When we released LLAMA-3,1-FOUNDATIONAI-SECURITYLM-BASE-8B (Foundation-SeC-8B) In April, we showed that a model-exclusive model with eight billion parameters on security data-outperform general purpos LLMS Mayry is gaining in size on cyber security benchmarks. He loved the results of practice, yet thesy still asked:
“Can I communicate with it as a chatgpt – no more scaffolding, just fast and go?”
Now you can. LLAMA-3,1-NALIDAIDAI-SECURITYLM-INTRUCT-8B (Foundation-Sec-8B-Instruct) Teaching layers of fine fine -tuning on the upper part of our Basel model focused on the domain, which gives you a chat copilot that understands the context of security and follows straight instructions of natural natural straight straight from the box.
Built for security, tuned to conversation
- The following instructions -Foundation-Sec-8B-Instruct is governed by daily challenges to summarize, answer questions, analysis of feeling and creative generation of text without any special tuning.
- Roll awareness -System assistant, users and roles, multiple cats allowing cats, searches-on-August generation and agent-style application.
- Compact track -A 4k-token context window allows Foundation-SEC-8B-Instructu-Instruct to run on one, highly commemorative GPU, while pushing to much longer windows.
- Open-source and light -Similarly to its sibling, Foundation-SEC-8B-Instropic ships as a fully open weight under a permissive license sufficiently sufficiently for deployment on bounds, air or edges without locking the supplier.
That it produces greater power output
Foundation-SeC-8B briefs heavy weight security in the form of eight billion parameters. In independent evaluations, it overcomes larger models while remaining small enough to run on one high -memorial GPU.
| Benchmark | Instruction Foundation-SEC-8B | LLAMA-3.1-8b-Instruct | LLAMA-3.1-70B-INSTRUCT | GPT-4-Mini |
|---|---|---|---|---|
| Honor | 0.692 | 0.558 | 0.623 | 0.655 |
Equally important is that the powerful performance of the general language based on-8b on wide tests, such as MMLU. This means that it can fluently explain the behavior of the threat, long-term narration and reaction naturally in the workflows of chat-inz threats to its safety focus or demanding excessive infrastructure.
Where is it suited today
Security Operation Cers (SOC) Facial warning noise, framed tools and chronic gaps in staff. For pilots across Cisco Csirt and Cisco XDR, Foundation-SeC-8B-Instruct-Instruct classifies notifications, maps observable for tactics, time-time reconstruction, and investigative reports that cut false positive and accelerating sorting, so they can focus on real threats.
Security Application (AppSEC) Teams juggle inspection code, threat models and red team planning with limited bandwidth. Early deployment on SBG, Meraki and S&P’s Web-Application Security Group uses Foundation-SEC-8B instruction to strengthen offensive paths, generate diagrams of threat model, control against OWASP instructions, and craft useful load, and shifting fixed to proactive design.
Becuse Foundation-SeC-8B-Instruct is tuned to instruction, most workflows need only a well-created challenge, or a simple search loop is required.
We look forward to seeing ahead
The Foundation-SEC-8B is a milestone, but it is far from the final goal. Our plan focuses on unlocking a richer context, stricter structures and wider data, so the model will be able to handle all the defenders of artifacts on which they rely on the next infrastructure. You can expect:
- Along the context windows – 16k+ tokens for full books, pooms and log webles
- Structured interaction -Schema proven JSON, explicit function calls and even code or binary generation
- Multimodal inputs – protocols, catches catches, screenshots and other artifacts in one conversation
- New generation models -Derate-Optimized variant of eight billion parameters and upcoming model seventy billion parameters
Start building today
- Grab the scales – Download Instruction Foundation-SEC-8B To hug the face and run it on-pray, in secure cloud enclaves or in laboratories with a gap in the air-no license restrictions.
- Watch recipes – Cookbook Foundation AI Cookbook Offs Deployment Guides, Statutes of Templates and Example Agent.
- Join the community -Pilot new workflows, calls on fast or fine tons and feeds your findings back to an open source code.
Instruction Foundation-SEC-8B It is alive, open and ready for defense. Download it, challenge it and help shape the future of cyber security on AI-peal.
Contributors: Sajana Weerawardhen, Paul Kassianik, Blaine Nelson, Batuly Saglam, Anu Vellore, Aman Priyanshu, Supriti Vijay, Massimo Aufifo, Fraser Burch, Arthur Goldblatt, Konstantin Goldin, Alie Fordyce, Alie Fordyce, Alie Ford Oshib
We would like to hear what you think! Ask and stay in conjunction with Cisco Security on social media.
Cisco Security Social Media
LinkedIn
Facebook
Instagram
X
Share:
(Tagstotranslate) Foundation AI