erukude/ cve-expert:70b

1,072 1 week ago

A fine-tuned DeepSeek-R1-70B model for CVE-worthiness review, exploit-path analysis, and JSON-only vulnerability triage.

thinking 70b 
ollama run erukude/cve-expert:70b

Details

1 week ago

808be7420620 · 43GB ·

model
llama
·
70.6B
·
Q4_K_M
license
MIT License Copyright (c) 2023 DeepSeek Permission is hereby granted, free of charge, to any person
system
You are a senior vulnerability researcher and CVE analyst with 15+ years of experience in applied se
params
{ "presence_penalty": 1.5, "stop": [ "<｜begin▁of▁sentence｜>", "<｜e
template
{{- if .System }}{{ .System }}{{ end }} {{- range $i, $_ := .Messages }} {{- $last := eq (len (slice

Readme

cve-expert

erukude/cve-expert is a fine-tuned DeepSeek-R1-based model for CVE analysis, vulnerability impact review, and advisory-quality security triage.

What It Does

This model acts as a senior vulnerability researcher and CVE analyst. It reviews security findings, checks whether they meet the CVE bar, and returns a conservative, JSON-only verdict.

Best For

  • CVE-worthiness review
  • Vulnerability impact assessment
  • Exploit-path analysis
  • False-positive reduction
  • Responsible disclosure and advisory triage

Verdicts

  • cve_worthy - realistic, novel, exploitable, meaningful impact
  • actionable - real issue, worth reporting, but not CVE-level
  • low_impact - genuine issue, limited practical impact
  • false_positive - not a real vulnerability

Run

ollama run erukude/cve-expert

Example Prompt

Assess this finding for CVE-worthiness.

Package: example-lib
Version: 1.2.3
Finding: ...
Reachability: ...
Attacker control: ...
Impact: ...
Prior analysis: ...

Model Details

  • Base: DeepSeek-R1
  • Architecture: llama
  • Parameters: 70.6B
  • Context length: 131072
  • Embedding length: 8192
  • Quantization: Q4_K_M
  • Format: GGUF
  • Temperature: 0.2
  • Top-p: 0.95
  • Top-k: 20
  • Output: valid JSON only
  • License: follow the upstream DeepSeek-R1 license and any adapter/model distribution terms

Developer

Created and maintained by Sai Teja Erukude.

Note

Built for defensive security research, responsible disclosure, and advisory triage. The model is intentionally conservative before marking anything as cve_worthy.