Krea-2-Turbo-OrbitQuant-W4A4 / quantization_manifest.json
WaveCut's picture
Finalize quality and inference optimized OrbitQuant release
f84e201 verified
Raw
History Blame Contribute Delete
34.9 kB
{
"artifact_format": "orbitquant-multicomponent-v1",
"source_model_id": "krea/Krea-2-Turbo",
"source_revision": "98e0fe118d17c9e3547fbb2e25acdbae2cadf7c7",
"source_license": "krea-2-community-license-agreement",
"repo_id": "WaveCut/Krea-2-Turbo-OrbitQuant-W4A4",
"visibility": "public-ungated",
"quant_method": "orbitquant",
"recipe": "w4a4-universal-main-ff-down-and-img-in-bf16",
"weight_bits": 4,
"activation_bits": 4,
"w4a4_components": [
"text_encoder",
"transformer"
],
"source_precision_components": [
"vae",
"scheduler",
"tokenizer"
],
"calibration_data": null,
"orbitquant_version": "0.9.2",
"orbitquant_revision": "bundled-runtime-wheel",
"orbitquant_runtime_wheel": "runtime/orbitquant-0.9.2-py3-none-any.whl",
"orbitquant_runtime_wheel_sha256": "acf31e139e630c1a8513f973d09ade28f22224b27daadbf83266858ef2ac8ffe",
"runtime": {
"package": "orbitquant",
"wheel": "runtime/orbitquant-0.9.2-py3-none-any.whl",
"wheel_sha256": "acf31e139e630c1a8513f973d09ade28f22224b27daadbf83266858ef2ac8ffe",
"packed_weight_path": "native_packed_matmul",
"activation_path": "direct_triton_int8_surrogate",
"shared_activation_quantization": true,
"attention": "strict_flash_sdpa_bf16",
"full_model_torch_compile": false,
"full_weight_dequantization_cache": false,
"decoded_int8_weight_cache": false
},
"diffusers_version": "0.39.0",
"components": [
{
"status": "complete",
"component": "transformer",
"framework": "diffusers",
"class_name": "Krea2Transformer2DModel",
"component_mode": "orbitquant_w4a4",
"source_weight_bytes": 26283288880,
"artifact_bytes": 10704580693,
"load_and_quantize_seconds": 249.18440662510693,
"save_seconds": 2.236414270475507,
"wall_seconds": 251.42434754874557,
"rss_before_bytes": 831778816,
"rss_after_bytes": 7679791104,
"released_source_tensor_bytes": 26118455344,
"source_page_release_failures": 0,
"gpu": {
"device": "NVIDIA RTX PRO 6000 Blackwell Server Edition",
"capability": [
12,
0
],
"free_bytes": 101384585216,
"total_bytes": 101974081536,
"allocated_bytes": 165376,
"reserved_bytes": 2097152,
"peak_allocated_bytes": 29526528,
"peak_reserved_bytes": 46137344
},
"orbitquant_module_count": 232,
"adaln_int4_module_count": 0,
"source_precision_linear_module_count": 32,
"orbitquant_modules": [
"time_mod_proj",
"text_fusion.layerwise_blocks.0.attn.to_q",
"text_fusion.layerwise_blocks.0.attn.to_k",
"text_fusion.layerwise_blocks.0.attn.to_v",
"text_fusion.layerwise_blocks.0.attn.to_gate",
"text_fusion.layerwise_blocks.0.attn.to_out.0",
"text_fusion.layerwise_blocks.0.ff.gate",
"text_fusion.layerwise_blocks.0.ff.up",
"text_fusion.layerwise_blocks.0.ff.down",
"text_fusion.layerwise_blocks.1.attn.to_q",
"text_fusion.layerwise_blocks.1.attn.to_k",
"text_fusion.layerwise_blocks.1.attn.to_v",
"text_fusion.layerwise_blocks.1.attn.to_gate",
"text_fusion.layerwise_blocks.1.attn.to_out.0",
"text_fusion.layerwise_blocks.1.ff.gate",
"text_fusion.layerwise_blocks.1.ff.up",
"text_fusion.layerwise_blocks.1.ff.down",
"text_fusion.projector",
"text_fusion.refiner_blocks.0.attn.to_q",
"text_fusion.refiner_blocks.0.attn.to_k",
"text_fusion.refiner_blocks.0.attn.to_v",
"text_fusion.refiner_blocks.0.attn.to_gate",
"text_fusion.refiner_blocks.0.attn.to_out.0",
"text_fusion.refiner_blocks.0.ff.gate",
"text_fusion.refiner_blocks.0.ff.up",
"text_fusion.refiner_blocks.0.ff.down",
"text_fusion.refiner_blocks.1.attn.to_q",
"text_fusion.refiner_blocks.1.attn.to_k",
"text_fusion.refiner_blocks.1.attn.to_v",
"text_fusion.refiner_blocks.1.attn.to_gate",
