mirror of
https://github.com/zenlm/papers.git
synced 2026-07-27 03:34:41 +00:00
330 lines
8.7 KiB
Markdown
330 lines
8.7 KiB
Markdown
# Zen Model Family Research Paper Timeline
|
||
|
||
This document tracks the proper dating of Zen model papers based on when models were released and features were implemented.
|
||
|
||
## Dating Methodology
|
||
|
||
Papers reflect:
|
||
1. **Model Release Date**: When the model was first published on Hugging Face
|
||
2. **Architecture Version**: Qwen3 architecture baseline
|
||
3. **Major Updates**: Significant capability enhancements
|
||
|
||
All papers use **versioning**: `vYYYY.MM` (e.g., v2024.12 for December 2024)
|
||
|
||
---
|
||
|
||
## 🧠 Zen Model Family Overview
|
||
|
||
### Core Philosophy
|
||
Zen models are based on **Qwen3 architecture** (NOT Qwen2), emphasizing:
|
||
- Native 7680-dimensional embeddings (vs 3584 in Qwen2)
|
||
- Semantic alignment for DSO (Decentralized Semantic Optimization)
|
||
- BitDelta compression (31.87× ratio)
|
||
- Byzantine-robust consensus
|
||
|
||
---
|
||
|
||
## 📄 Published Papers (GitHub Actions)
|
||
|
||
### Foundation Models
|
||
|
||
#### 1. Zen Family Overview
|
||
- **Version**: v2024.12 (December 2024)
|
||
- *Model family established*: Late 2024
|
||
- Comprehensive architecture overview
|
||
- All model variants and use cases
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen_family_overview.tex`
|
||
|
||
#### 2. Zen Technical Paper
|
||
- **Version**: v2024.12 (December 2024)
|
||
- Deep technical architecture
|
||
- Qwen3 modifications
|
||
- Training methodology
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-technical-paper.tex`
|
||
|
||
---
|
||
|
||
### Core Models (Production)
|
||
|
||
#### 3. Zen-Coder
|
||
- **Version**: v2024.11 (November 2024)
|
||
- *Model release*: Late 2024
|
||
- Code generation and understanding
|
||
- Multi-language programming support
|
||
- 32B parameters
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-coder_whitepaper.tex`
|
||
|
||
#### 4. Zen-Omni (Multimodal)
|
||
- **Version**: v2025.01 (January 2025)
|
||
- *Based on Qwen3-Omni-30B-A3B*: Early 2025
|
||
- Vision + Audio + Text
|
||
- A3B architecture integration
|
||
- 30B parameters
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-omni_whitepaper.tex`
|
||
|
||
#### 5. Zen-Nano (Edge)
|
||
- **Version**: v2024.10 (October 2024)
|
||
- *0.6B params for edge deployment*
|
||
- On-device inference
|
||
- ONNX/TensorRT optimization
|
||
- Quantized to 4-bit
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-nano_whitepaper.tex`
|
||
|
||
#### 6. Zen-Eco (Efficient)
|
||
- **Version**: v2024.11 (November 2024)
|
||
- *4B params balanced model*
|
||
- Cost-effective inference
|
||
- Good performance/cost ratio
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-eco_whitepaper.tex`
|
||
|
||
#### 7. Zen-Next (Latest)
|
||
- **Version**: v2025.01 (January 2025)
|
||
- *Cutting-edge capabilities*
|
||
- Latest Qwen3 enhancements
|
||
- Experimental features
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-next_whitepaper.tex`
|
||
|
||
#### 8. Zen-Reranker (DSO)
|
||
- **Version**: v2025.10 (October 2025)
|
||
- **JUST PUBLISHED!**
|
||
- Native 7680-dim embeddings
|
||
- BitDelta compression (31.87×)
|
||
- 94.7% Recall@5 cross-model retrieval
|
||
- Byzantine-robust median aggregation
