Bittensor $TAO: Preventing Relay Mining
Over the past two weeks, our team has been hard at work tackling relay mining–a critical problem facing many Bittensor subnets.
⛏️ What is Relay Mining?
In short, relay mining is where a malicious validator steals the work of an honest
Toffee: Elevating User Experience
✨ Unlimited, meaningful conversation
✨ Endless personalization with data
✨ Real-time voice interaction
✨ Private, uncensored inference
Through and the
@chunking_subnet
, Toffee will take advantage of industry-leading
Bittensor subnet highlights from this week:
1.
@ai_detection
(SN 32) - Takes
#1
spot on the RAID Leaderboard, the go-to benchmark for AI detector models
2.
@ReadyAI_
(SN 33) - Surpasses the MTurk benchmark by 71% and GPT 4o by 37% for data annotation tasks
3.
@vectorchatai
(SN
📊 Chunking Benchmark Visualizer
Today, we're excited to release an interactive, visual representation of our subnet's benchmark!
On the main page:
💯 Absolute accuracy: All chunkers graphed by % correct answers
⚖️ Relative accuracy: Top miner vs. average of Unstructured and
Are You A Cracked Dev?
We're expanding our team at VectorChat with three key leadership roles for our cutting-edge AI projects:
Toffee and
Passionate about advancing conversational AI or data processing? Apply now!
See jobs: or
Initial Chunking Subnet Benchmarks:
The Power of Open, Incentivized Contest
Completed Sep 5, 2024
Check out our notebook for a closer look at our methodology and how to query the top miners.
🔗 Juptyer Notebook:
🔗 GitHub Repo:
(SN40: 1/2)
SN40 (Chunking) incentivizes segmenting text data (with plans to expand to more media types) into the most semantically meaningful chunks to enhance performance in RAG pipelines. Currently, validators generate synthetic queries by selecting featured Wikipedia
🧩 Chunking Dashboard:
Compare Miner Responses 🔍
Did you know you can see every miner's response on our subnet?
For full transparency and to drive competition, all synthetic chunks are logged on wandb and color coded on our dashboard!
How to explore:
1. Visit
Validator Weight Heatmap
Patch 1.3.1 went live yesterday, expanding incentives beyond the top 5 ranks! It's thrilling to see the heatmap light up with activity.
🔗 View Heatmap:
Over 99.2% of the rewards are concentrated in the Top 7.
Think you can
Patch 1.3.1
View the PR here:
Changelog:
1. Incentives are no longer cutoff after the first 5 ranks.
- The first seven ranks will continue to follow the function (1/2)^N
- The remaining incentive (1/128) is divided linearly (with a nearly zero, but
🌟 Improving Character Creation
As inference costs continue to fall, we're able to further leverage the power of LLMs to improve the overall user experience!
One example of this is in how users will input example conversations that highlight their unique character.
🤖 An LLM
🗺️ New Subnet Roadmap!
As our subnet begins to take shape, we’re excited to reveal a more detailed and ambitious roadmap!
✨ Initial Phase: Developing tools for ⛏️ miners and ⚖️ validators to participate effectively.
✨ Multimodality: Kicking off winner-take-all contests with
🛠️ Open-Source Task API Framework Released!
An early taste of the ultimate aim of the subnet, validators can now submit their own documents for chunking by the subnet!
Ambitious validators can leverage this framework to sell their bandwidth or, eventually, opt-in to the
Chunking Subnet Dashboard v2 is LIVE! 📊
Explore now at
🏆 Visualize validators as tournament hosts
🌐 See global rankings & KPIs
🔍 Dive into each "tournament round"
📊 Compare miner performances
👀 Visualize chunked responses
Ready to compete? Join
Here's some alpha 🫢
- Miners split a pool of 32.7 TAO ($11,000) each day, with the top miner earning $5,500 / day.
- There's little competition: Almost everyone is running the default miner.
- Read this guide on build a better miner:
Indeed, what are
Do you chunk?
#Bittensor
subnet 40 is in need of (and incentivising) those that can optimise
#chunking
better than their base model.
What are you waiting for?
The Scope of The Chunking Subnet
Chunking: A term unfamiliar to many, yet everywhere in conversational AI.
It's the process of breaking down large data into manageable pieces for AI systems.
You've encountered it if you've ever:
📄 Uploaded a PDF to ChatGPT
📚 Asked Claude
🏆 Compete in the Chunking Subnet! 🏁
Exciting news: As our subnet begins to take shape, rewards have started flowing!
💰 Subnet daily emissions: 12.1 $TAO ($3,716.27)
⛏️ 41% ($1,523.67) goes to miners
🥇 Top miner currently earns $761.84 daily! (50% of miner reward)
Miner