trend-cybertron/Llama-Primus-Nemotron-70B-Base

TEXT GENERATIONPricing:Input $2.88 / Output $2.88Concurrent Unit Cost:4Model Size:70BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 21, 2025License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Llama-Primus-Nemotron-70B-Base is a 70 billion parameter language model developed by Trend Micro, built upon the NVIDIA Llama-3.1-Nemotron-70B-Instruct architecture. It has been pre-trained on over 10 billion tokens of large-scale cybersecurity corpora, including Primus-Seed-V2, Primus-FineWeb, and Primus-Nemotron-CC, and features a 32768 token context length. This model is specifically optimized for cybersecurity tasks, demonstrating an 11.19% aggregate improvement across various public cybersecurity benchmarks compared to its base model.

Loading preview...

Llama-Primus-Nemotron-70B-Base: Cybersecurity-Optimized LLM

Llama-Primus-Nemotron-70B-Base is a 70 billion parameter language model developed by Trend Micro, extending the nvidia/Llama-3.1-Nemotron-70B-Instruct architecture. This model is part of the Primus series, which focuses on open cybersecurity language models and datasets.

Key Capabilities & Optimizations

  • Cybersecurity Specialization: The model underwent extensive pre-training on over 10 billion tokens of large-scale cybersecurity corpora, including proprietary datasets like Primus-Seed-V2, Primus-FineWeb, and Primus-Nemotron-CC.
  • Enhanced Performance: It achieves an 11.19% aggregate improvement across several public cybersecurity benchmarks compared to the original Llama-3.1-Nemotron-70B-Instruct. This includes notable gains in CTI-Bench (CVE → CWE), CTI-Bench (ATE), and CISSP (Exam Questions).
  • Context Length: Features a substantial context window of 32768 tokens, suitable for processing detailed cybersecurity reports and logs.

Training Details

The pre-training utilized datasets specifically curated for cybersecurity knowledge:

  • Primus-Seed-V2: An enhanced version of Primus-Seed, incorporating blogs, news, books, websites, Wikipedia, MITRE, and Trend Micro knowledge.
  • Primus-FineWeb: Cybersecurity-specific text filtered from FineWeb-edu-score-2.
  • Primus-Nemotron-CC: Cybersecurity text filtered from Nemotron-CC.

Ideal Use Cases

This model is particularly well-suited for applications requiring deep understanding and generation within the cybersecurity domain, such as:

  • Cyber threat intelligence analysis.
  • Vulnerability assessment and reporting.
  • Security incident response.
  • Cybersecurity education and training platforms.

Finetunes
1 models