"text_fusion.refiner_blocks.1.attn.to_out.0",
"text_fusion.refiner_blocks.1.ff.gate",
"text_fusion.refiner_blocks.1.ff.up",
"text_fusion.refiner_blocks.1.ff.down",
"txt_in.linear_1",
"txt_in.linear_2",
"transformer_blocks.0.attn.to_q",
"transformer_blocks.0.attn.to_k",
"transformer_blocks.0.attn.to_v",
"transformer_blocks.0.attn.to_gate",
"transformer_blocks.0.attn.to_out.0",
"transformer_blocks.0.ff.gate",
"transformer_blocks.0.ff.up",
"transformer_blocks.1.attn.to_q",
"transformer_blocks.1.attn.to_k",
"transformer_blocks.1.attn.to_v",
"transformer_blocks.1.attn.to_gate",
"transformer_blocks.1.attn.to_out.0",
"transformer_blocks.1.ff.gate",
"transformer_blocks.1.ff.up",
"transformer_blocks.2.attn.to_q",
"transformer_blocks.2.attn.to_k",
"transformer_blocks.2.attn.to_v",
"transformer_blocks.2.attn.to_gate",
"transformer_blocks.2.attn.to_out.0",
"transformer_blocks.2.ff.gate",
"transformer_blocks.2.ff.up",
"transformer_blocks.3.attn.to_q",
"transformer_blocks.3.attn.to_k",
"transformer_blocks.3.attn.to_v",
"transformer_blocks.3.attn.to_gate",
"transformer_blocks.3.attn.to_out.0",
"transformer_blocks.3.ff.gate",
"transformer_blocks.3.ff.up",
"transformer_blocks.4.attn.to_q",
"transformer_blocks.4.attn.to_k",
"transformer_blocks.4.attn.to_v",
"transformer_blocks.4.attn.to_gate",
"transformer_blocks.4.attn.to_out.0",
"transformer_blocks.4.ff.gate",
"transformer_blocks.4.ff.up",
"transformer_blocks.5.attn.to_q",
"transformer_blocks.5.attn.to_k",
"transformer_blocks.5.attn.to_v",
"transformer_blocks.5.attn.to_gate",
"transformer_blocks.5.attn.to_out.0",
"transformer_blocks.5.ff.gate",
"transformer_blocks.5.ff.up",
"transformer_blocks.6.attn.to_q",
"transformer_blocks.6.attn.to_k",
"transformer_blocks.6.attn.to_v",
"transformer_blocks.6.attn.to_gate",
"transformer_blocks.6.attn.to_out.0",
"transformer_blocks.6.ff.gate",
"transformer_blocks.6.ff.up",
"transformer_blocks.7.attn.to_q",
"transformer_blocks.7.attn.to_k",
"transformer_blocks.7.attn.to_v",
"transformer_blocks.7.attn.to_gate",
"transformer_blocks.7.attn.to_out.0",
"transformer_blocks.7.ff.gate",
"transformer_blocks.7.ff.up",
"transformer_blocks.8.attn.to_q",
"transformer_blocks.8.attn.to_k",
"transformer_blocks.8.attn.to_v",
"transformer_blocks.8.attn.to_gate",
"transformer_blocks.8.attn.to_out.0",
"transformer_blocks.8.ff.gate",
"transformer_blocks.8.ff.up",
"transformer_blocks.9.attn.to_q",
"transformer_blocks.9.attn.to_k",
"transformer_blocks.9.attn.to_v",
"transformer_blocks.9.attn.to_gate",
"transformer_blocks.9.attn.to_out.0",
"transformer_blocks.9.ff.gate",
"transformer_blocks.9.ff.up",
"transformer_blocks.10.attn.to_q",
"transformer_blocks.10.attn.to_k",
"transformer_blocks.10.attn.to_v",
"transformer_blocks.10.attn.to_gate",
"transformer_blocks.10.attn.to_out.0",
"transformer_blocks.10.ff.gate",
"transformer_blocks.10.ff.up",
"transformer_blocks.11.attn.to_q",
"transformer_blocks.11.attn.to_k",
"transformer_blocks.11.attn.to_v",
"transformer_blocks.11.attn.to_gate",
"transformer_blocks.11.attn.to_out.0",
"transformer_blocks.11.ff.gate",
"transformer_blocks.11.ff.up",
"transformer_blocks.12.attn.to_q",
"transformer_blocks.12.attn.to_k",
"transformer_blocks.12.attn.to_v",
"transformer_blocks.12.attn.to_gate",
"transformer_blocks.12.attn.to_out.0",
"transformer_blocks.12.ff.gate",
"transformer_blocks.12.ff.up",
"transformer_blocks.13.attn.to_q",
"transformer_blocks.13.attn.to_k",
"transformer_blocks.13.attn.to_v",
"transformer_blocks.13.attn.to_gate",
"transformer_blocks.13.attn.to_out.0",
"transformer_blocks.13.ff.gate",
"transformer_blocks.13.ff.up",
"transformer_blocks.14.attn.to_q",
"transformer_blocks.14.attn.to_k",
