Llama-Mimi 1.3B
Llama-Mimi 1.3B uses a single Llama decoder to model interleaved semantic and acoustic audio tokens. Its card identifies Meta Llama 3.2 1B as the base.
Research release
One decoder models semantic and acoustic tokens
The method uses the Mimi tokenizer rather than separate language and acoustic model stages. The released usage example continues an audio prompt. This is distinct from LLM-jp's Japanese conversational Moshi adaptation.Llama-Mimi 1.3B official model card
Evaluate the continuation task
Practical guidance. Use the smaller release to study speech continuation with less model memory. Judge speaker and acoustic continuity separately from whether the spoken content is useful.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | Llama-Mimi-1.3B |
| Training | Approximately 240,000 hours of English audio.Llama-Mimi 1.3B official model card |
| License | Llama 3.2 Community License.Llama-Mimi 1.3B official model card |
| Input and output | Audio prompt to generated audio continuation.Llama-Mimi 1.3B official model card |
Related models
- Llama-Mimi 8B
LLM-jp / NII LLMC · Llama-Mimi
An English speech-continuation model derived from Meta Llama 3.1 8B and Mimi audio tokenization.
Launch date not confirmedResearch releaseDownloadable - LLM-jp Moshi v1
LLM-jp / NII LLMC · LLM-jp Moshi
A Japanese full-duplex speech-dialogue derivative of Kyutai's Moshi model.
Launch date not confirmedResearch releaseDownloadable