
"down_blocks.0.attentions.0.transformer_blocks.0.attn1.to_out.0": "int4",
"down_blocks.0.attentions.0.transformer_blocks.0.attn1.to_q": "int4",
"down_blocks.0.attentions.0.transformer_blocks.0.attn1.to_v": "int4",
"down_blocks.0.attentions.0.transformer_blocks.0.attn2.to_k": "int8",
"down_blocks.0.attentions.0.transformer_blocks.0.attn2.to_out.0": "int8",
"down_blocks.0.attentions.0.transformer_blocks.0.attn2.to_q": "int4",
"down_blocks.0.attentions.0.transformer_blocks.0.attn2.to_v": "int8",
"down_blocks.0.attentions.0.transformer_blocks.0.ff.net.0.proj": "int4",
"down_blocks.0.attentions.0.transformer_blocks.0.ff.net.2": "fp16",
"down_blocks.0.attentions.1.proj_in": "int4",
"down_blocks.0.attentions.1.proj_out": "int4",
"down_blocks.0.attentions.1.transformer_blocks.0.attn1.to_k": "int4",
"down_blocks.0.attentions.1.transformer_blocks.0.attn1.to_out.0": "int4",
"down_blocks.0.attentions.1.transformer_blocks.0.attn1.to_q": "int4",
"down_blocks.0.attentions.1.transformer_blocks.0.attn1.to_v": "int4",
"down_blocks.0.attentions.1.transformer_blocks.0.attn2.to_k": "int8",
"down_blocks.0.attentions.1.transformer_blocks.0.attn2.to_out.0": "int8",
"down_blocks.0.attentions.1.transformer_blocks.0.attn2.to_q": "int4",
"down_blocks.0.attentions.1.transformer_blocks.0.attn2.to_v": "int8",
"down_blocks.0.attentions.1.transformer_blocks.0.ff.net.0.proj": "int4",
"down_blocks.0.attentions.1.transformer_blocks.0.ff.net.2": "int8",
"down_blocks.1.attentions.0.proj_in": "fp16",
"down_blocks.1.attentions.0.proj_out": "fp16",
"down_blocks.1.attentions.0.transformer_blocks.0.attn1.to_k": "int8",
"down_blocks.1.attentions.0.transformer_blocks.0.attn1.to_out.0": "fp16",
"down_blocks.1.attentions.0.transformer_blocks.0.attn1.to_q": "int8",
"down_blocks.1.attentions.0.transformer_blocks.0.attn1.to_v": "fp16",
"down_blocks.1.attentions.0.transformer_blocks.0.attn2.to_k": "int8",
"down_blocks.1.attentions.0.transformer_blocks.0.attn2.to_out.0": "int8",
"down_blocks.1.attentions.0.transformer_blocks.0.attn2.to_q": "int8",
"down_blocks.1.attentions.0.transformer_blocks.0.attn2.to_v": "int8",
"down_blocks.1.attentions.0.transformer_blocks.0.ff.net.0.proj": "fp16",
"down_blocks.1.attentions.0.transformer_blocks.0.ff.net.2": "fp16",
"down_blocks.1.attentions.1.proj_in": "fp16",
"down_blocks.1.attentions.1.proj_out": "fp16",
"down_blocks.1.attentions.1.transformer_blocks.0.attn1.to_k": "int8",
"down_blocks.1.attentions.1.transformer_blocks.0.attn1.to_out.0": "int8",
"down_blocks.1.attentions.1.transformer_blocks.0.attn1.to_q": "int8",
"down_blocks.1.attentions.1.transformer_blocks.0.attn1.to_v": "int8",
"down_blocks.1.attentions.1.transformer_blocks.0.attn2.to_k": "int8",
"down_blocks.1.attentions.1.transformer_blocks.0.attn2.to_out.0": "int8",
"down_blocks.1.attentions.1.transformer_blocks.0.attn2.to_q": "int8",
