license: apache-2.0
language:
- en
- fr
- it
- de
- es
base_model:
- PleIAs/Pleias-350m-Preview
Pleias-RAG-350m
Pleias-RAG-350m is a phone-sized Small Reasoning Model, trained for retrieval-augmented general (RAG), search and source summarization. Along with Pleias-RAG-1B it belongs to the first generation of Pleias reasoning models for RAG.
Pleias-RAG-350m outperform most SLMs below 4 billion parameters on standardized benchmarks for retrieval-augmented general (HotPotQA, 2wiki) and is competitive with popular larger models, including Qwen-2.5-7B, Llama-3.1-8B and Gemma-3-4B. It is the only SLM to date to maintain consistent RAG performance across leading European languages and to ensure systematic reference grounding for statements.
Due to its size, and ease of deployment on constrained infrastructure (even mobile phone) and built-in support for factual and accurate information, Pleias-RAG-350m unlock a range of new use cases for generative AI. It is ideally deployed in the context of externalized memory, where the model is primarily conceived to process external sources rather than relying on its internal knowledge.
Training
Pleias-RAG-350m is trained on large synthetic dataset emulating retrieval of wide variety of multilingual open sources from Common Corpus. They provide native support for citation and grounding with literal quotes. Following on the latest trends of agentification, the models reintegrate multiple features associated with RAG workflows such as query routing, query reformulation, source reranking.