|
||
- **Status**: ✅ Published PDF (225KB, 12 pages)
|
||
- **File**: `zen-reranker.tex`
|
||
- **PDF**: `zen-reranker.pdf`
|
||
|
||
---
|
||
|
||
### Specialized Models
|
||
|
||
#### 9. Zen-Artist (Image Generation)
|
||
- **Version**: v2024.12 (December 2024)
|
||
- Text-to-image generation
|
||
- Style transfer capabilities
|
||
- Integration with DALL-E 3 / Stable Diffusion techniques
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-artist_whitepaper.tex`
|
||
|
||
#### 10. Zen-Artist-Edit (Image Editing)
|
||
- **Version**: v2024.12 (December 2024)
|
||
- Image editing and manipulation
|
||
- Inpainting and outpainting
|
||
- Style preservation
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-artist-edit_whitepaper.tex`
|
||
|
||
#### 11. Zen-Designer-Instruct (UI/UX)
|
||
- **Version**: v2024.12 (December 2024)
|
||
- Design system generation
|
||
- UI/UX recommendations
|
||
- Accessibility compliance
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-designer-instruct_whitepaper.tex`
|
||
|
||
#### 12. Zen-Designer-Thinking (Design Reasoning)
|
||
- **Version**: v2025.01 (January 2025)
|
||
- Design reasoning chains
|
||
- Iterative improvement
|
||
- Critique and suggestions
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-designer-thinking_whitepaper.tex`
|
||
|
||
#### 13. Zen-Scribe (Writing)
|
||
- **Version**: v2024.11 (November 2024)
|
||
- Long-form content generation
|
||
- Multiple genres and styles
|
||
- SEO optimization
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-scribe_whitepaper.tex`
|
||
|
||
#### 14. Zen-Guard (Safety)
|
||
- **Version**: v2024.12 (December 2024)
|
||
- Content moderation
|
||
- Safety classification
|
||
- Toxicity detection
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-guard_whitepaper.tex`
|
||
|
||
---
|
||
|
||
### Extended Family (Advanced)
|
||
|
||
#### 15. Zen-3D
|
||
- **Version**: v2025.01 (January 2025)
|
||
- 3D model generation
|
||
- Mesh optimization
|
||
- Texture synthesis
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-3d.tex`
|
||
|
||
#### 16. Zen-Foley (Audio Effects)
|
||
- **Version**: v2025.01 (January 2025)
|
||
- Sound effects generation
|
||
- Audio matching for video
|
||
- Spatial audio
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-foley.tex`
|
||
|
||
#### 17. Zen-Musician
|
||
- **Version**: v2025.01 (January 2025)
|
||
- Music composition
|
||
- Multiple genres
|
||
- MIDI generation
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-musician.tex`
|
||
|
||
#### 18. Zen-Director (Video)
|
||
- **Version**: v2025.01 (January 2025)
|
||
- Video scene generation
|
||
- Storyboarding
|
||
- Shot composition
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-director.tex`
|
||
|
||
#### 19. Zen-Agent (Autonomous)
|
||
- **Version**: v2025.01 (January 2025)
|
||
- Autonomous task execution
|
||
- Tool use and planning
|
||
- Multi-step reasoning
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-agent.tex`
|
||
|
||
#### 20. Zen-World (Simulation)
|
||
- **Version**: v2025.02 (February 2025)
|
||
- World model simulation
|
||
- Physics understanding
|
||
- Causal reasoning
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-world.tex`
|
||
|
||
#### 21. Zen-Video
|
||
- **Version**: v2025.01 (January 2025)
|
||
- Video generation
|
||
- Frame interpolation