"down_blocks.1.attentions.1.transformer_blocks.0.attn2.to_v": "int8",
"down_blocks.1.attentions.1.transformer_blocks.0.ff.net.0.proj": "int8",
"down_blocks.1.attentions.1.transformer_blocks.0.ff.net.2": "int8",
"down_blocks.2.attentions.0.proj_in": "int8",
"down_blocks.2.attentions.0.proj_out": "int8",
"down_blocks.2.attentions.0.transformer_blocks.0.attn1.to_k": "int8",
"down_blocks.2.attentions.0.transformer_blocks.0.attn1.to_out.0": "int8",
"down_blocks.2.attentions.0.transformer_blocks.0.attn1.to_q": "int8",
"down_blocks.2.attentions.0.transformer_blocks.0.attn1.to_v": "int8",
"down_blocks.2.attentions.0.transformer_blocks.0.attn2.to_k": "int8",
"down_blocks.2.attentions.0.transformer_blocks.0.attn2.to_out.0": "int4",
"down_blocks.2.attentions.0.transformer_blocks.0.attn2.to_q": "int8",
"down_blocks.2.attentions.0.transformer_blocks.0.attn2.to_v": "int8",
"down_blocks.2.attentions.0.transformer_blocks.0.ff.net.0.proj": "fp16",
"down_blocks.2.attentions.0.transformer_blocks.0.ff.net.2": "int8",
"down_blocks.2.attentions.1.proj_in": "int8",
"down_blocks.2.attentions.1.proj_out": "int8",
"down_blocks.2.attentions.1.transformer_blocks.0.attn1.to_k": "int8",
"down_blocks.2.attentions.1.transformer_blocks.0.attn1.to_out.0": "int8",
"down_blocks.2.attentions.1.transformer_blocks.0.attn1.to_q": "int8",
"down_blocks.2.attentions.1.transformer_blocks.0.attn1.to_v": "int8",
"down_blocks.2.attentions.1.transformer_blocks.0.attn2.to_k": "int8",
"down_blocks.2.attentions.1.transformer_blocks.0.attn2.to_out.0": "int4",
"down_blocks.2.attentions.1.transformer_blocks.0.attn2.to_q": "int8",
"down_blocks.2.attentions.1.transformer_blocks.0.attn2.to_v": "int8",
"down_blocks.2.attentions.1.transformer_blocks.0.ff.net.0.proj": "int8",
"down_blocks.2.attentions.1.transformer_blocks.0.ff.net.2": "int8",
"mid_block.attentions.0.proj_in": "int8",
"mid_block.attentions.0.proj_out": "int8",
"mid_block.attentions.0.transformer_blocks.0.attn1.to_k": "int8",
"mid_block.attentions.0.transformer_blocks.0.attn1.to_out.0": "int8",
"mid_block.attentions.0.transformer_blocks.0.attn1.to_q": "int8",
"mid_block.attentions.0.transformer_blocks.0.attn1.to_v": "int8",
"mid_block.attentions.0.transformer_blocks.0.attn2.to_k": "int8",
"mid_block.attentions.0.transformer_blocks.0.attn2.to_out.0": "int8",
"mid_block.attentions.0.transformer_blocks.0.attn2.to_q": "int8",
"mid_block.attentions.0.transformer_blocks.0.attn2.to_v": "int8",
"mid_block.attentions.0.transformer_blocks.0.ff.net.0.proj": "int8",
"mid_block.attentions.0.transformer_blocks.0.ff.net.2": "int8",
"up_blocks.1.attentions.0.proj_in": "int8",
"up_blocks.1.attentions.0.proj_out": "int8",
"up_blocks.1.attentions.0.transformer_blocks.0.attn1.to_k": "int8",
"up_blocks.1.attentions.0.transformer_blocks.0.attn1.to_out.0": "int8",
