Psyfighter v2 13B

Model Overview

Property	Value
Model Name	Psyfighter v2 13B
Inference Model ID	koboldai/psyfighter-13b-2
Creator/Organization	KoboldAI
Authors	Jeb Carter, TwistedShadows (community members)
Release Date	December 8, 2023
Model Type	Text Generation / Creative Writing

Description

A specialized merged model designed for enhanced fictional storytelling with supplementary medical knowledge. The model combines three base models to balance creative narrative generation with anatomical and mental health terminology.

Key Purpose: Enhancing fictional narratives with medical terminology for realistic descriptions of anatomy and mental states in storytelling contexts.

Important Safety Warning

This model should NOT be used for medical advice or therapy due to its high likelihood of incorporating fictional data alongside medical information. Users should treat medical outputs as creative content only, not as authoritative medical guidance.

Technical Specifications

Specification	Details
Base Architecture	LLaMA 2 13B variant
Parameter Count	13 Billion
Context Window	4,096 tokens
Input Modality	Text
Output Modality	Text
Instruction Format	Alpaca
Default Stop Sequences	###, </s>

Capabilities

Capability	Status
Text Generation	Yes
Creative Writing	Yes
Medical Knowledge	Yes (supplementary, not authoritative)
Conversation	Yes
Reasoning	No
Function Calling	No
Vision/Images	No

Related Models

Similar Creative Writing Models:

• Tiefighter (base model, pure creative writing focus)
• OpenChat (general purpose)
• Mistral 7B (smaller, faster alternative)

Alternative Medical-Aware Models:

• Medalpaca (medical domain-specific)
• BioGPT (biomedical text generation)
• PubMedBERT (biomedical understanding)

Component Models

The Psyfighter v2 13B is a merge of three base models:

1. KoboldAI/LLaMA2-13B-Tiefighter

   • Base creative writing model

2. Doctor-Shotgun/cat-v1.0-13b

   • Medical and anatomical knowledge

3. Doctor-Shotgun/llama-2-13b-chat-limarp-v2-merged

   • Conversational capabilities with refined responses

Model Weights

Model weights are publicly available on Hugging Face:

• Repository: KoboldAI/LLaMA2-13B-Psyfighter2
• License: Check Hugging Face repository for current licensing information

Performance Characteristics

• Supports Reasoning: No
• Deprecation Status: Active (not deprecated as of December 2023)
• Usage Analytics: Insufficient activity for detailed usage data display

Recommended Use Cases

Ideal For:

• Fictional storytelling and creative writing
• Character development with realistic medical/anatomical descriptions
• Narrative generation for games, novels, and creative projects
• Dialogue generation for fictional characters
• World-building with medically accurate terminology

Not Recommended For:

• Medical advice or diagnosis
• Healthcare applications
• Mental health therapy or counseling
• Legal or regulatory compliance
• Production systems requiring high reliability
• Real-time applications (context window limitation)

Model Configuration

Alpaca Instruction Format

The model uses the Alpaca instruction template:

Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
{instruction}

### Response:
{response}

Stop Sequences

Generate text until encountering:

• ### (instruction boundary marker)
• </s> (end-of-sequence token)

Set both as stop tokens in your inference engine.

Implementation Notes

Merging Technique

The model is created using model merging/mixing techniques that combine the strengths of multiple base models. This allows it to:

• Retain creative writing capabilities from Tiefighter
• Incorporate medical knowledge from cat-v1.0
• Maintain conversational quality from limarp

Knowledge Integration

Medical terminology is integrated into the model's knowledge base through the merging process. The quality of medical information should be verified before use in creative contexts where accuracy matters.

Pricing & Availability

• Access: Available through LangMart and other providers
• Pricing: Check current provider listings for token pricing (input/output rates vary by provider)
• Inference Providers: LangMart (primary), compatible with other providers supporting LLaMA 2 13B variants

Resource Requirements

Hardware Specifications

For local deployment:

• Minimum VRAM: 27-32GB (full precision) or 13-16GB (quantized)
• Optimal Setup: NVIDIA GPU with 48GB+ VRAM or multiple GPUs
• CPU Fallback: Possible but very slow (not recommended for production)

Performance Metrics

• Context Window: 4,096 tokens (relatively small, suitable for short to medium documents)
• Throughput: Variable (depends on hardware and inference engine)
• Latency: Moderate (13B is considered mid-size for inference speed)

Integration Guide

LangMart API Example

curl https://api.langmart.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "koboldai/psyfighter-13b-2",
    "messages": [
      {
        "role": "user",
        "content": "Write a scene where a character discovers they have an unusual medical condition."
      }
    ],
    "temperature": 0.7,
    "max_tokens": 1024
  }'

Local Deployment (llama.cpp)

# Download GGUF quantized version
# Run with llama.cpp
./main -m psyfighter-13b-2-q4_K_M.gguf \
  -n 1024 \
  -p "Below is an instruction..." \
  --repeat_penalty 1.1

Version History

Version	Release Date	Notes
v2	December 8, 2023	Current version - merged model with 3 base models
v1	Earlier	Original Tiefighter model (predecessor)

Resources & References

• Documentation: https://langmart.ai/model-docs
• Model Weights: https://huggingface.co/KoboldAI/LLaMA2-13B-Psyfighter2
• Base Architecture: LLaMA 2 Research Paper (Meta AI)
• Creator: KoboldAI Community

FAQ

Q: Can I use this for medical applications?

A: No. While the model has medical knowledge, it's trained to blend medical terminology with fiction. It should never be used for actual medical advice, diagnosis, or treatment.

Q: What's the difference from base LLaMA 2 13B?

A: Psyfighter v2 is a specialized merge that adds medical knowledge and enhanced conversational abilities while maintaining creative writing strengths.

Q: How does the context window compare?

A: At 4,096 tokens, it's standard for base LLaMA 2 but shorter than newer models (many are 8K-200K). Good for short stories and dialogues, limited for long documents.

Q: Can it generate images?

A: No, this is a text-only model. Use DALL-E, Stable Diffusion, or other vision models for image generation.

Q: Is it suitable for production use?

A: This depends on your use case. For creative writing and storytelling, yes. For reliability-critical systems, consider more established models with better documentation and support.

Updates & Maintenance

• Last Updated: Based on LangMart listing (December 2023)
• Status: Active community model
• Support: Community-supported through KoboldAI channels

---

Documentation generated from LangMart model listing. For the most current information, visit the official LangMart page or Hugging Face repository.