Models, not just a resume line
I fine-tune open-weight language models for security and geopolitical risk analysis, then ship the weights publicly on Hugging Face. Below are the active families; the full catalog (40+ models, including course exercises and paper reproductions) is on my Hugging Face profile.
Qwen2.5 Risk Models
Three sizes of Qwen2.5, supervised fine-tuned (TRL/SFT) on geopolitical and security risk data, so the right size can be picked for the deployment: a 3B model for accuracy-sensitive workloads, a 1.5B GGUF build for fast local/edge inference, and a 0.5B model for tight latency or cost budgets.
Qwen2.5 Risk 3B
The largest of the three, merged and ready to serve, for workloads where accuracy matters more than latency.
View on Hugging FaceQwen2.5 Risk 1.5B
Quantized (GGUF) for fast local and edge inference, same risk-analysis fine-tune at a lighter footprint.
View on Hugging FaceQwen2.5 Risk 0.5B
The smallest of the family, merged and ready to serve, for tight latency or cost budgets.
View on Hugging FaceTravel Risk Analyst Models
Two different base architectures fine-tuned on the same travel-risk analysis task, comparing Qwen2.5 against Nvidia's Nemotron line for this use case.
Qwen2.5-1.5B Travel Risk
Fine-tuned for travel risk analysis: reads a destination and context, returns a structured risk read.
View on Hugging FaceNemotron-mini-4B Travel Risk
The same travel-risk task, fine-tuned on Nvidia's Nemotron-mini-4B for a larger-capacity comparison point.
View on Hugging FaceConflict-Risk Models
Qwen3.5 fine-tunes for geopolitical conflict risk, the research line behind Conflict GPT.
Conflict-Risk-9B
Supervised fine-tune of Qwen3.5-9B for geopolitical conflict and risk assessment.
View on Hugging FaceConflict-Risk-4B
The lighter-weight counterpart, supervised fine-tune of Qwen3.5-4B for the same conflict-risk task.
View on Hugging FaceWant the full catalog, including reinforcement-learning course models and paper-reproduction work? Browse huggingface.co/Firemedic15, or ask my chat assistant.