quant-FlashTalk / bnb4 /config.json
Yuvrajxms09's picture
Upload BNB4 quantized FlashTalk DiT (bitsandbytes)
336629b verified
raw
history blame contribute delete
980 Bytes
{
"_class_name": "WanModel",
"_diffusers_version": "0.37.0.dev0",
"_name_or_path": "/models/SoulX-FlashTalk-14B",
"audio_window": 5,
"context_tokens": 32,
"dim": 5120,
"eps": 1e-06,
"ffn_dim": 13824,
"freq_dim": 256,
"in_dim": 36,
"intermediate_dim": 512,
"model_type": "i2v",
"norm_input_visual": true,
"norm_output_audio": true,
"num_heads": 40,
"num_layers": 40,
"out_dim": 16,
"output_dim": 768,
"quantization_config": {
"_load_in_4bit": true,
"_load_in_8bit": false,
"bnb_4bit_compute_dtype": "bfloat16",
"bnb_4bit_quant_storage": "uint8",
"bnb_4bit_quant_type": "nf4",
"bnb_4bit_use_double_quant": false,
"llm_int8_enable_fp32_cpu_offload": false,
"llm_int8_has_fp16_weight": false,
"llm_int8_skip_modules": null,
"llm_int8_threshold": 6.0,
"load_in_4bit": true,
"load_in_8bit": false,
"quant_method": "bitsandbytes"
},
"text_len": 512,
"vae_scale": 4,
"weight_init": false
}