README.md

---
library_name: transformers
tags:
- TravelRisk
- Conflict
- Finetune
- ConflictAI
license: mit
datasets:
- Firemedic15/travel-risk-conflict-combined
language:
- en
metrics:
- accuracy
base_model:
- Qwen/Qwen3.5-9B
---

# Model Card for Firemedic15/Conflict-Risk-9B-Full

## Model Details

### Model Description

Conflict-Risk-9B-Full is a fine-tuned language model based on Qwen/Qwen3.5-9B. It was trained on approximately 50,000 records from a combined conflict and travel risk dataset to support risk-aware conversational analysis. The Qwen3.5-9B base model is a 9B-parameter model with native 262,144-token context support, published by Qwen on Hugging Face [web:27].

The model is intended to help users explore travel risk, conflict context, regional instability, and related analytical questions in a conversational format.

- **Developed by:** Firemedic15
- **Model type:** Causal language model
- **Language(s):** English
- **License:** MIT
- **Base model:** Qwen/Qwen3.5-9B
- **Dataset:** Firemedic15/travel-risk-conflict-combined

### Model Sources

- **Base model:** [Qwen/Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B)
- **Dataset:** [Firemedic15/travel-risk-conflict-combined](https://huggingface.co/datasets/Firemedic15/travel-risk-conflict-combined)
- **Model repository:** [Firemedic15/Conflict-Risk-9B-Full](https://huggingface.co/Firemedic15/Conflict-Risk-9B-Full)

## Uses

This model is designed to support risk-aware dialogue related to travel and conflict. The training objective was to improve performance on prompts involving regional instability, conflict dynamics, and travel-related threat awareness.

### Direct Use

Potential direct uses include:

- Conversational analysis of regional and global conflict conditions
- Travel risk awareness and qualitative threat discussion
- Summarization of conflict-related patterns and drivers
- Research support for risk-focused exploratory tasks

### Downstream Use

Potential downstream uses include:

- Risk intelligence assistants
- Travel advisory research prototypes
- Conflict monitoring and summarization tools
- Retrieval-augmented applications that combine model output with current external data

### Out-of-Scope Use

This model should not be used for:

- Real-time emergency response or evacuation decisions
- Sole-source operational security decisions
- Military targeting, surveillance, or harmful applications
- Legal, medical, or life-safety advice
- Any use case requiring guaranteed real-time factual accuracy without external verification

## Bias, Risks, and Limitations

This model may reflect biases present in the training data, including reporting bias, uneven regional coverage, and differing source quality across conflict and travel datasets.

Because the model was trained to be risk aware, it may overemphasize threat, instability, or negative framing in some contexts. It may also hallucinate specific facts, events, timelines, or causal explanations when asked for highly detailed or current information.

The model is not a substitute for verified reporting, official advisories, or expert human judgment.

### Recommendations

Users should treat this model as a supporting analytical tool rather than a source of truth.

Recommended safeguards include:

- Use current verified data sources for time-sensitive decisions
- Apply human review before acting on outputs
- Pair responses with retrieval or external validation where possible
- Avoid use in high-stakes environments without additional controls

## How to Get Started with the Model

Use the example below to load the model with Transformers:

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "Firemedic15/Conflict-Risk-9B-Full"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

prompt = "What are the main risk factors affecting travel in a conflict-affected region?"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(
    **inputs,
    max_new_tokens=200,
    temperature=0.7,
    do_sample=True
)

print(tokenizer.decode(outputs, skip_special_tokens=True))
```

## Training Details

### Training Data

The model was fine-tuned on approximately 50,000 records from the dataset `Firemedic15/travel-risk-conflict-combined`.

The dataset combines conflict-oriented records and travel risk data to help the model produce more context-aware and risk-sensitive responses.

### Training Procedure

The model was fine-tuned from Qwen/Qwen3.5-9B and later merged into a standalone checkpoint for deployment.

#### Preprocessing

Preprocessing likely included:

- Formatting examples into instruction or conversational pairs
- Normalizing text fields
- Combining conflict and travel-risk records into a unified training structure

#### Training Hyperparameters

- **Training regime:** bf16 mixed precision

Additional hyperparameters such as learning rate, batch size, LoRA rank, number of epochs, and optimizer settings can be added here if available.

#### Speeds, Sizes, Times

- **Training time:** 26 hours
- **Training environment:** Local machine

## Evaluation

### Testing Data, Factors, and Metrics

#### Testing Data

Evaluation data has not yet been fully documented. A recommended evaluation setup would include held-out prompts covering conflict analysis, regional instability, and travel-risk reasoning.

#### Factors

Relevant evaluation factors may include:

- Geographic region
- Conflict severity
- Prompt wording and complexity
- Time sensitivity
- Degree of uncertainty in the prompt

#### Metrics

- Accuracy

### Results

Formal benchmark results are not yet documented in this card.

The model is intended to improve risk-aware reasoning over the base model for conflict and travel-risk related prompts, but outputs should still be validated by human review and external sources where reliability matters.

## Environmental Impact

Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute).

- **Hardware Type:** NVIDIA GeForce RTX 5070 Ti Blackwell
- **Hours used:** 26
- **Cloud Provider:** Local machine
- **Compute Region:** Local
- **Carbon Emitted:** Not estimated

## Technical Specifications

### Model Architecture and Objective

This model is a fine-tuned causal language model based on Qwen/Qwen3.5-9B. According to the base model page, Qwen/Qwen3.5-9B is a 9B-parameter model with 32 layers and a native context length of 262,144 tokens [web:27]. Its objective in this fine-tuned form is to provide improved performance on prompts related to conflict awareness, travel risk, and risk-informed dialogue.

### Compute Infrastructure

The model was trained locally on consumer GPU hardware for approximately 26 hours.

#### Hardware

- NVIDIA GeForce RTX 5070 Ti Blackwell

#### Software

- Transformers
- PyTorch
- PEFT
- Accelerate

## Citation

**BibTeX:**

```bibtex
@misc{firemedic15conflictrisk9bfull,
  title        = {Conflict-Risk-9B-Full},
  author       = {Firemedic15},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/Firemedic15/Conflict-Risk-9B-Full}},
  note         = {Fine-tuned from Qwen/Qwen3.5-9B on combined conflict and travel-risk data}
}
```

**APA:**

Firemedic15. (2026). *Conflict-Risk-9B-Full* [Computer software]. Hugging Face. https://huggingface.co/Firemedic15/Conflict-Risk-9B-Full

## More Information

This model is best used as part of a broader analytical workflow that includes current data sources, human review, and domain-specific validation.

## Model Card Authors

Firemedic15

## Model Card Contact

Firemedic15 via Hugging Face Hub