dufuspaelli/Llama-Poro-2-8B-Long-Instruct-heretic

TEXT GENERATIONPricing:Input $0.37 / Cached $0.074 / Output $0.38Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Aug 26, 2026License:llama3.3Architecture:Transformer Featherless Exclusive Cold

dufuspaelli/Llama-Poro-2-8B-Long-Instruct-heretic is an 8 billion parameter instruction-following chatbot model, based on LumiOpen's Llama-Poro-2-8B-Long-Instruct, with an extended context length of 8192 tokens. This version is a decensored variant created using Heretic v1.4.0, demonstrating significantly reduced refusals compared to the original. It is designed for conversational AI applications and instruction following in both Finnish and English, supporting long-context tasks.

Loading preview...

Model Overview

dufuspaelli/Llama-Poro-2-8B-Long-Instruct-heretic is a decensored version of the LumiOpen/Llama-Poro-2-8B-Long-Instruct model, created using Heretic v1.4.0. This 8 billion parameter model is an instruction-following chatbot with extended context support, designed for conversational AI in both Finnish and English. It supports context lengths up to 128K tokens, though this specific model's context is noted as 8192 tokens.

Key Differentiators

  • Decensored Variant: This model is explicitly noted as a decensored version, achieving a refusal rate of 5/100 compared to the original model's 87/100 refusals.
  • Bilingual Support: Optimized for instruction following and conversation in both English and Finnish.
  • Long Context: The base model family supports up to 128K tokens, with this variant leveraging that capability for tasks like document summarization and multi-document Q&A.

Performance Highlights

While the original Poro 2 Long Instruct model shows strong performance in Finnish instruction-following benchmarks (e.g., MTBench FI: 6.18, IFEval FI: 63.4), it generally performs lower than Llama 3.1 8B Instruct in English benchmarks and long-context evaluations like HELMET and OneRULER. The primary distinction of this 'heretic' version lies in its significantly reduced refusal rate.

Intended Use Cases

  • Conversational AI applications in Finnish and English
  • Long-context tasks such as document summarization and analysis
  • Question answering and information retrieval
  • Content generation and creative writing
  • Educational applications and customer service

Finetunes
2 models