mirror of
https://github.com/zenlm/zen-docs.git
synced 2026-10-30 14:58:13 +00:00
zeekay
8ef8f3e801
Updated 8 existing stubs and added 4 new model pages: - zen-nano: 0.6B edge model specs, throughput table, benchmarks - zen-scribe: 4B writing model, multilingual, use cases - zen-omni: 30B hypermodal Thinker-Talker MoE, speech-to-speech - zen-vl: 4B/8B/30B VL family with function calling and agent variants - zen-embedding: MTEB #1 multilingual (70.58), MRL support - zen-coder: 4B/31B/480B MoE family, FIM, 358 languages - zen-designer: 235B VL MoE, document parsing, GUI interaction - zen-max: 1T+ MoE, 256K context, 200+ tool calls, heavy mode - zen-pro: 8B instruct/thinking/agent variants - zen-reranker: 8B dual-task, native 7680-dim for DSO - zen-translator: real-time translation pipeline with voice cloning - zen-guard: removed upstream model reference No upstream model names in any page (Zen MoDE branding throughout).
98 lines
3.0 KiB
Plaintext
98 lines
3.0 KiB
Plaintext
---
|
|
title: Zen Reranker
|
|
description: 8B dual-task embedding and reranking model with native 7680-dim output for Decentralized Semantic Optimization
|
|
---
|
|
|
|
# Zen Reranker
|
|
|
|
Zen Reranker is an 8B dual-task embedding and reranking model built on the Zen Embedding architecture. It natively outputs 7680-dimensional embeddings — the canonical dimension for the Hanzo and Zoo Networks — eliminating alignment overhead in Decentralized Semantic Optimization (DSO) pipelines. It supports 100+ languages, 8K context, and BitDelta compression.
|
|
|
|
## Specifications
|
|
|
|
| Property | Value |
|
|
|----------|-------|
|
|
| Parameters | 8.2B (8B base + 200M projection head) |
|
|
| Embedding Dimension | 7680 (native, no alignment needed) |
|
|
| Context Length | 8,192 tokens |
|
|
| Languages | 100+ |
|
|
| License | Apache 2.0 |
|
|
| Tasks | Embedding + Reranking (dual-task) |
|
|
|
|
## Model Variants
|
|
|
|
| Variant | Context | Optimized For |
|
|
|---------|---------|---------------|
|
|
| zen-reranker-8b | 8K | General embedding and reranking |
|
|
| zen-reranker-8b-code | 16K | Code search (Hanzo Network) |
|
|
| zen-reranker-8b-research | 16K | Scientific papers (Zoo Network) |
|
|
| zen-reranker-1.5b | 4K | Edge devices (distilled) |
|
|
|
|
## Quick Start
|
|
|
|
```python
|
|
from zen_reranker import ZenReranker
|
|
|
|
model = ZenReranker.from_pretrained("zenlm/zen-reranker-8b")
|
|
|
|
# Generate 7680-dim embeddings
|
|
texts = [
|
|
"How to handle async errors in Rust?",
|
|
"Explain gradient descent",
|
|
]
|
|
|
|
embeddings = model.encode(texts)
|
|
print(embeddings.shape) # (2, 7680)
|
|
```
|
|
|
|
## Reranking
|
|
|
|
```python
|
|
query = "async error handling in Rust"
|
|
docs = [
|
|
"Use Result<T, E> with ? operator",
|
|
"Tokio provides timeout utilities",
|
|
"Python has try-except blocks",
|
|
"Result type propagates errors"
|
|
]
|
|
|
|
scores = model.rerank(query, docs)
|
|
ranked = sorted(zip(docs, scores), key=lambda x: x[1], reverse=True)
|
|
```
|
|
|
|
## Use Cases
|
|
|
|
- Decentralized Semantic Optimization (DSO) retrieval with no alignment overhead
|
|
- Cross-model experience retrieval (Hanzo + Zoo Networks)
|
|
- Code search and documentation (coding variant)
|
|
- Research paper retrieval (research variant)
|
|
- Semantic similarity and clustering
|
|
|
|
## Architecture
|
|
|
|
Zen Reranker extends Zen Embedding 8B with:
|
|
|
|
1. **Expanded projection head** — 4096 → 7680 dimensions via learned linear projection
|
|
2. **Dual-task training** — Contrastive embedding loss + cross-encoder reranking loss
|
|
3. **BitDelta optimization** — 31.87x compression (964 bytes per embedding) with 97.3% quality retention
|
|
|
|
## Performance
|
|
|
|
### MTEB Benchmark
|
|
|
|
| Model | Avg | Retrieval | Reranking | Clustering | STS |
|
|
|-------|-----|-----------|-----------|------------|-----|
|
|
| Zen-Embedding-8B | 65.1 | 63.2 | 60.1 | 53.8 | 83.4 |
|
|
| **Zen-Reranker-8B** | **67.3** | **65.8** | **62.4** | **55.2** | **85.1** |
|
|
|
|
### DSO Cross-Model Retrieval (Precision@5)
|
|
|
|
| Model | Precision@5 |
|
|
|-------|-------------|
|
|
| Zen-Embedding-8B | 88% |
|
|
| **Zen-Reranker-8B** | **98%** (native 7680-dim) |
|
|
|
|
## Get Started
|
|
|
|
- [HuggingFace](https://huggingface.co/zenlm/zen-reranker-8b)
|
|
- [GitHub](https://github.com/zenlm/zen-reranker)
|