This model card introduces MistralCowboy-7b-v0.1, a state-of-the-art language model created by Phanerozoic. It stands as our finest achievement, expertly fine-tuned to emulate the distinctive style and lingo of a classic cowboy. This model excels in bringing the unique cowboy vernacular and worldview to life in engaging and entertaining conversations.
Model Description
Developer: Phanerozoic
License: cc-by-nc-4.0
Finetuned from: OpenHermes 2.5
Primary Use: Ideal for applications that require dialogues in the distinct cowboy style, such as interactive storytelling, virtual role-playing games, and educational platforms exploring historical and cultural aspects of the American West.
Direct Use
MistralCowboy-7b-v0.1 is intended for immersive experiences where users interact with a virtual cowboy, gaining insights into the lifestyle, language, and ethos of the old West.
Downstream Use
While primarily designed for direct interaction, MistralCowboy-7b-v0.1 can be adapted for tasks that benefit from a rugged, Western-themed narrative style.
Out-of-Scope Use
This model is not suited for modern, factual, or technical topics unrelated to the cowboy theme. Its expertise lies in historical and cultural representations of the cowboy era.
Bias, Risks, and Limitations
As it's trained on stylized, historical language, MistralCowboy-7b-v0.1 may reflect the biases and limitations of its era. This should be considered in educational and cultural settings.
Recommendations
Users are encouraged to provide context at the start of conversations for optimal performance. Custom stopping strings are advised to prevent overrun and maintain relevancy.
Custom Stopping Strings Usage
Recommended stopping strings include:
"},"
"User:"
"You:"
""\n"
"\nUser"
"\nUser:"
These aid in delineating responses and maintaining the cowboy dialogue structure.
Training Data
The model was trained on a mix of historical texts, cowboy literature, and scripted dialogues, ensuring a rich and authentic cowboy vocabulary and style.
Preprocessing
Datasets were formatted for consistent, structured input, focusing on emulating the cowboy communication style.
Training Hyperparameters
Training Regime: FP32
Warmup Steps: 1
Per Device Train Batch Size: 1
Gradient Accumulation Steps: 32
Max Steps: 1000
Learning Rate: 0.0002
Logging Steps: 1
Save Steps: 1
Lora Alpha: 16
Dimension Count: 8
Speeds, Sizes, Times
Training was completed in about 10 minutes using an RTX 6000 Ada GPU.
Testing Data
Evaluated against a range of cowboy-themed texts, demonstrating an excellent grasp of the style and content.
Factors
Focus on maintaining coherent, era-appropriate responses in cowboy vernacular.
Metrics
Emphasis was on model's ability to accurately and engagingly replicate cowboy speech.
Results
The model excels in consistently delivering authentic, engaging cowboy-style dialogues.
Performance Highlights
MistralCowboy-7b-v0.1 is our best performing model to date, expertly capturing the essence of cowboy talk in its responses.
Summary
MistralCowboy-7b-v0.1 is a significant leap forward in creating specialized language models that can authentically represent cultural and historical figures, in this case, the iconic American cowboy.
Model Architecture and Objective
Built upon the Mistral model architecture, further refined with LoRA modifications to encapsulate the cowboy communication style.
Compute Infrastructure
Hardware Type: RTX 6000 Ada GPU
Training Duration: Approximately 10 minutes
Acknowledgments
Special thanks to the Mistral team and OpenHermes 2.5 team. Our innovative use of LoRA techniques showcases the collaborative advancement in AI and language modeling, bringing the cowboy era to virtual life.
