From the model card
Fields below are copied from the tags and counters on the HuggingFace repository LiquidAI/LFM2.5-1.2B-Instruct at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- LiquidAI
- Pipeline tag
- text-generation
- Library
- Transformers
- Weight formats
- safetensors
- License tag
other— read the license file in the repo before relying on it- Lineage
-
- base model LiquidAI/LFM2.5-1.2B-Base
- fine-tune of LiquidAI/LFM2.5-1.2B-Base
- Language tags
- English (en), Arabic (ar), Chinese (zh), French (fr), German (de), Japanese (ja), Korean (ko), Spanish (es)
- Papers cited
- arXiv:2511.23404
- Downloads (HF counter at last fetch)
- 408,065
- Likes (HF counter at last fetch)
- 661
- Model card
- https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct
Use cases
- Edge inference where transformer alternatives are being evaluated
- Long-context tasks that benefit from recurrent memory efficiency
- Multilingual instruction following across 9 languages at small model size
- Research comparing hybrid recurrent-attention architectures
Pros
- Hybrid recurrent-attention architecture offers alternative to pure transformers
- 9-language multilingual support
- 1.2B size suitable for constrained hardware
- Transformers-compatible inference
Cons
- 'Other' license — verify commercial use rights with Liquid AI
- 1.2B parameters limits practical task complexity
- Architecture novelty means less community tooling and fine-tuning recipes
- Eval results tag present but not independently verified against standard benchmarks
Tags
transformerssafetensorslfm2text-generationliquidlfm2.5edgeconversationalenarzhfrdejakoesarxiv:2511.23404base_model:LiquidAI/LFM2.5-1.2B-Basebase_model:finetune:LiquidAI/LFM2.5-1.2B-Baselicense:other