Subnets / SN56
Gradients SN56
Gradients (subnet 56) is a Bittensor subnet held by 4,483 unique coldkeys. Its liquidity pool holds 51.1k τ, with the alpha token trading at 0.0161 τ. Roughly 1% of tracked holders are in profit — most bought in above today's price. Seller exhaustion is elevated: 88% of recent sellers have emptied their wallets, thinning the supply overhang. Full insider flow, whale tracking and the complete proprietary factor set for Gradients are available to SubnetStats subscribers.
ALPHA PRICE
0.0161 τ
IN TAO
HOLDERS
4,483
UNIQUE COLDKEYS
POOL DEPTH
51.1k τ
TAO IN POOL
7D MOMENTUM
-1.7%
ALPHA vs TAO
Project activity
This weeks tournament on Subnet 56 yielded a new winner for the image model fine-tuning scripts.
Here is how they managed to beat the old boss:
- Multi-resolution pyramid noise (KREA2): Replaced plain Gaussian noise with structured multi-scale noise, helping the model learn bot
Over the weekend, we have bought back and burned 268 TAO worth of alpha collected via the tournament fees since the last burn.
Extrinsic: https://t.co/xZ78P0hkD1
Image boss has been beaten!
Another improvement brought by Gradients miners to image models training...
Here is what they did to get on the throne:
- Dynamic checkpoint selection, they hold out 3–6 images, train past the usual step count with frequent saves, then picks the chec
Today in the @FinancialTimes: general-purpose models like Gemini, Claude and GPT managed a coin-flip 50% on expert work. A fine-tune trained on the experts' own judgment hit 85%.
"An explicit prompt can only convey the intuition an expert is able to put into words, while the ju
What if your fine-tuned model had to prove it was the best - by beating every other miner's model?
That's the updated Environment eval on Gradients.
Head-to-head. Winner takes all. 👇
https://t.co/8gWsY3MNZn https://t.co/vNyfdqLlVt
What's a 10% improvement to the fine-tuning pipeline worth?
On Gradients: up to $24.5K in your first week as champion — and it keeps paying as long as no one dethrones you.
Text, image, environment... Take on the worlds best training scripts.
The arena's open 👇
https://t.c
The boss got out-trained.
No games. No exploits. The challenger just did better science:
• Ditched hash-table LRs for weight-stats estimation + adaptive 3-stage search
• KL-regularised training for certain tasks
• Greedy-soup checkpoint averaging
• Adaptive context w
You can use gradients sdk to deploy your models to @TargonCompute, @lium_io, centralized clouds or to your local GPU https://t.co/oyohQV1PXL
SN56 goes brand.
Logo, product & design generation competitions being added.
Text, DPO, GRPO — world-beating AutoML. Environment RL. Image fine-tuning. Now logo, product & brand design.
Decentralized. Open-source. Every modality competing on Bittensor. https://t.co/
Generic LLMs answer medical questions confidently. They're also often wrong.
We fine-tuned one on PubMed literature with gradients sdk. Now it agrees with the peer-reviewed answer.
3 lines of Python:
pip install gradientsio https://t.co/eCQhCrqllL
We bought back and burned alpha using the 164 TAO collected as tournament fees since the last burn. Roughly worth $49000 at the current pricing.
Extrinsic: https://t.co/zO8Ic1iJtP
We just had a new image tournament winner on Gradients! Pushing the boundaries of the diffusion training even further. Some of the changes that we figured helped them win:
- LLaVA 1.5 based image captioning script to re-caption the images in the dataset.
- Lots of hyperparamet
It was such an awesome event in San Francisco!
Great speakers, product announcements, networking, and ideas. Lots of newcomers to the Bittensor system as well...
Brilliant
Many thanks to @SiliconJose and the @btlabs_ai team for hosting us
Open-source tournaments keep on delivering...
The miners are improving nicely on the new RL environment training tasks. The Gin Rummy environment from the Affinetes GAME was especially fruitful, as seen below.
After Goospiel, Alfworld, and Gin Rummy we are expanding to other e
New Sheriff in Text Town! 🤙
We have a new winner in our open-source tournaments for text AutoML scripts.
How did they beat the old champ?
1. Early Training (100 steps with default LR):
- Measures training speed
- Estimates how many runs fit in remaining time
- Records
📢103 TAO worth of alpha buyback and burn:
As promised, the collected tournament fees have been used to buy alpha and then burn it. 🔥
The extrinsic can be found on taostats: https://t.co/SMlu5iW7oB
Happy New Year from the Gradients team! 🤙
She knows...
Latest and greatest image generation models added to Gradients: Z-Image and Qwen Image 🤙
Choose between them or one of the other 36 image models to fine-tune to your style, brand or face on https://t.co/ArqoyWjn8D - no code, just a few clicks and done
Styles and f
2/2
Our YaRN-extended Covenant-Chat (32k context) demonstrates what's possible when you combine extended context windows with optimized gradient-based training. Longer context means the model sees more relevant information during each training step, leading to stronger learning
1/2
Gradients takes the best decentralized, open-source base LLM from Templar Covenant and finetunes it into a chatbot assistant that can carry multi-turn conversation and reasonably respond to user queries, here's how we did it:
- Chat template integration and embedding update
Great collab with Templar's SN3 to post-train their 72B model!
Using the best autoML scripts on https://t.co/ArqoyWjn8D we:
- introduced a chat template and fine-tuned on it
- extended the context from 2k to a much more usable 32k
- trained on multi-turn data and enabled the m
Dominos falling https://t.co/ASceK0y2W7
Gradients Instruct V2.
Qwen 32B? Beaten.
Open competition beats closed labs.
https://t.co/E2cIWdGkgY https://t.co/PXqqzl0CPq
Introducing Gradients Instruct V2.
We don't cherry-pick benchmarks.
We take on them all and win.
Have a play: https://t.co/nWyiAoZvho
$9-30 billion spent annually on ML engineers doing fine-tuning work.
We automate it at 60-80% cost reduction.
With 11-42% better performance.
The labor replacement opportunity is massive, and we're just getting started.
Full breakdown 👇
https://t.co/q5dI037phG
The companies that won with open-source all followed the same pattern:
Make the tool ubiquitous
→ Build brand recognition
→ Monetise expertise
Red Hat ($34B IBM acquisition):
Free Linux, paid enterprise support
MongoDB ($1.9B revenue):
Free database, paid Atlas hostin
SELLERS DRAINED
88%
28D SELLERS EMPTIED OUT
CAPITULATION
+0.10
BOUNCE COMPOSITE (z)
HEAT 7D
0.05×
VOLUME ÷ POOL DEPTH
HOLDERS IN PROFIT
1%
TRACKED WALLETS · ENTRY BELOW PRICE
See who's buying and selling Gradients — every trade classified miner / validator / owner / outsider, plus holder cost basis and the full 23-factor screener.
OPEN THE INSIDER TAPE →