"transformer_blocks.14.attn.to_v",
"transformer_blocks.14.attn.to_gate",
"transformer_blocks.14.attn.to_out.0",
"transformer_blocks.14.ff.gate",
"transformer_blocks.14.ff.up",
"transformer_blocks.15.attn.to_q",
"transformer_blocks.15.attn.to_k",
"transformer_blocks.15.attn.to_v",
"transformer_blocks.15.attn.to_gate",
"transformer_blocks.15.attn.to_out.0",
"transformer_blocks.15.ff.gate",
"transformer_blocks.15.ff.up",
"transformer_blocks.16.attn.to_q",
"transformer_blocks.16.attn.to_k",
"transformer_blocks.16.attn.to_v",
"transformer_blocks.16.attn.to_gate",
"transformer_blocks.16.attn.to_out.0",
"transformer_blocks.16.ff.gate",
"transformer_blocks.16.ff.up",
"transformer_blocks.17.attn.to_q",
"transformer_blocks.17.attn.to_k",
"transformer_blocks.17.attn.to_v",
"transformer_blocks.17.attn.to_gate",
"transformer_blocks.17.attn.to_out.0",
"transformer_blocks.17.ff.gate",
"transformer_blocks.17.ff.up",
"transformer_blocks.18.attn.to_q",
"transformer_blocks.18.attn.to_k",
"transformer_blocks.18.attn.to_v",
"transformer_blocks.18.attn.to_gate",
"transformer_blocks.18.attn.to_out.0",
"transformer_blocks.18.ff.gate",
"transformer_blocks.18.ff.up",
"transformer_blocks.19.attn.to_q",
"transformer_blocks.19.attn.to_k",
"transformer_blocks.19.attn.to_v",
"transformer_blocks.19.attn.to_gate",
"transformer_blocks.19.attn.to_out.0",
"transformer_blocks.19.ff.gate",
"transformer_blocks.19.ff.up",
"transformer_blocks.20.attn.to_q",
"transformer_blocks.20.attn.to_k",
"transformer_blocks.20.attn.to_v",
"transformer_blocks.20.attn.to_gate",
"transformer_blocks.20.attn.to_out.0",
"transformer_blocks.20.ff.gate",
"transformer_blocks.20.ff.up",
"transformer_blocks.21.attn.to_q",
"transformer_blocks.21.attn.to_k",
"transformer_blocks.21.attn.to_v",
"transformer_blocks.21.attn.to_gate",
"transformer_blocks.21.attn.to_out.0",
"transformer_blocks.21.ff.gate",
"transformer_blocks.21.ff.up",
"transformer_blocks.22.attn.to_q",
"transformer_blocks.22.attn.to_k",
"transformer_blocks.22.attn.to_v",
"transformer_blocks.22.attn.to_gate",
"transformer_blocks.22.attn.to_out.0",
"transformer_blocks.22.ff.gate",
"transformer_blocks.22.ff.up",
"transformer_blocks.23.attn.to_q",
"transformer_blocks.23.attn.to_k",
"transformer_blocks.23.attn.to_v",
"transformer_blocks.23.attn.to_gate",
"transformer_blocks.23.attn.to_out.0",
"transformer_blocks.23.ff.gate",
"transformer_blocks.23.ff.up",
"transformer_blocks.24.attn.to_q",
"transformer_blocks.24.attn.to_k",
"transformer_blocks.24.attn.to_v",
"transformer_blocks.24.attn.to_gate",
"transformer_blocks.24.attn.to_out.0",
"transformer_blocks.24.ff.gate",
"transformer_blocks.24.ff.up",
"transformer_blocks.25.attn.to_q",
"transformer_blocks.25.attn.to_k",
"transformer_blocks.25.attn.to_v",
"transformer_blocks.25.attn.to_gate",
"transformer_blocks.25.attn.to_out.0",
"transformer_blocks.25.ff.gate",
"transformer_blocks.25.ff.up",
"transformer_blocks.26.attn.to_q",
"transformer_blocks.26.attn.to_k",
"transformer_blocks.26.attn.to_v",
"transformer_blocks.26.attn.to_gate",
"transformer_blocks.26.attn.to_out.0",
"transformer_blocks.26.ff.gate",
"transformer_blocks.26.ff.up",
"transformer_blocks.27.attn.to_q",
"transformer_blocks.27.attn.to_k",
"transformer_blocks.27.attn.to_v",
"transformer_blocks.27.attn.to_gate",
"transformer_blocks.27.attn.to_out.0",
"transformer_blocks.27.ff.gate",
"transformer_blocks.27.ff.up"
],
"adaln_int4_modules": [],
"source_precision_linear_modules": [
"final_layer.linear",
"img_in",
"time_embed.linear_1",
"time_embed.linear_2",
"transformer_blocks.0.ff.down",
"transformer_blocks.1.ff.down",
"transformer_blocks.10.ff.down",
"transformer_blocks.11.ff.down",