"up_blocks.1.attentions.0.transformer_blocks.0.attn1.to_q": "int8",
"up_blocks.1.attentions.0.transformer_blocks.0.attn1.to_v": "int8",
"up_blocks.1.attentions.0.transformer_blocks.0.attn2.to_k": "int8",
"up_blocks.1.attentions.0.transformer_blocks.0.attn2.to_out.0": "int4",
"up_blocks.1.attentions.0.transformer_blocks.0.attn2.to_q": "int8",
"up_blocks.1.attentions.0.transformer_blocks.0.attn2.to_v": "int8",
"up_blocks.1.attentions.0.transformer_blocks.0.ff.net.0.proj": "int8",
"up_blocks.1.attentions.0.transformer_blocks.0.ff.net.2": "int8",
"up_blocks.1.attentions.1.proj_in": "int8",
"up_blocks.1.attentions.1.proj_out": "int8",
"up_blocks.1.attentions.1.transformer_blocks.0.attn1.to_k": "int8",
"up_blocks.1.attentions.1.transformer_blocks.0.attn1.to_out.0": "int8",
"up_blocks.1.attentions.1.transformer_blocks.0.attn1.to_q": "int8",
"up_blocks.1.attentions.1.transformer_blocks.0.attn1.to_v": "int8",
"up_blocks.1.attentions.1.transformer_blocks.0.attn2.to_k": "int8",
"up_blocks.1.attentions.1.transformer_blocks.0.attn2.to_out.0": "int4",
"up_blocks.1.attentions.1.transformer_blocks.0.attn2.to_q": "int8",
"up_blocks.1.attentions.1.transformer_blocks.0.attn2.to_v": "int8",
"up_blocks.1.attentions.1.transformer_blocks.0.ff.net.0.proj": "int8",
"up_blocks.1.attentions.1.transformer_blocks.0.ff.net.2": "int8",
"up_blocks.1.attentions.2.proj_in": "fp16",
"up_blocks.1.attentions.2.proj_out": "fp16",
"up_blocks.1.attentions.2.transformer_blocks.0.attn1.to_k": "int8",
"up_blocks.1.attentions.2.transformer_blocks.0.attn1.to_out.0": "fp16",
"up_blocks.1.attentions.2.transformer_blocks.0.attn1.to_q": "int8",
"up_blocks.1.attentions.2.transformer_blocks.0.attn1.to_v": "fp16",
"up_blocks.1.attentions.2.transformer_blocks.0.attn2.to_k": "int8",
"up_blocks.1.attentions.2.transformer_blocks.0.attn2.to_out.0": "int4",
"up_blocks.1.attentions.2.transformer_blocks.0.attn2.to_q": "int8",
"up_blocks.1.attentions.2.transformer_blocks.0.attn2.to_v": "int8",
"up_blocks.1.attentions.2.transformer_blocks.0.ff.net.0.proj": "fp16",
"up_blocks.1.attentions.2.transformer_blocks.0.ff.net.2": "int8",
"up_blocks.2.attentions.0.proj_in": "fp16",
"up_blocks.2.attentions.0.proj_out": "fp16",
"up_blocks.2.attentions.0.transformer_blocks.0.attn1.to_k": "int8",
"up_blocks.2.attentions.0.transformer_blocks.0.attn1.to_out.0": "fp16",
"up_blocks.2.attentions.0.transformer_blocks.0.attn1.to_q": "int8",
"up_blocks.2.attentions.0.transformer_blocks.0.attn1.to_v": "int8",
"up_blocks.2.attentions.0.transformer_blocks.0.attn2.to_k": "int8",
"up_blocks.2.attentions.0.transformer_blocks.0.attn2.to_out.0": "int8",
"up_blocks.2.attentions.0.transformer_blocks.0.attn2.to_q": "int8",
"up_blocks.2.attentions.0.transformer_blocks.0.attn2.to_v": "int8",
"up_blocks.2.attentions.0.transformer_blocks.0.ff.net.0.proj": "fp16",
