Datasets:
Download codec-pilot/tts.toml from JonesLin/nano-duplex-data-v1: direct link, hf CLI and curl.
- Browser
- Download file 795 Bytes
-
https://huggingface.co/datasets/JonesLin/nano-duplex-data-v1/resolve/main/codec-pilot/tts.toml
- Command line
-
hf download hf://datasets/JonesLin/nano-duplex-data-v1/codec-pilot/tts.toml
-
curl -L -o tts.toml https://huggingface.co/datasets/JonesLin/nano-duplex-data-v1/resolve/main/codec-pilot/tts.toml
795 Bytes
| policy = "Qwen/Qwen3-0.6B-Base" | |
| policy_revision = "da87bfb608c14b7cf20ba1ce41287e8de496c0cd" | |
| mimi_repo = "kyutai/moshiko-pytorch-bf16" | |
| mimi_revision = "2bfc9ae6e89079a5cc7ed2a68436010d91a3d289" | |
| device = "cuda" | |
| audio_dim = 512 | |
| vision_dim = 768 | |
| codebooks = 8 | |
| codebook_size = 2048 | |
| audio_head_width = 256 | |
| audio_head_layers = 2 | |
| audio_head_attention_heads = 4 | |
| max_context_tokens = 2048 | |
| precision = "bfloat16" | |
| freeze_policy = true | |
| batch_size = 8 | |
| gradient_accumulation = 1 | |
| learning_rate = 0.0003 | |
| weight_decay = 0.01 | |
| steps = 400 | |
| save_every = 400 | |
| dataset = "codec-pilot/tts-train.jsonl" | |
| output_dir = "runs/hf-pilot-tts" | |
| stage = "alignment_tts" | |
| # Opt-in: use a new run; effective batch differs from the baseline. | |
| alignment_batching = "batched" | |
| frozen_linear_dtype = "bfloat16" | |