"transformer_blocks.12.ff.down",
"transformer_blocks.13.ff.down",
"transformer_blocks.14.ff.down",
"transformer_blocks.15.ff.down",
"transformer_blocks.16.ff.down",
"transformer_blocks.17.ff.down",
"transformer_blocks.18.ff.down",
"transformer_blocks.19.ff.down",
"transformer_blocks.2.ff.down",
"transformer_blocks.20.ff.down",
"transformer_blocks.21.ff.down",
"transformer_blocks.22.ff.down",
"transformer_blocks.23.ff.down",
"transformer_blocks.24.ff.down",
"transformer_blocks.25.ff.down",
"transformer_blocks.26.ff.down",
"transformer_blocks.27.ff.down",
"transformer_blocks.3.ff.down",
"transformer_blocks.4.ff.down",
"transformer_blocks.5.ff.down",
"transformer_blocks.6.ff.down",
"transformer_blocks.7.ff.down",
"transformer_blocks.8.ff.down",
"transformer_blocks.9.ff.down"
],
"quantized_linear_weight_parameters": 9959899148,
"source_precision_linear_weight_parameters": 2858680320,
"linear_weight_parameters": 12818579468,
"linear_parameter_coverage": 0.7769893046935238,
"packed_module_state_bytes": 6393049096,
"full_dequantized_cache_count": 0,
"quality_protection": {
"patterns": [
"transformer_blocks.*.ff.down",
"img_in"
],
"module_count": 29,
"source_precision_parameter_count": 2818971648,
"selection": "component-isolated native-size image, latent-trajectory, and smooth-gradient ablations"
}
},
{
"status": "complete",
"component": "text_encoder",
"framework": "transformers",
"class_name": "Qwen3VLModel",
"component_mode": "orbitquant_w4a4",
"source_weight_bytes": 8875715136,
"artifact_bytes": 2811934088,
"load_and_quantize_seconds": 104.62921418622136,
"save_seconds": 0.3852194733917713,
"wall_seconds": 105.29201615415514,
"rss_before_bytes": 7679791104,
"rss_after_bytes": 12943683584,
"released_source_tensor_bytes": 8088715264,
"source_page_release_failures": 0,
"gpu": {
"device": "NVIDIA RTX PRO 6000 Blackwell Server Edition",
"capability": [
12,
0
],
"free_bytes": 101340545024,
"total_bytes": 101974081536,
"allocated_bytes": 241152,
"reserved_bytes": 46137344,
"peak_allocated_bytes": 17674752,
"peak_reserved_bytes": 46137344
},
"orbitquant_module_count": 356,
"adaln_int4_module_count": 0,
"source_precision_linear_module_count": 0,
"orbitquant_modules": [
"visual.blocks.0.attn.qkv",
"visual.blocks.0.attn.proj",
"visual.blocks.0.mlp.linear_fc1",
"visual.blocks.0.mlp.linear_fc2",
"visual.blocks.1.attn.qkv",
"visual.blocks.1.attn.proj",
"visual.blocks.1.mlp.linear_fc1",
"visual.blocks.1.mlp.linear_fc2",
"visual.blocks.2.attn.qkv",
"visual.blocks.2.attn.proj",
"visual.blocks.2.mlp.linear_fc1",
"visual.blocks.2.mlp.linear_fc2",
"visual.blocks.3.attn.qkv",
"visual.blocks.3.attn.proj",
"visual.blocks.3.mlp.linear_fc1",
"visual.blocks.3.mlp.linear_fc2",
"visual.blocks.4.attn.qkv",
"visual.blocks.4.attn.proj",
"visual.blocks.4.mlp.linear_fc1",
"visual.blocks.4.mlp.linear_fc2",
"visual.blocks.5.attn.qkv",
"visual.blocks.5.attn.proj",
"visual.blocks.5.mlp.linear_fc1",
"visual.blocks.5.mlp.linear_fc2",
"visual.blocks.6.attn.qkv",
"visual.blocks.6.attn.proj",
"visual.blocks.6.mlp.linear_fc1",
"visual.blocks.6.mlp.linear_fc2",
"visual.blocks.7.attn.qkv",
"visual.blocks.7.attn.proj",
"visual.blocks.7.mlp.linear_fc1",
"visual.blocks.7.mlp.linear_fc2",
"visual.blocks.8.attn.qkv",
"visual.blocks.8.attn.proj",
"visual.blocks.8.mlp.linear_fc1",
"visual.blocks.8.mlp.linear_fc2",
"visual.blocks.9.attn.qkv",
"visual.blocks.9.attn.proj",
"visual.blocks.9.mlp.linear_fc1",
"visual.blocks.9.mlp.linear_fc2",
"visual.blocks.10.attn.qkv",
"visual.blocks.10.attn.proj",
"visual.blocks.10.mlp.linear_fc1",
"visual.blocks.10.mlp.linear_fc2",
"visual.blocks.11.attn.qkv",
"visual.blocks.11.attn.proj",
"visual.blocks.11.mlp.linear_fc1",
