FireRedTTS2 safetensors export
- Export date (UTC): 2025-12-03T15:33:21Z
- Source repo: FireRedTeam/FireRedTTS2 @ 4af3f5cc4963373b86b52d750220d4de85261f05
- Files:
- llm_pretrain.safetensors: ~7.70 GB, 392 tensors
metadata: llm_pretrain_metadata.json
- llm_posttrain.safetensors: ~7.70 GB, 392 tensors
metadata: llm_posttrain_metadata.json
- codec.safetensors: ~4.00 GB, 1202 tensors
metadata: codec_metadata.json
Notes:
- Tensor names are identical to the original PyTorch state_dict keys.
- Configs
config_llm.json and config_codec.json are copied verbatim.
- Qwen2.5 tokenizer assets are copied into
Qwen2.5-1.5B/ for Rust-side tokenization.
Verification (optional):
uv run python - <<'PY'
from safetensors.torch import load_file
sd = load_file('llm_pretrain.safetensors')
print(len(sd), list(sd)[:5])
PY
