Maintained implementation availablepytorchPretrained Models Available

LLaMA-Omni2: LLM-based Real-time Spoken Chatbot with Autoregressive Streaming Speech Synthesis

May 1, 2025arXiv: 2505.02625

1 repo268 stars~a few hours to reproduce

Abstract

Task	Dataset	Metric	Value
Representation learning	LLaMA-Omni2-7B	ASR-WER.	3.26
Representation learning	w/o Gate Fusion	ASR-WER.	4.89
Representation learning	w/o Text Embedding	ASR-WER.	6.83
Language modeling	Streaming TTS	ASR-WER.	3.26
Language modeling	Offline TTS	ASR-WER.	3.51
Language modeling	Text Pretrained	ASR-WER.	10.34

ictnlp/LLaMA-Omni2

268 27 May 2025

License –

CI –

Deps ✓

Docker –

Selected ictnlp/llama-omni2 as the strongest maintained implementation for new work.
Includes dependency/environment manifest signals.
Repository activity is within the last 24 months.

1
Start with ictnlp/llama-omni2 and validate setup instructions in README.
2
Reproduce the baseline result with the provided defaults before modifying hyperparameters.
3
Log exact dependency versions and runtime environment for reproducibility.

Time to first repro: a few hoursLicense metadata missingNo CI workflows detected

No additional verified repositories beyond the primary recommendation.

No direct paper-linked artifacts were found. Showing strongest curated related artifacts.

Curated Related