"visual.blocks.11.mlp.linear_fc2",
"visual.blocks.12.attn.qkv",
"visual.blocks.12.attn.proj",
"visual.blocks.12.mlp.linear_fc1",
"visual.blocks.12.mlp.linear_fc2",
"visual.blocks.13.attn.qkv",
"visual.blocks.13.attn.proj",
"visual.blocks.13.mlp.linear_fc1",
"visual.blocks.13.mlp.linear_fc2",
"visual.blocks.14.attn.qkv",
"visual.blocks.14.attn.proj",
"visual.blocks.14.mlp.linear_fc1",
"visual.blocks.14.mlp.linear_fc2",
"visual.blocks.15.attn.qkv",
"visual.blocks.15.attn.proj",
"visual.blocks.15.mlp.linear_fc1",
"visual.blocks.15.mlp.linear_fc2",
"visual.blocks.16.attn.qkv",
"visual.blocks.16.attn.proj",
"visual.blocks.16.mlp.linear_fc1",
"visual.blocks.16.mlp.linear_fc2",
"visual.blocks.17.attn.qkv",
"visual.blocks.17.attn.proj",
"visual.blocks.17.mlp.linear_fc1",
"visual.blocks.17.mlp.linear_fc2",
"visual.blocks.18.attn.qkv",
"visual.blocks.18.attn.proj",
"visual.blocks.18.mlp.linear_fc1",
"visual.blocks.18.mlp.linear_fc2",
"visual.blocks.19.attn.qkv",
"visual.blocks.19.attn.proj",
"visual.blocks.19.mlp.linear_fc1",
"visual.blocks.19.mlp.linear_fc2",
"visual.blocks.20.attn.qkv",
"visual.blocks.20.attn.proj",
"visual.blocks.20.mlp.linear_fc1",
"visual.blocks.20.mlp.linear_fc2",
"visual.blocks.21.attn.qkv",
"visual.blocks.21.attn.proj",
"visual.blocks.21.mlp.linear_fc1",
"visual.blocks.21.mlp.linear_fc2",
"visual.blocks.22.attn.qkv",
"visual.blocks.22.attn.proj",
"visual.blocks.22.mlp.linear_fc1",
"visual.blocks.22.mlp.linear_fc2",
"visual.blocks.23.attn.qkv",
"visual.blocks.23.attn.proj",
"visual.blocks.23.mlp.linear_fc1",
"visual.blocks.23.mlp.linear_fc2",
"visual.merger.linear_fc1",
"visual.merger.linear_fc2",
"visual.deepstack_merger_list.0.linear_fc1",
"visual.deepstack_merger_list.0.linear_fc2",
"visual.deepstack_merger_list.1.linear_fc1",
"visual.deepstack_merger_list.1.linear_fc2",
"visual.deepstack_merger_list.2.linear_fc1",
"visual.deepstack_merger_list.2.linear_fc2",
"language_model.layers.0.self_attn.q_proj",
"language_model.layers.0.self_attn.k_proj",
"language_model.layers.0.self_attn.v_proj",
"language_model.layers.0.self_attn.o_proj",
"language_model.layers.0.mlp.gate_proj",
"language_model.layers.0.mlp.up_proj",
"language_model.layers.0.mlp.down_proj",
"language_model.layers.1.self_attn.q_proj",
"language_model.layers.1.self_attn.k_proj",
"language_model.layers.1.self_attn.v_proj",
"language_model.layers.1.self_attn.o_proj",
"language_model.layers.1.mlp.gate_proj",
"language_model.layers.1.mlp.up_proj",
"language_model.layers.1.mlp.down_proj",
"language_model.layers.2.self_attn.q_proj",
"language_model.layers.2.self_attn.k_proj",
"language_model.layers.2.self_attn.v_proj",
"language_model.layers.2.self_attn.o_proj",
"language_model.layers.2.mlp.gate_proj",
"language_model.layers.2.mlp.up_proj",
"language_model.layers.2.mlp.down_proj",
"language_model.layers.3.self_attn.q_proj",
"language_model.layers.3.self_attn.k_proj",
"language_model.layers.3.self_attn.v_proj",
"language_model.layers.3.self_attn.o_proj",
"language_model.layers.3.mlp.gate_proj",
"language_model.layers.3.mlp.up_proj",
"language_model.layers.3.mlp.down_proj",
"language_model.layers.4.self_attn.q_proj",
"language_model.layers.4.self_attn.k_proj",
"language_model.layers.4.self_attn.v_proj",
"language_model.layers.4.self_attn.o_proj",
"language_model.layers.4.mlp.gate_proj",
"language_model.layers.4.mlp.up_proj",
"language_model.layers.4.mlp.down_proj",
"language_model.layers.5.self_attn.q_proj",
"language_model.layers.5.self_attn.k_proj",
"language_model.layers.5.self_attn.v_proj",
"language_model.layers.5.self_attn.o_proj",
"language_model.layers.5.mlp.gate_proj",
"language_model.layers.5.mlp.up_proj",
