| { |
| "format": "audio8_asr_onnx_bundle", |
| "format_version": 1, |
| "sampling_rate": 16000, |
| "max_audio_seconds": 30, |
| "prompt_mode": "arkasr_chat", |
| "response_prefix": "", |
| "language_hint_enabled": false, |
| "runtime_notes": { |
| "audio_loader": "librosa_first_for_hf_eval_parity" |
| }, |
| "prompt_audio": { |
| "token_count_rule": "arkasr_processor_floor_mel_downsample_merge", |
| "merge_factor": 4, |
| "hop_length": 160 |
| }, |
| "tokens": { |
| "user_token": "<|user|>", |
| "assistant_token": "<|assistant|>", |
| "bos_audio_token": "<|begin_of_audio|>", |
| "audio_token": "<|audio|>", |
| "eos_audio_token": "<|end_of_audio|>", |
| "assistant_end_token": "<|im_end|>", |
| "audio_token_id": 151646, |
| "pad_token_id": 151643, |
| "eos_token_ids": [ |
| 151645 |
| ], |
| "asr_block_token_id_from": -1, |
| "extra_block_token_ids": [ |
| 151647, |
| 151650, |
| 151648, |
| 151646, |
| 151649 |
| ], |
| "vocab_size": 151652 |
| }, |
| "audio_encoder": { |
| "type": "qwen3_asr_mlp_tower", |
| "qwen3_asr_audio_output_mode": "mlp_tower", |
| "prompt_token_count_uses_arkasr_processor_rule": true |
| }, |
| "graphs": { |
| "audio_hidden": { |
| "path": "audio_hidden.onnx", |
| "max_frames": 3000, |
| "frames_padded": 3000, |
| "feature_size": 128, |
| "max_audio_tokens": 390, |
| "hidden_size": 1024, |
| "tokens_per_chunk": 13, |
| "valid_mask_sum_at_max": 390 |
| }, |
| "audio_hidden_int8": { |
| "path": "audio_hidden_int8.onnx", |
| "quantization": "dynamic_int8_matmul_quint8_per_channel", |
| "source_graph": "audio_hidden.onnx", |
| "max_frames": 3000, |
| "frames_padded": 3000, |
| "feature_size": 128, |
| "max_audio_tokens": 390, |
| "hidden_size": 1024, |
| "tokens_per_chunk": 13, |
| "valid_mask_sum_at_max": 390 |
| }, |
| "lm_logits": { |
| "path": "lm_logits.onnx", |
| "hidden_size": 512, |
| "dummy_sequence_length": 16, |
| "patched_negative_transpose_perms": 0 |
| }, |
| "lm_cache_prefill": { |
| "path": "lm_cache_prefill.onnx", |
| "hidden_size": 512, |
| "num_layers": 8, |
| "num_attention_heads": 8, |
| "num_key_value_heads": 8, |
| "head_dim": 64, |
| "max_total_len": 512, |
| "dummy_sequence_length": 16, |
| "patched_negative_transpose_perms": 0 |
| }, |
| "lm_cache_decode": { |
| "path": "lm_cache_decode.onnx", |
| "hidden_size": 512, |
| "num_layers": 8, |
| "num_attention_heads": 8, |
| "num_key_value_heads": 8, |
| "head_dim": 64, |
| "max_total_len": 512, |
| "patched_negative_transpose_perms": 0 |
| } |
| }, |
| "weights": { |
| "token_embedding": "weights/token_embedding.npy", |
| "audio_projector": "weights/audio_projector.npz" |
| }, |
| "generation_defaults": { |
| "max_new_tokens": 128, |
| "do_sample": false, |
| "temperature": 0.5, |
| "repetition_penalty": 1.0 |
| } |
| } |
|
|
