Deepsex

The 1st
RolePlay Model

Based on Deepseek-R1 with superpower enhanced role-playing capabilities, novel text generation, and Chain of Thought (CoT) reasoning.
We are proud to present the Tifa-Deepsex model!

NEW: Try DeepSex With Ollama

Tifa-DeepsexV2-7b-Cot-0301-F16NEW
huanlin/Tifa-DeepsexV2-7b-Cot-0301-F16
Tifa-DeepsexV2-7b-Cot-0301-Q4_KMNEW
huanlin/Tifa-DeepsexV2-7b-Cot-0301-Q4_KM
Tifa-DeepsexV2-7b-Cot-0301-Q8NEW
huanlin/Tifa-DeepsexV2-7b-Cot-0301-Q8
Tifa-DeepsexV2-7b-0218-Q4_KM
huanlin/Tifa-DeepsexV2-7b-0218-Q4_KM
Tifa-DeepsexV2-7b-F16
huanlin/Tifa-DeepsexV2-7b-F16
Tifa-DeepsexV2-7b-NoCot-0222-Q4_KM
huanlin/Tifa-DeepsexV2-7b-NoCot-0222-Q4_KM
Tifa-DeepsexV2-7b-Q4_KM
huanlin/Tifa-DeepsexV2-7b-Q4_KM
Tifa-DeepsexV2-7b-0218-F16
huanlin/Tifa-DeepsexV2-7b-0218-F16
Tifa-DeepsexV2-7b-Q8
huanlin/Tifa-DeepsexV2-7b-Q8
Tifa-Deepsex-14b-CoT-Q8
huanlin/Tifa-Deepsex-14b-CoT-Q8
Tifa-Deepsex-14b-CoT-Q4_K_M
huanlin/Tifa-Deepsex-14b-CoT-Q4_K_M
Tifa-Deepsex-14b-CoT-Crazy-Q8
huanlin/Tifa-Deepsex-14b-CoT-Crazy-Q8
Tifa-Deepsex-14b-CoT-Crazy-IQ4_NL
huanlin/Tifa-Deepsex-14b-CoT-Crazy-IQ4_NL
Tifa-Deepsex-14b-CoT-Chat-Q8
huanlin/Tifa-Deepsex-14b-CoT-Chat-Q8
...

Key Features

🤖 Enhanced Role-Playing
Improved character interactions and story generation

Trained on 0.4T tokens of novel content and 100K high-quality role-playing data for immersive character experiences.

🧠 Chain of Thought
Advanced reasoning capabilities

Implements sophisticated thought processes for complex problem-solving and storytelling coherence.

📜 128K Context
Extended context window

Supports ultra-long context for maintaining coherence in extended conversations and stories.

Model Versions

Tifa-Deepseek-14b-CoT
Verification model for testing RL reward algorithms

Initial version focused on flexible outputs and research purposes. Includes:

  • Base testing implementation
  • Uncontrolled but flexible responses
  • Research-oriented features

This series has many models, please visit here for details

Try Now!

Tifa-Deepsex-Cot-14B
HuggingFace Space

Technical Specifications

Base Architecture

Deepseek-R1-14B

Max Context

128k tokens

Training Data

0.4T novels + 100K SFT

Hardware

8×H20 GPU cluster

Acknowledgments
  • Shanghai Left-North Technology - Algorithms & Computing Power
  • Deepseek Team - GRPO Algorithm Sharing
  • Qwen Team - Excellent Open Source Foundation
  • Shanghai Fudan University
  • PRIME Team - Optimization Strategy