| { |
| "_class_name": "Krea2Transformer2DModel", |
| "_diffusers_version": "0.39.0.dev0", |
| "attention_head_dim": 128, |
| "axes_dims_rope": [ |
| 32, |
| 48, |
| 48 |
| ], |
| "in_channels": 64, |
| "intermediate_size": 16384, |
| "norm_eps": 1e-05, |
| "num_attention_heads": 48, |
| "num_key_value_heads": 12, |
| "num_layers": 28, |
| "num_layerwise_text_blocks": 2, |
| "num_refiner_text_blocks": 2, |
| "num_text_layers": 12, |
| "rope_theta": 1000.0, |
| "text_hidden_dim": 2560, |
| "text_intermediate_size": 6912, |
| "text_num_attention_heads": 20, |
| "text_num_key_value_heads": 20, |
| "timestep_embed_dim": 256, |
| "quantization_config": { |
| "weight_dtype": "int6", |
| "group_size": 32, |
| "scale_dtype": "bfloat16", |
| "dynamic": { |
| "-:final_layer\\.linear$": {}, |
| "-:img_in$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.0\\.attn\\.to_gate$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.0\\.attn\\.to_k$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.0\\.attn\\.to_out\\.0$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.0\\.attn\\.to_q$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.0\\.attn\\.to_v$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.0\\.ff\\.down$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.0\\.ff\\.gate$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.0\\.ff\\.up$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.1\\.attn\\.to_gate$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.1\\.attn\\.to_k$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.1\\.attn\\.to_out\\.0$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.1\\.attn\\.to_q$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.1\\.attn\\.to_v$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.1\\.ff\\.down$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.1\\.ff\\.gate$": {}, |
| "-:text_fusion\\.layerwise_blocks\\.1\\.ff\\.up$": {}, |
| "-:text_fusion\\.projector$": {}, |
| "-:text_fusion\\.refiner_blocks\\.0\\.attn\\.to_gate$": {}, |
| "-:text_fusion\\.refiner_blocks\\.0\\.attn\\.to_k$": {}, |
| "-:text_fusion\\.refiner_blocks\\.0\\.attn\\.to_out\\.0$": {}, |
| "-:text_fusion\\.refiner_blocks\\.0\\.attn\\.to_q$": {}, |
| "-:text_fusion\\.refiner_blocks\\.0\\.attn\\.to_v$": {}, |
| "-:text_fusion\\.refiner_blocks\\.0\\.ff\\.down$": {}, |
| "-:text_fusion\\.refiner_blocks\\.0\\.ff\\.gate$": {}, |
| "-:text_fusion\\.refiner_blocks\\.0\\.ff\\.up$": {}, |
| "-:text_fusion\\.refiner_blocks\\.1\\.attn\\.to_gate$": {}, |
| "-:text_fusion\\.refiner_blocks\\.1\\.attn\\.to_k$": {}, |
| "-:text_fusion\\.refiner_blocks\\.1\\.attn\\.to_out\\.0$": {}, |
| "-:text_fusion\\.refiner_blocks\\.1\\.attn\\.to_q$": {}, |
| "-:text_fusion\\.refiner_blocks\\.1\\.attn\\.to_v$": {}, |
| "-:text_fusion\\.refiner_blocks\\.1\\.ff\\.down$": {}, |
| "-:text_fusion\\.refiner_blocks\\.1\\.ff\\.gate$": {}, |
| "-:text_fusion\\.refiner_blocks\\.1\\.ff\\.up$": {}, |
| "-:time_embed\\.linear_1$": {}, |
| "-:time_embed\\.linear_2$": {}, |
| "-:time_mod_proj$": {}, |
| "-:txt_in\\.linear_1$": {}, |
| "-:txt_in\\.linear_2$": {}, |
| "+:transformer_blocks\\.0\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.0\\.attn\\.to_k$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.0\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.0\\.attn\\.to_q$": { |
| "weight_dtype": "int3" |
| }, |
| "+:transformer_blocks\\.0\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.0\\.ff\\.down$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.0\\.ff\\.gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.0\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.1\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.1\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.1\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.1\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.1\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.1\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.1\\.ff\\.gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.1\\.ff\\.up$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.10\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.10\\.attn\\.to_k$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.10\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.10\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.10\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.10\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.10\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.10\\.ff\\.up$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.11\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.11\\.attn\\.to_k$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.11\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.11\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.11\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.11\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.11\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.11\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.12\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.12\\.attn\\.to_k$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.12\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.12\\.attn\\.to_q$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.12\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.12\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.12\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.12\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.13\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.13\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.13\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.13\\.attn\\.to_q$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.13\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.13\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.13\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.13\\.ff\\.up$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.14\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.14\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.14\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.14\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.14\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.14\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.14\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.14\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.15\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.15\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.15\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.15\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.15\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.15\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.15\\.ff\\.gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.15\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.16\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.16\\.attn\\.to_k$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.16\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.16\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.16\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.16\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.16\\.ff\\.gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.16\\.ff\\.up$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.17\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.17\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.17\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.17\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.17\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.17\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.17\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.17\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.18\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.18\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.18\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.18\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.18\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.18\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.18\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.18\