"up_blocks.2.attentions.0.transformer_blocks.0.ff.net.2": "fp16",
"up_blocks.2.attentions.1.proj_in": "fp16",
"up_blocks.2.attentions.1.proj_out": "fp16",
"up_blocks.2.attentions.1.transformer_blocks.0.attn1.to_k": "int8",
"up_blocks.2.attentions.1.transformer_blocks.0.attn1.to_out.0": "fp16",
"up_blocks.2.attentions.1.transformer_blocks.0.attn1.to_q": "int8",
"up_blocks.2.attentions.1.transformer_blocks.0.attn1.to_v": "fp16",
"up_blocks.2.attentions.1.transformer_blocks.0.attn2.to_k": "int8",
"up_blocks.2.attentions.1.transformer_blocks.0.attn2.to_out.0": "int8",
"up_blocks.2.attentions.1.transformer_blocks.0.attn2.to_q": "int8",
"up_blocks.2.attentions.1.transformer_blocks.0.attn2.to_v": "fp16",
"up_blocks.2.attentions.1.transformer_blocks.0.ff.net.0.proj": "fp16",
"up_blocks.2.attentions.1.transformer_blocks.0.ff.net.2": "fp16",
"up_blocks.2.attentions.2.proj_in": "fp16",
"up_blocks.2.attentions.2.proj_out": "fp16",
"up_blocks.2.attentions.2.transformer_blocks.0.attn1.to_k": "int8",
"up_blocks.2.attentions.2.transformer_blocks.0.attn1.to_out.0": "fp16",
"up_blocks.2.attentions.2.transformer_blocks.0.attn1.to_q": "int8",
"up_blocks.2.attentions.2.transformer_blocks.0.attn1.to_v": "fp16",
"up_blocks.2.attentions.2.transformer_blocks.0.attn2.to_k": "int8",
"up_blocks.2.attentions.2.transformer_blocks.0.attn2.to_out.0": "fp16",
"up_blocks.2.attentions.2.transformer_blocks.0.attn2.to_q": "int8",
"up_blocks.2.attentions.2.transformer_blocks.0.attn2.to_v": "int8",
"up_blocks.2.attentions.2.transformer_blocks.0.ff.net.0.proj": "fp16",
"up_blocks.2.attentions.2.transformer_blocks.0.ff.net.2": "fp16",
"up_blocks.3.attentions.0.proj_in": "int4",
"up_blocks.3.attentions.0.proj_out": "int4",
"up_blocks.3.attentions.0.transformer_blocks.0.attn1.to_k": "int4",
"up_blocks.3.attentions.0.transformer_blocks.0.attn1.to_out.0": "int4",
"up_blocks.3.attentions.0.transformer_blocks.0.attn1.to_q": "int4",
"up_blocks.3.attentions.0.transformer_blocks.0.attn1.to_v": "int4",
"up_blocks.3.attentions.0.transformer_blocks.0.attn2.to_k": "int8",
"up_blocks.3.attentions.0.transformer_blocks.0.attn2.to_out.0": "int8",
"up_blocks.3.attentions.0.transformer_blocks.0.attn2.to_q": "int4",
"up_blocks.3.attentions.0.transformer_blocks.0.attn2.to_v": "int8",
"up_blocks.3.attentions.0.transformer_blocks.0.ff.net.0.proj": "int4",
"up_blocks.3.attentions.0.transformer_blocks.0.ff.net.2": "fp16",
"up_blocks.3.attentions.1.proj_in": "int4",
"up_blocks.3.attentions.1.proj_out": "int4",
"up_blocks.3.attentions.1.transformer_blocks.0.attn1.to_k": "int4",
"up_blocks.3.attentions.1.transformer_blocks.0.attn1.to_out.0": "int4",
"up_blocks.3.attentions.1.transformer_blocks.0.attn1.to_q": "int4",
"up_blocks.3.attentions.1.transformer_blocks.0.attn1.to_v": "int4",
