diff --git "a/ui.html" "b/ui.html" new file mode 100644--- /dev/null +++ "b/ui.html" @@ -0,0 +1,799 @@ + + + + + +🔐 CyberLog-GPT + + + + + +
+ +
+ Loading model... + CPU +
+
+ +
+ + + + + +
+ + + + + +
+
Generated Security Logs
+
+ + +
+
+ + +
+ + + + + +
+Configure settings in the sidebar and click ⚡ Generate Logs + +Supported attack types: + • SSH Brute Force • Port Scan + • Firewall Block • Web Attack (SQLi/XSS) + • Malware C2 Beacon • Privilege Escalation + • Data Exfiltration • Ransomware Activity + • Vulnerability Exploit • SIEM Alert + +Useful for: + → Testing SIEM detection rules + → SOC analyst training + → CTF challenge preparation + → Security demo datasets +
+ +
+
+ + + +YPE html> + + + + +🔐 CyberLog-GPT + + + + + +
+ +
+ Loading model... + CPU +
+
+ +
+ + + + + +
+ + + + + +
+
Generated Security Logs
+
+ + +
+
+ + +
+ + + + + +
+Configure settings in the sidebar and click ⚡ Generate Logs + +Supported attack types: + • SSH Brute Force • Port Scan + • Firewall Block • Web Attack (SQLi/XSS) + • Malware C2 Beacon • Privilege Escalation + • Data Exfiltration • Ransomware Activity + • Vulnerability Exploit • SIEM Alert + +Useful for: + → Testing SIEM detection rules + → SOC analyst training + → CTF challenge preparation + → Security demo datasets +
+ +
+
+ + + + +
+
+
+ + Model Training Results + +
+ + +
+
+
0.24
+
Final Loss
+
+
+
94%
+
From Baseline
+
+
+
5.2M
+
Parameters
+
+
+
~15m
+
Train Time
+
+
+ + +
+
+
+ Training Loss Curve +
+ Training Loss +
+
+
+ Attention Weights — Layer 0 +
+ Attention Weights +
+
+ +
+ 6-layer GPT transformer · 8 attention heads · 256 embedding dim · trained on T4 GPU · PyTorch from scratch +
+
+
+ + + +
+
+ + +
+
+
+
Training Results
+
Model trained from scratch · PyTorch · T4 GPU
+
+
+ + +
+
+
0.24
+
Final Loss
+
+
+
94%
+
↓ From Random
+
+
+
5.2M
+
Parameters
+
+
+
~15m
+
Train Time
+
+
+ + +
+ +
+
+ Loss Curve + 4.41 → 0.24 +
+ Training Loss Curve +
+ Train loss (blue) and val loss (red dashed) nearly identical → no overfitting. + Model learned 94% of log structure from random baseline. +
+
+ +
+
+ Attention Weights + Layer 0 · 8 Heads +
+ Attention Weight Heatmaps +
+ Each head attends to different character patterns. Diagonal = local attention. + Red = strong attention. Different heads learn different log structure rules. +
+
+ +
+ + +
+ 6 transformer layers + · + 8 attention heads + · + 256 embedding dim + · + 256 char context + · + character-level tokenizer + · + same architecture as GPT-2 +
+ +
+
+ + + \ No newline at end of file