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.19\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.19\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.19\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.19\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.19\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.19\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.19\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.19\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.2\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.2\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.2\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.2\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.2\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.2\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.2\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.2\\.ff\\.up$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.20\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.20\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.20\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.20\\.attn\\.to_q$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.20\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.20\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.20\\.ff\\.gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.20\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.21\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.21\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.21\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.21\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.21\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.21\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.21\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.21\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.22\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.22\\.attn\\.to_k$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.22\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.22\\.attn\\.to_q$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.22\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.22\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.22\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.22\\.ff\\.up$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.23\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.23\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.23\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.23\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.23\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.23\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.23\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.23\\.ff\\.up$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.24\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.24\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.24\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.24\\.attn\\.to_q$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.24\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.24\\.ff\\.down$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.24\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.24\\.ff\\.up$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.25\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.25\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.25\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.25\\.attn\\.to_q$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.25\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.25\\.ff\\.down$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.25\\.ff\\.gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.25\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.26\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.26\\.attn\\.to_k$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.26\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.26\\.attn\\.to_q$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.26\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.26\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.26\\.ff\\.gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.26\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.27\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.27\\.attn\\.to_k$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.27\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.27\\.attn\\.to_q$": { |
| "weight_dtype": "int4" |
| }, |
| "+:transformer_blocks\\.27\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.27\\.ff\\.down$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.27\\.ff\\.gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.27\\.ff\\.up$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.3\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.3\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.3\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.3\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.3\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.3\\.ff\\.down$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.3\\.ff\\.gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.3\\.ff\\.up$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.4\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.4\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.4\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.4\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.4\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.4\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.4\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.4\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.5\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.5\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.5\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.5\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.5\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.5\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.5\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.5\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.6\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.6\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.6\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.6\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.6\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.6\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.6\\.ff\\.gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.6\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.7\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.7\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.7\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.7\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.7\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.7\\.ff\\.down$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.7\\.ff\\.gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.7\\.ff\\.up$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.8\\.attn\\.to_gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.8\\.attn\\.to_k$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.8\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.8\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.8\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.8\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.8\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.8\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.9\\.attn\\.to_gate$": { |
| "weight_dtype": "int5" |
| }, |
| "+:transformer_blocks\\.9\\.attn\\.to_k$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.9\\.attn\\.to_out\\.0$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.9\\.attn\\.to_q$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.9\\.attn\\.to_v$": { |
| "weight_dtype": "int8" |
| }, |
| "+:transformer_blocks\\.9\\.ff\\.down$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.9\\.ff\\.gate$": { |
| "weight_dtype": "int6" |
| }, |
| "+:transformer_blocks\\.9\\.ff\\.up$": { |
| "weight_dtype": "int6" |
| } |
| }, |
| "quant_method": "humming", |
| "input_dtype": "int8", |
| "input_scale_group_size": 0 |
| } |
| } |
