Tiny Aya L2-Thinker

People can use it to solve tasks such as math, cultural reasoning, instruction following, and open-ended writing while inspecting reasoning traces in the prompt language.

Availability

Model weights released on Hugging Face; try it on CohereLabs Space

Pricing model

No paid plan required

Deployment

Weights released on Hugging Face

Privacy

Not established

Licence

CC-BY-NC; also requires adherence to Cohere Labs' Acceptable Use Policy

• It supports in-language reasoning for 44 languages plus English. ^[1]

  It is trained to support in-language reasoning for 44 languages plus English, with coverage extending to 20+ more through additional non-reasoning instruction data.

• The model supports dual-mode reasoning, with thinking enabled by default and an option to disable it. ^[1]
• The model repository is publicly accessible, but users must accept the conditions to access its files and content. ^[1]

  This repository is publicly accessible, but you have to accept the conditions to access its files and content .

• The model has a 32K context length covering input and output. ^[1]

  Context Length: 32K (input + output)

• The stated license is CC-BY-NC and also requires adherence to Cohere Labs' Acceptable Use Policy. ^[1]

  License: CC-BY-NC , requires also adhering to Cohere Lab's Acceptable Use Policy

• The model is licensed under CC-BY-NC and requires adherence to Cohere Labs' Acceptable Use Policy. ^[2]
• It supports mathematics, science, general reasoning, instruction following, and multilingual open-ended generation. ^[2]

  The model is designed to support mathematics, science, and general reasoning tasks, as well as instruction following and multilingual open-ended generation.

• The model supports dual-mode reasoning, and users can disable thinking mode. ^[2]

  The model supports dual-mode reasoning .

• The model files are available from the public Hugging Face repository after accepting the repository conditions. ^[2]
• On five of six benchmarks, accuracy drops by at most two to three points compared with Tiny Aya En-Thinker. ^[3]

  On five of six benchmarks it drops by at most two to three points, and on MIST-OEG, our open-ended writing benchmark, it actually goes up slightly.

• Cohere released Tiny Aya L2-Thinker weights on Hugging Face. ^[3]
• Tiny Aya L2-Thinker can be tried on CohereLabs Space. ^[3]

  Tiny Aya L2-Thinker: https://huggingface.co/CohereLabs/tiny-aya-l2-thinker

• The model reasons in the prompt language over 93% of the time across 60 languages. ^[3]

  Our resulting model, Tiny Aya L2-Thinker (3.35B), reasons in the prompt language over 93% of the time across 60 languages.

• It supports 45 languages in training. ^[3]

  It also supports 45 languages in training, with evidence that it can carry this ability over to languages it was never shown reasoning examples in.