diffusers/FLUX.1-dev-torchao-int4

1{2  "architectures": [3    "T5EncoderModel"4  ],5  "classifier_dropout": 0.0,6  "d_ff": 10240,7  "d_kv": 64,8  "d_model": 4096,9  "decoder_start_token_id": 0,10  "dense_act_fn": "gelu_new",11  "dropout_rate": 0.1,12  "eos_token_id": 1,13  "feed_forward_proj": "gated-gelu",14  "initializer_factor": 1.0,15  "is_encoder_decoder": true,16  "is_gated_act": true,17  "layer_norm_epsilon": 1e-06,18  "model_type": "t5",19  "num_decoder_layers": 24,20  "num_heads": 64,21  "num_layers": 24,22  "output_past": true,23  "pad_token_id": 0,24  "quantization_config": {25    "include_input_output_embeddings": false,26    "modules_to_not_convert": null,27    "quant_method": "torchao",28    "quant_type": "int4_weight_only",29    "quant_type_kwargs": {},30    "untie_embedding_weights": false31  },32  "relative_attention_max_distance": 128,33  "relative_attention_num_buckets": 32,34  "tie_word_embeddings": false,35  "torch_dtype": "bfloat16",36  "transformers_version": "4.52.0.dev0",37  "use_cache": true,38  "vocab_size": 3212839}40