{ "quantization_method": "bitsandbytes", "bits": 4, "quant_type": "nf4", "double_quant": true, "modules_not_quantized": [ "acoustic_tokenizer_encoder", "semantic_tokenizer_encoder", "acoustic_projection", "semantic_projection", "lm_head" ], "modules_quantized": [ "language_model (excluding lm_head)" ], "base_model": "microsoft/VibeVoice-ASR-HF", "transformers_version": "5.3.0" }