"language_model.layers.5.mlp.down_proj",
"language_model.layers.6.self_attn.q_proj",
"language_model.layers.6.self_attn.k_proj",
"language_model.layers.6.self_attn.v_proj",
"language_model.layers.6.self_attn.o_proj",
"language_model.layers.6.mlp.gate_proj",
"language_model.layers.6.mlp.up_proj",
"language_model.layers.6.mlp.down_proj",
"language_model.layers.7.self_attn.q_proj",
"language_model.layers.7.self_attn.k_proj",
"language_model.layers.7.self_attn.v_proj",
"language_model.layers.7.self_attn.o_proj",
"language_model.layers.7.mlp.gate_proj",
"language_model.layers.7.mlp.up_proj",
"language_model.layers.7.mlp.down_proj",
"language_model.layers.8.self_attn.q_proj",
"language_model.layers.8.self_attn.k_proj",
"language_model.layers.8.self_attn.v_proj",
"language_model.layers.8.self_attn.o_proj",
"language_model.layers.8.mlp.gate_proj",
"language_model.layers.8.mlp.up_proj",
"language_model.layers.8.mlp.down_proj",
"language_model.layers.9.self_attn.q_proj",
"language_model.layers.9.self_attn.k_proj",
"language_model.layers.9.self_attn.v_proj",
"language_model.layers.9.self_attn.o_proj",
"language_model.layers.9.mlp.gate_proj",
"language_model.layers.9.mlp.up_proj",
"language_model.layers.9.mlp.down_proj",
"language_model.layers.10.self_attn.q_proj",
"language_model.layers.10.self_attn.k_proj",
"language_model.layers.10.self_attn.v_proj",
"language_model.layers.10.self_attn.o_proj",
"language_model.layers.10.mlp.gate_proj",
"language_model.layers.10.mlp.up_proj",
"language_model.layers.10.mlp.down_proj",
"language_model.layers.11.self_attn.q_proj",
"language_model.layers.11.self_attn.k_proj",
"language_model.layers.11.self_attn.v_proj",
"language_model.layers.11.self_attn.o_proj",
"language_model.layers.11.mlp.gate_proj",
"language_model.layers.11.mlp.up_proj",
"language_model.layers.11.mlp.down_proj",
"language_model.layers.12.self_attn.q_proj",
"language_model.layers.12.self_attn.k_proj",
"language_model.layers.12.self_attn.v_proj",
"language_model.layers.12.self_attn.o_proj",
"language_model.layers.12.mlp.gate_proj",
"language_model.layers.12.mlp.up_proj",
"language_model.layers.12.mlp.down_proj",
"language_model.layers.13.self_attn.q_proj",
"language_model.layers.13.self_attn.k_proj",
"language_model.layers.13.self_attn.v_proj",
"language_model.layers.13.self_attn.o_proj",
"language_model.layers.13.mlp.gate_proj",
"language_model.layers.13.mlp.up_proj",
"language_model.layers.13.mlp.down_proj",
"language_model.layers.14.self_attn.q_proj",
"language_model.layers.14.self_attn.k_proj",
"language_model.layers.14.self_attn.v_proj",
"language_model.layers.14.self_attn.o_proj",
"language_model.layers.14.mlp.gate_proj",
"language_model.layers.14.mlp.up_proj",
"language_model.layers.14.mlp.down_proj",
"language_model.layers.15.self_attn.q_proj",
"language_model.layers.15.self_attn.k_proj",
"language_model.layers.15.self_attn.v_proj",
"language_model.layers.15.self_attn.o_proj",
"language_model.layers.15.mlp.gate_proj",
"language_model.layers.15.mlp.up_proj",
"language_model.layers.15.mlp.down_proj",
"language_model.layers.16.self_attn.q_proj",
"language_model.layers.16.self_attn.k_proj",
"language_model.layers.16.self_attn.v_proj",
"language_model.layers.16.self_attn.o_proj",
"language_model.layers.16.mlp.gate_proj",
"language_model.layers.16.mlp.up_proj",
"language_model.layers.16.mlp.down_proj",
"language_model.layers.17.self_attn.q_proj",
"language_model.layers.17.self_attn.k_proj",
"language_model.layers.17.self_attn.v_proj",
"language_model.layers.17.self_attn.o_proj",
"language_model.layers.17.mlp.gate_proj",
"language_model.layers.17.mlp.up_proj",
"language_model.layers.17.mlp.down_proj",
