@kadirnardev: From now on, I will work on accelerating the LFM2.5-Audio-1.5B model. I will also work on RL methods and synthetic data…
Summary
The developer announces plans to accelerate the LFM2.5-Audio-1.5B model using reinforcement learning methods and synthetic data, with a focus on English, and shares optimizations for the Mimi codec model to improve speed and quality for open-source release.
View Cached Full Text
Cached at: 09/13/26, 01:06 AM
From now on, I will work on accelerating the LFM2.5-Audio-1.5B model. I will also work on RL methods and synthetic data for this model. My priority right now will be the English language. In my initial attempts, the model runs 3-4x faster.
Kadir Nar (@kadirnardev): After 2 weeks of development, I finished optimizing the Mimi codec model, and it runs 15.4x faster for FP32.
Many omni and TTS models use the Mimi codec model. So I will fine-tune it to improve the WER and DNSMOS quality of the Mimi codec model and release it as open source.
Similar Articles
New audio model: LFM3-Audio-2B
Liquid AI released LFM3-Audio-2B, a compact 2B-parameter audio model that runs locally on laptops and edge devices, supporting voice chat, multilingual speech across 18 languages, audio understanding, and tool calling with an in-house audio encoder and LiquidCodec tokenizer.
@kadirnardev: I had been very focused on continuous pretraining and collecting real-world voice data, but after talking with Maxime L…
An AI developer shares that incorporating reinforcement learning and synthetic data improved their voice model's accuracy and inference speed, with plans to open-source the project.
@paulabartabajo_: Advice for AI engineers If you're building voice agents, stop wiring up 3 separate models, for audio-to-text, text-to-a…
Announces liquid-audio, an open-source repository for Liquid AI's end-to-end speech-to-speech LFM models (LFM2-Audio-1.5B and LFM2.5-Audio-1.5B) with interleaved and sequential generation modes and fine-tuning support.
@maximelabonne: We just released two new encoder models (MLM) in 2026 They're super fast, easy to train, and strongly multilingual. Try…
Liquid AI released two new encoder models, LFM2.5-Encoder-230M and LFM2.5-Encoder-350M, that are fast, easy to train, and strongly multilingual, with speed benchmarks showing over 3.7x improvement on CPU compared to ModernBERT-base.
@maximelabonne: It didn't take two days to get the first fine-tunes of LFM2.5-2.6B. This one looks cool!
Maxime Labonne highlights early fine-tunes of LFM2.5-2.6B, while Bad Theory Labs releases two open-weight models: BTL-4 35B, a frontier agentic reasoning model, and Macaw 2.7B, an on-device Mac agent, with notable BFCL v4 performance.