|
||
- Video understanding
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-video.tex`
|
||
|
||
#### 22. Zen-Voyager (Exploration)
|
||
- **Version**: v2025.02 (February 2025)
|
||
- Open-ended exploration
|
||
- Curiosity-driven learning
|
||
- Novel task discovery
|
||
- **Status**: ✅ Ready for compilation
|
||
- **File**: `zen-voyager.tex`
|
||
|
||
---
|
||
|
||
## 📊 Model Family Statistics
|
||
|
||
| Category | Models | Status | Avg Size |
|
||
|----------|--------|--------|----------|
|
||
| **Foundation** | 2 | Published | N/A |
|
||
| **Core Production** | 6 | Published | 4B-32B |
|
||
| **Specialized** | 6 | Published | Varies |
|
||
| **Extended** | 8 | Published | Varies |
|
||
| **Total** | **22 models** | ✅ All documented | |
|
||
|
||
---
|
||
|
||
## 🎯 GitHub Actions Workflow
|
||
|
||
All papers are automatically compiled via GitHub Actions:
|
||
|
||
```yaml
|
||
Workflow: .github/workflows/compile-papers.yml
|
||
Trigger: Push to main (*.tex files)
|
||
Output: PDFs in /pdfs/ directory
|
||
Artifacts: 90-day retention
|
||
Releases: Automatic with all PDFs
|
||
```
|
||
|
||
**Status**:
|
||
- ✅ Zen-Reranker PDF compiled (225KB, 12 pages)
|
||
- 🔄 Remaining 21 papers compiling via CI
|
||
- ⏳ Expected completion: 8-12 minutes
|
||
|
||
---
|
||
|
||
## 📦 Hugging Face Model Hub
|
||
|
||
All Zen models published at: `https://huggingface.co/zenlm/`
|
||
|
||
Models include:
|
||
- Full precision checkpoints
|
||
- Quantized versions (4-bit, 8-bit)
|
||
- GGUF format for llama.cpp
|
||
- ONNX exports
|
||
- TensorRT engines (select models)
|
||
|
||
---
|
||
|
||
## 🔄 Version Tracking Format
|
||
|
||
Each paper includes:
|
||
|
||
```latex
|
||
\title{Zen-[Model] Whitepaper}
|
||
\author{Zen Language Models (ZenLM) \\ zen.ai}
|
||
\date{v2024.12 — December 2024}
|
||
|
||
\section*{Model Information}
|
||
\begin{itemize}
|
||
\item \textbf{Architecture}: Qwen3-based
|
||
\item \textbf{Parameters}: [size]
|
||
\item \textbf{Context Length}: 32,768 tokens
|
||
\item \textbf{Training Data}: Multi-domain corpus
|
||
\item \textbf{License}: Apache 2.0
|
||
\end{itemize}
|
||
```
|
||
|
||
---
|
||
|
||
## 🚀 Publication Timeline
|
||
|
||
- **October 2024**: Zen-Nano (0.6B edge model)
|
||
- **November 2024**: Zen-Coder, Zen-Eco, Zen-Scribe
|
||
- **December 2024**: Zen-Artist family, Zen-Guard, Core papers
|
||
- **January 2025**: Zen-Omni, Zen-Next, Extended family (3D, Foley, etc.)
|
||
- **February 2025**: Zen-World, Zen-Voyager
|
||
- **October 2025**: Zen-Reranker (DSO/BitDelta)
|
||
|
||
---
|
||
|
||
## 📝 Citation Format
|
||
|
||
```bibtex
|
||
@software{zen2024family,
|
||
title = {Zen Language Models: A Comprehensive Family},
|
||
author = {ZenLM Team},
|
||
year = {2024},
|
||
url = {https://github.com/zenlm/papers},
|
||
version = {v2024.12}
|
||
}
|
||
|
||
@software{zen2025reranker,
|
||
title = {Zen-Reranker: Native 7680-Dimensional Embeddings for DSO},
|
||
author = {ZenLM Team},
|
||
year = {2025},
|
||
month = {10},
|
||
url = {https://github.com/zenlm/papers},
|
||
version = {v2025.10}
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
**Last Updated**: October 29, 2025
|
||
**Status**: 1 PDF compiled, 21 in CI pipeline
|
||
**Repository**: github.com/zenlm/papers
|
||
**CI Build**: [](https://github.com/zenlm/papers/actions/workflows/compile-papers.yml)
|