Hugging Face's logo

umer07
/
fathom-mixtral

PEFT
Safetensors
English
cybersecurity
malware-analysis
att&ck
threat-intelligence
mixtral
lora
expert-adapters
cape-sandbox
digital-forensics
Model card Files
xet
 Community
1
umer07 commited on
Commit
662fc41
·
verified ·
1 Parent(s): b9d9d3f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -33,7 +33,7 @@ metrics:
33
  ## Model Overview
34
 
35
  - **Base:** Mixtral-8x7B-Instruct-v0.1 (full bf16, no quantization)
36
- - **Training:** Direct PEFT+TRL (LlamaFactory dropped due to ROCm issues)
37
  - **Adapters:** 1 unified + 9 expert LoRA adapters (all rank=32, α=16)
38
  - **Hardware:** AMD MI300X (205.8 GB VRAM) — full bf16 training
39
  - **Key Innovation:** Evidence extraction layer + structured behavioral prompts → **9× improvement** in real ATT&CK mapping
 
33
  ## Model Overview
34
 
35
  - **Base:** Mixtral-8x7B-Instruct-v0.1 (full bf16, no quantization)
36
+ - **Training:** Direct PEFT+TRL
37
  - **Adapters:** 1 unified + 9 expert LoRA adapters (all rank=32, α=16)
38
  - **Hardware:** AMD MI300X (205.8 GB VRAM) — full bf16 training
39
  - **Key Innovation:** Evidence extraction layer + structured behavioral prompts → **9× improvement** in real ATT&CK mapping