WeDLM: The fastest diffusion language model with standard causal attention and native KV cache compatibility, delivering real speedups over vLLM-optimized baselines.
The first-ever vast natural language processing benchmark for Indonesian Language. We provide multiple downstream tasks, pre-trained IndoBERT models, and a starter code. (AACL-IJCNLP 2020).
Directly used on Hugging Face **Codec**: xcodec2 **TTS inference: Llasa-collections**: Llasa-collections **Update (2025-06-09):** Add Llasa finetune instruction. Awesome work based on Llasa LLaSa GRPO tuning : Special thanks to Seungyoun Shin from Channel Corp. for sharing the Llasa GRPO RL-tuning script based on Verl. You can try the finetuning results here: LLaSA 1B Multi-Speakers (Genshin-zh-en-ja-ko) LLaSA 1B Finetuned for Two Speakers **Update (2025-02-07):** Our paper has been released! Training
or Data You can download tokenized open-source speech data here. This includes LibriHeavy, Emilia (in both Chinese and English), and WenetSpeech4TTS, totaling approximately 160,000 hours of open-source data. Our models are trained on 250,000 hours of speech data. Of this, 160,000 hours come from the open-source datasets mentioned above, while the remaining 90,000 hours are from internal datasets, which are not yet available for open-source release. Data instruction Text_sequence is encoded
by the text tokenizer from Llama, for example, Llama-3. 2-1B-Instruct Speech_sequence is extrated through X-codec2 We change the value of speech tokens by adding len(text tokenizer) +8 special tokens thereby forming a unified tokenizer that encompasses both speech and text.
Frequently asked questions
How many stars and forks are recorded?
zhenye234/LLaSA_training has 657 stars and 50 forks.
What is the repository rank?
zhenye234/LLaSA_training is ranked 15,408.
What language and category are recorded?
zhenye234/LLaSA_training has 657 stars, uses Python and belongs to Models.
Which repository is closest in rank?
zhenye234/LLaSA_training is compared with bytedance/vidi, ranked 15,413 with 656 stars.
You can download tokenized open-source speech data here. This includes LibriHeavy, Emilia (in both Chinese and English), and WenetSpeech4TTS, totaling approximately 160,000 hours of open-source data.
Our models are trained on 250,000 hours of speech data. Of this, 160,000 hours come from the open-source datasets mentioned above, while the remaining 90,000 hours are from internal datasets, which are not yet available for open-source release.
Speech_sequence is extrated through X-codec2 We change the value of speech tokens by adding len(text tokenizer) +8 special tokens thereby forming a unified tokenizer that encompasses both speech and text.
