Instructions to use RetentionLabs/personaplex-ko-s6-text with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Moshi
How to use RetentionLabs/personaplex-ko-s6-text with Moshi:
# pip install moshi # Run the interactive web server python -m moshi.server --hf-repo "RetentionLabs/personaplex-ko-s6-text" # Then open https://localhost:8998 in your browser
# pip install moshi import torch from moshi.models import loaders # Load checkpoint info from HuggingFace checkpoint = loaders.CheckpointInfo.from_hf_repo("RetentionLabs/personaplex-ko-s6-text") # Load the Mimi audio codec mimi = checkpoint.get_mimi(device="cuda") mimi.set_num_codebooks(8) # Encode audio (24kHz, mono) wav = torch.randn(1, 1, 24000 * 10) # [batch, channels, samples] with torch.no_grad(): codes = mimi.encode(wav.cuda()) decoded = mimi.decode(codes) - Notebooks
- Google Colab
- Kaggle
personaplex-ko-s6-text (stage6 step 1600)
ํ๊ตญ์ด ํ ์คํธ ๋ฅ๋ ฅ ์์ฑ ์ง์ : ํ ์คํธ ํ ์ด๋ธ lr 2e-5(ร20), ๋๋จธ์ง LoRA 1e-6. text_ppl 6,140 โ 13.7. ์ค๋์ค ๊ฒฝ๋ก๋ ์ฌ์ค์ ๋ฌดํ์ต ์ํ(๋ฐํ ํ๋ ์ CE 4.23โ3.63)๋ผ ์์ฑ์ ์น์์ด. ์ดํ ๋ชจ๋ ์ค๋์ค ์คํ ์ด์ง(s9~)์ ๋ถ๊ธฐ ๋ฒ ์ด์ค.
๋ก๋ฉ
- ํ์ต/๊ฒ์ดํธ์ ๋์ผํ moshi ๋ฐํ์ ๊ถ์ฅ:
moshi.models.loaders.get_moshi_lm(model.safetensors)+tokenizer_spm_ko.model - HF ๋ฐํ์: dev transformers์
PersonaPlexForConditionalGeneration.from_pretrained(<dir>)(config.json ํฌํจ๋จ). ๋จ, HF ๊ฒฝ๋ก์ selfplay ์์ฑ์ ๋ฌด์ ์ด์๊ฐ ์์(๋ฆฌํฌํธ ยง5 ์ฐธ๊ณ ) โ ํ๊ฐ๋ moshi ๋ฐํ์ ๊ธฐ์ค. - ํ ์คํธ ์ํ๋ง์ temp 0.95 / top_k 60 ๊ถ์ฅ (0.7/25๋ ๋ฐ๋ณต ๋ฃจํ ์ ๋ฐ).
License / ์ฌ์ฉ ๊ณ ์ง
์ด ๊ฐ์ค์น๋ nvidia/personaplex-7b-v1์ ๋ฒ ์ด์ค๋ก NIKL(๊ตญ๋ฆฝ๊ตญ์ด์) ์ผ์๋ํ ์ฝํผ์ค๋ก ํ์ธํ๋ํ ํ์๋ฌผ์
๋๋ค.
๋ฒ ์ด์ค ๋ชจ๋ธ์ ๋ผ์ด์ ์ค๊ฐ ์น๊ณ๋๋ฉฐ, NIKL ์ฝํผ์ค ํ์๋ฌผ์ด๋ผ๋ ์ ์ ๊ณ ๋ คํด ์ฐ๊ตฌ ๋ชฉ์ ์ฌ์ฉ์ ๊ถ์ฅํฉ๋๋ค.
์ ์ฝํผ์ค ๋ฐ์ดํฐ ์์ฒด๋ ํฌํจ๋์ด ์์ง ์์ต๋๋ค.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support