SmolLM3
SmolLM3 is a 3B open model with dual-mode reasoning: the same weights answer directly or produce an extended chain of thought depending on a flag in the chat template. It handles six languages, supports 64K context extendable to 128K, ships with the full training recipe published, and runs locally on a laptop or a small GPU under Apache 2.0.
- Type
- Text
- Weights
- Open
More from Hugging Face
6
-
1.2k DL
Carbon-8B
Text
Carbon-8B is an open-source 8B generative DNA foundation model for DNA/RNA sequence generation, scoring, and…
-
37 DL
Distil Whisper
Multimodal
Distil Whisper is a distilled version of OpenAI's Whisper speech recognition model, roughly six times faster…
-
918 DL
nanowhale
Text
nanowhale is a miniature DeepSeek-V4-style language model project, built to reproduce the core DeepSeek-V4 ar…
-
SmolLM2
Text
SmolLM2 is a family of very small open language models at 135M, 360M and 1.7B parameters, trained on 11 trill…
About Hugging Face
1
Hugging Face is an Artificial Intelligence company that specializes in Machine Learning, Natural Language Processing, and AI Chatbots.