"up_blocks.3.attentions.1.transformer_blocks.0.attn2.to_k": "int8",
"up_blocks.3.attentions.1.transformer_blocks.0.attn2.to_out.0": "fp16",
"up_blocks.3.attentions.1.transformer_blocks.0.attn2.to_q": "int4",
"up_blocks.3.attentions.1.transformer_blocks.0.attn2.to_v": "fp16",
"up_blocks.3.attentions.1.transformer_blocks.0.ff.net.0.proj": "int4",
"up_blocks.3.attentions.1.transformer_blocks.0.ff.net.2": "fp16",
"up_blocks.3.attentions.2.proj_in": "int4",
"up_blocks.3.attentions.2.proj_out": "int4",
"up_blocks.3.attentions.2.transformer_blocks.0.attn1.to_k": "int4",
"up_blocks.3.attentions.2.transformer_blocks.0.attn1.to_out.0": "int4",
"up_blocks.3.attentions.2.transformer_blocks.0.attn1.to_q": "int4",
"up_blocks.3.attentions.2.transformer_blocks.0.attn1.to_v": "int4",
"up_blocks.3.attentions.2.transformer_blocks.0.attn2.to_k": "int8",
"up_blocks.3.attentions.2.transformer_blocks.0.attn2.to_out.0": "int8",
"up_blocks.3.attentions.2.transformer_blocks.0.attn2.to_q": "int4",
"up_blocks.3.attentions.2.transformer_blocks.0.attn2.to_v": "fp16",
"up_blocks.3.attentions.2.transformer_blocks.0.ff.net.0.proj": "int4",
"up_blocks.3.attentions.2.transformer_blocks.0.ff.net.2": "fp16"
},
"size_reduction": 0.3753969219677904,
"model_dtype": "bfloat16"
},
"model_config": {
"sample_size": 64,
"in_channels": 4,
"out_channels": 4,
"center_input_sample": false,
"flip_sin_to_cos": true,
"freq_shift": 0,
"down_block_types": [
"CrossAttnDownBlock2D",
"CrossAttnDownBlock2D",
"CrossAttnDownBlock2D",
"DownBlock2D"
],
"mid_block_type": "UNetMidBlock2DCrossAttn",
"up_block_types": [
"UpBlock2D",
"CrossAttnUpBlock2D",
"CrossAttnUpBlock2D",
"CrossAttnUpBlock2D"
],
"only_cross_attention": false,
"block_out_channels": [
320,
640,
1280,
1280
],
"layers_per_block": 2,
"downsample_padding": 1,
"mid_block_scale_factor": 1,
"dropout": 0.0,
"act_fn": "silu",
"norm_num_groups": 32,
"norm_eps": 1e-05,
"cross_attention_dim": 1024,
"transformer_layers_per_block": 1,
"reverse_transformer_layers_per_block": null,
"encoder_hid_dim": null,
"encoder_hid_dim_type": null,
"attention_head_dim": [
5,
10,
20,
20
],
"num_attention_heads": null,
"dual_cross_attention": false,
"use_linear_projection": true,
"class_embed_type": null,
"addition_embed_type": null,
"addition_time_embed_dim": null,
"num_class_embeds": null,
"upcast_attention": null,
"resnet_time_scale_shift": "default",
"resnet_skip_time_act": false,
"resnet_out_scale_factor": 1.0,
"time_embedding_type": "positional",
"time_embedding_dim": null,
"time_embedding_act_fn": null,
"timestep_post_act": null,
"time_cond_proj_dim": null,
"conv_in_kernel": 3,
"conv_out_kernel": 3,
"projection_class_embeddings_input_dim": null,
"attention_type": "default",
"class_embeddings_concat": false,
"mid_block_only_cross_attention": null,
"cross_attention_norm": null,
"addition_embed_type_num_heads": 64
}
}