Skip to content

EthenEthenEthen

LoginPlatformTry Ethen

Open Source Model Profile · NbAiLab

nb-notram-llama-3.2-3b-instruct

nb-notram-llama-3.2-3b-instruct is a 3.21B-parameter Llama fine-tune from NbAiLab. According to the model card, it adapts Llama-3.2-3B-Instruct for Norwegian Bokmal and Nynorsk instruction following.

Publisher
NbAiLab
Task
text-generation
Model type
llama
License
llama3.2
Library
transformers
Publication status
Approved for indexing

Model overview

The model is published by NbAiLab as part of the NB-Llama-3.x and NoTraM lines. The captured configuration identifies LlamaForCausalLM with model type llama, and Safetensors metadata reports 3,212,749,824 parameters. According to the model card, it is fine-tuned from Meta's Llama-3.2-3B-Instruct using only publicly available data.

Recorded capabilities

Norwegian instruction focus

According to the model card, the model improves instruction following in Norwegian Bokmal and Nynorsk while aiming to preserve English performance.

SFT plus light DPO recipe

According to the model card, alignment is primarily supervised fine-tuning with a very light DPO step to stabilize instruction following.

Concise response style

According to the model card, the model tends toward shorter, more concise answers reflecting its instruction-tuning mix.

Public-data training scope

According to the model card, training uses only publicly available datasets and synthetic data, excluding legal-deposit-only material.

Use cases in the source record

  • Norwegian dialogue and assistant applications in Bokmal or Nynorsk, plus English assistance, as listed in the card's intended use.
  • Bokmal and Nynorsk summarization and Q&A workflows covered by the card's suitable-use scope.

Limitations and unknowns

  • According to the model card, Norwegian cultural and historical knowledge may be uneven or prompt-sensitive, described as knowledge pocketing.
  • According to the model card, the model can hallucinate and should not be relied on as a sole source of truth or for high-stakes domains without controls.
  • No evaluation results were extracted as structured data from this record.
  • Provider state is historical snapshot data, not independently refreshed current availability.

Source and provenance

Source: NbAiLab/nb-notram-llama-3.2-3b-instruct

Captured: Unknown. Processed: 2026-09-07T19:34:34.708283+00:00.

Model Card: "nb-notram-llama-3.2-3b-instruct" Model overview "NbAiLab/nb-notram-llama-3.2-3b-instruct" is part of the "NB-Llama-3.x" series (covering "Llama 3.1", "Llama 3.2", and "Llama 3.3" based releases) and the "NoTraM" line of work, trained on top of Meta’s "Llama-3.2-3B-Instruct": https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct The model is fine-tuned to improve instruction-following behavior in Norwegian Bokmål and Norwegian Nynorsk, while aiming to preserve strong English performance. (It may also handle some Swedish and Danish in practice, but those are not primary targets for this release.) This release is an expe…

F001F002F003F004F005F006F007F009F010F011F012F013F014F016F019F020F025F026F027