"language_model.layers.18.self_attn.q_proj",
"language_model.layers.18.self_attn.k_proj",
"language_model.layers.18.self_attn.v_proj",
"language_model.layers.18.self_attn.o_proj",
"language_model.layers.18.mlp.gate_proj",
"language_model.layers.18.mlp.up_proj",
"language_model.layers.18.mlp.down_proj",
"language_model.layers.19.self_attn.q_proj",
"language_model.layers.19.self_attn.k_proj",
"language_model.layers.19.self_attn.v_proj",
"language_model.layers.19.self_attn.o_proj",
"language_model.layers.19.mlp.gate_proj",
"language_model.layers.19.mlp.up_proj",
"language_model.layers.19.mlp.down_proj",
"language_model.layers.20.self_attn.q_proj",
"language_model.layers.20.self_attn.k_proj",
"language_model.layers.20.self_attn.v_proj",
"language_model.layers.20.self_attn.o_proj",
"language_model.layers.20.mlp.gate_proj",
"language_model.layers.20.mlp.up_proj",
"language_model.layers.20.mlp.down_proj",
"language_model.layers.21.self_attn.q_proj",
"language_model.layers.21.self_attn.k_proj",
"language_model.layers.21.self_attn.v_proj",
"language_model.layers.21.self_attn.o_proj",
"language_model.layers.21.mlp.gate_proj",
"language_model.layers.21.mlp.up_proj",
"language_model.layers.21.mlp.down_proj",
"language_model.layers.22.self_attn.q_proj",
"language_model.layers.22.self_attn.k_proj",
"language_model.layers.22.self_attn.v_proj",
"language_model.layers.22.self_attn.o_proj",
"language_model.layers.22.mlp.gate_proj",
"language_model.layers.22.mlp.up_proj",
"language_model.layers.22.mlp.down_proj",
"language_model.layers.23.self_attn.q_proj",
"language_model.layers.23.self_attn.k_proj",
"language_model.layers.23.self_attn.v_proj",
"language_model.layers.23.self_attn.o_proj",
"language_model.layers.23.mlp.gate_proj",
"language_model.layers.23.mlp.up_proj",
"language_model.layers.23.mlp.down_proj",
"language_model.layers.24.self_attn.q_proj",
"language_model.layers.24.self_attn.k_proj",
"language_model.layers.24.self_attn.v_proj",
"language_model.layers.24.self_attn.o_proj",
"language_model.layers.24.mlp.gate_proj",
"language_model.layers.24.mlp.up_proj",
"language_model.layers.24.mlp.down_proj",
"language_model.layers.25.self_attn.q_proj",
"language_model.layers.25.self_attn.k_proj",
"language_model.layers.25.self_attn.v_proj",
"language_model.layers.25.self_attn.o_proj",
"language_model.layers.25.mlp.gate_proj",
"language_model.layers.25.mlp.up_proj",
"language_model.layers.25.mlp.down_proj",
"language_model.layers.26.self_attn.q_proj",
"language_model.layers.26.self_attn.k_proj",
"language_model.layers.26.self_attn.v_proj",
"language_model.layers.26.self_attn.o_proj",
"language_model.layers.26.mlp.gate_proj",
"language_model.layers.26.mlp.up_proj",
"language_model.layers.26.mlp.down_proj",
"language_model.layers.27.self_attn.q_proj",
"language_model.layers.27.self_attn.k_proj",
"language_model.layers.27.self_attn.v_proj",
"language_model.layers.27.self_attn.o_proj",
"language_model.layers.27.mlp.gate_proj",
"language_model.layers.27.mlp.up_proj",
"language_model.layers.27.mlp.down_proj",
"language_model.layers.28.self_attn.q_proj",
"language_model.layers.28.self_attn.k_proj",
"language_model.layers.28.self_attn.v_proj",
"language_model.layers.28.self_attn.o_proj",
"language_model.layers.28.mlp.gate_proj",
"language_model.layers.28.mlp.up_proj",
"language_model.layers.28.mlp.down_proj",
"language_model.layers.29.self_attn.q_proj",
"language_model.layers.29.self_attn.k_proj",
"language_model.layers.29.self_attn.v_proj",
"language_model.layers.29.self_attn.o_proj",
"language_model.layers.29.mlp.gate_proj",
"language_model.layers.29.mlp.up_proj",
"language_model.layers.29.mlp.down_proj",
