| { |
| "_class_name": "SanaTransformer2DModel", |
| "_diffusers_version": "0.34.0.dev0", |
| "_name_or_path": "Efficient-Large-Model/Sana_600M_1024px_diffusers", |
| "attention_bias": false, |
| "attention_head_dim": 32, |
| "caption_channels": 2304, |
| "cross_attention_dim": 1152, |
| "cross_attention_head_dim": 72, |
| "dropout": 0.0, |
| "guidance_embeds": false, |
| "guidance_embeds_scale": 0.1, |
| "in_channels": 32, |
| "interpolation_scale": null, |
| "mlp_ratio": 2.5, |
| "norm_elementwise_affine": false, |
| "norm_eps": 1e-06, |
| "num_attention_heads": 36, |
| "num_cross_attention_heads": 16, |
| "num_layers": 28, |
| "out_channels": 32, |
| "patch_size": 1, |
| "qk_norm": null, |
| "quantization_config": { |
| "algorithm": "max", |
| "block_quantize": null, |
| "channel_quantize": null, |
| "modelopt_config": { |
| "algorithm": "max", |
| "quant_cfg": { |
| "*block_sparse_moe.gate*": { |
| "enable": false |
| }, |
| "*input_quantizer": { |
| "enable": false |
| }, |
| "*k_bmm_quantizer": { |
| "enable": false |
| }, |
| "*lm_head*": { |
| "enable": false |
| }, |
| "*mlp.gate.*": { |
| "enable": false |
| }, |
| "*mlp.shared_expert_gate.*": { |
| "enable": false |
| }, |
| "*output_layer*": { |
| "enable": false |
| }, |
| "*output_quantizer": { |
| "enable": false |
| }, |
| "*proj_out.*": { |
| "enable": false |
| }, |
| "*q_bmm_quantizer": { |
| "enable": false |
| }, |
| "*router*": { |
| "enable": false |
| }, |
| "*softmax_quantizer": { |
| "enable": false |
| }, |
| "*v_bmm_quantizer": { |
| "enable": false |
| }, |
| "*weight_quantizer": { |
| "fake_quant": false, |
| "num_bits": [ |
| 4, |
| 3 |
| ] |
| }, |
| "default": { |
| "enable": false |
| }, |
| "nn.BatchNorm1d": { |
| "*": { |
| "enable": false |
| } |
| }, |
| "nn.BatchNorm2d": { |
| "*": { |
| "enable": false |
| } |
| }, |
| "nn.BatchNorm3d": { |
| "*": { |
| "enable": false |
| } |
| }, |
| "nn.LeakyReLU": { |
| "*": { |
| "enable": false |
| } |
| }, |
| "output.*": { |
| "enable": false |
| } |
| } |
| }, |
| "modules_to_not_convert": null, |
| "quant_method": "modelopt", |
| "quant_type": "FP8", |
| "type_bit_map": { |
| "FP8": [ |
| 4, |
| 3 |
| ], |
| "INT4": 4, |
| "NVFP4": [ |
| 2, |
| 1 |
| ] |
| }, |
| "weight_only": true |
| }, |
| "sample_size": 32, |
| "timestep_scale": 1.0 |
| } |
|
|