"language_model.layers.30.self_attn.q_proj",
"language_model.layers.30.self_attn.k_proj",
"language_model.layers.30.self_attn.v_proj",
"language_model.layers.30.self_attn.o_proj",
"language_model.layers.30.mlp.gate_proj",
"language_model.layers.30.mlp.up_proj",
"language_model.layers.30.mlp.down_proj",
"language_model.layers.31.self_attn.q_proj",
"language_model.layers.31.self_attn.k_proj",
"language_model.layers.31.self_attn.v_proj",
"language_model.layers.31.self_attn.o_proj",
"language_model.layers.31.mlp.gate_proj",
"language_model.layers.31.mlp.up_proj",
"language_model.layers.31.mlp.down_proj",
"language_model.layers.32.self_attn.q_proj",
"language_model.layers.32.self_attn.k_proj",
"language_model.layers.32.self_attn.v_proj",
"language_model.layers.32.self_attn.o_proj",
"language_model.layers.32.mlp.gate_proj",
"language_model.layers.32.mlp.up_proj",
"language_model.layers.32.mlp.down_proj",
"language_model.layers.33.self_attn.q_proj",
"language_model.layers.33.self_attn.k_proj",
"language_model.layers.33.self_attn.v_proj",
"language_model.layers.33.self_attn.o_proj",
"language_model.layers.33.mlp.gate_proj",
"language_model.layers.33.mlp.up_proj",
"language_model.layers.33.mlp.down_proj",
"language_model.layers.34.self_attn.q_proj",
"language_model.layers.34.self_attn.k_proj",
"language_model.layers.34.self_attn.v_proj",
"language_model.layers.34.self_attn.o_proj",
"language_model.layers.34.mlp.gate_proj",
"language_model.layers.34.mlp.up_proj",
"language_model.layers.34.mlp.down_proj",
"language_model.layers.35.self_attn.q_proj",
"language_model.layers.35.self_attn.k_proj",
"language_model.layers.35.self_attn.v_proj",
"language_model.layers.35.self_attn.o_proj",
"language_model.layers.35.mlp.gate_proj",
"language_model.layers.35.mlp.up_proj",
"language_model.layers.35.mlp.down_proj"
],
"adaln_int4_modules": [],
"source_precision_linear_modules": [],
"quantized_linear_weight_parameters": 4044357632,
"source_precision_linear_weight_parameters": 0,
"linear_weight_parameters": 4044357632,
"linear_parameter_coverage": 1.0,
"packed_module_state_bytes": 2025381888,
"full_dequantized_cache_count": 0
}
],
"totals": {
"source_weight_bytes": 35159004016,
"artifact_bytes": 13516514781,
"quantized_linear_weight_parameters": 14004256780,
"source_precision_linear_weight_parameters": 2858680320,
"orbitquant_module_count": 588,
"adaln_int4_module_count": 0,
"source_precision_linear_module_count": 32,
"linear_parameter_coverage": 0.8304755391633406
},
"comparison_matrix": "assets/original_vs_orbitquant_w4a4.webp",
"raw_generation_artifacts": "artifacts/generations",
"benchmark_summary": "benchmark/summary.json",
"quality_protection": {
"patterns": [
"transformer_blocks.*.ff.down",
"img_in"
],
"module_count": 29,
"source_precision_parameter_count": 2818971648,
"selection": "component-isolated native-size image, latent-trajectory, and smooth-gradient ablations"
},
"benchmark": {
"hardware": "NVIDIA A40 48 GB",
"torch": "2.8.0+cu128",
"width": 2048,
"height": 2048,
"steps": 8,
"guidance_scale": 0.0,
"attention": "strict_bf16_flash_sdpa",
"component_staging": "qwen_then_release_then_dit_and_tiled_vae",
"prompt_compaction": "lossless_valid_tokens",
"bf16_hot_end_to_end_median_seconds": 54.57881709188223,
"orbitquant_hot_end_to_end_median_seconds": 49.98552576825023,
"hot_end_to_end_speedup": 1.0918924279185949,
"orbitquant_empty_triton_cache_first_seconds": 68.87463613599539,
"bf16_nvml_peak_mib": 31441,
"orbitquant_nvml_peak_mib": 19177,
"bf16_torch_peak_mib": 28982.806640625,
"orbitquant_torch_peak_mib": 16798.73583984375,
"raw_summary": "benchmark/summary.json"
},
"diagnostic_artifacts": "artifacts/diagnostics"
}