← Retour au fil
TPN (SN65) Is Shrinking AI Models for Browsers and Phones
TAO Daily14 nov., 09h · il y a 2j

TPN (SN65) Is Shrinking AI Models for Browsers and Phones

TPN uses miners to compress AI models for phones and browsers while preserving their capabilities. Its ‘Gratis AI’ lets users try the models locally in a browser. […]

Bittensor

Détails

Source
TAO Daily
Publication
14 nov. à 09h38

Contenu source (brut)

<p class="wp-block-paragraph">An AI model can fit on a phone and still lose too much of its ability to be useful. <a href="https://taodaily.io/tpn-sn65s-bet-on-making-ai-models-fit-the-hardware-users-have/">TPN(SN65)</a> is using Bittensor to find out how much of that capability miners can preserve as they shrink open models such as Mistral, Llama and Gemma.</p> <p class="wp-block-paragraph">The aim is to make those models practical to run on devices with limited memory.</p> <figure class="wp-block-embed is-type-video is-provider-youtube wp-block-embed-youtube wp-embed-aspect-16-9 wp-has-aspect-ratio"><div class="wp-block-embed__wrapper"> <iframe title="Hash Rate - Ep. 191: TPN Subnet 65" width="500" height="281" src="https://www.youtube.com/embed/DMwFHDIHHrY?feature=oembed" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen></iframe> </div><figcaption class="wp-element-caption">TPN Founder talks with Mark Jeffrey</figcaption></figure> <p class="wp-block-paragraph">On <a href="https://x.com/markjeffrey/status/2120687358022676776">Hash Rate Pod</a>, TPN’s co-founder, <a href="https://taodaily.io/meet-mentor-the-tpn-co-founder-who-turned-a-vpn-into-a-llm-compression-subnet/">Mentor Palokaj</a>, called the competition “model squashing.” Miners can use quantization to store model weights at lower precision or distillation to train a smaller model to imitate a larger one.</p> <p class="wp-block-paragraph">Half the score comes from <a href="https://taodaily.io/tpn-v0-2-changes-how-optimized-ai-models-are-scored-on-bittensor/">Fluid Bench</a>, which refreshes its test questions to limit overfitting. The other half measures how closely the compressed model matches the original’s predictions across all possible next tokens.</p> <figure class="wp-block-image size-large is-resized"><img loading="lazy" decoding="async" width="1024" height="330" src="https://taodaily.io/wp-content/uploads/2026/11/image-116-1024x330.png" alt="" class="wp-image-26364" style="aspect-ratio:3.1044776119402986;width:624px;height:auto" srcset="https://taodaily.io/wp-content/uploads/2026/11/image-116-1024x330.png 1024w, https://taodaily.io/wp-content/uploads/2026/11/image-116-300x97.png 300w, https://taodaily.io/wp-content/uploads/2026/11/image-116-766x247.png 766w, https://taodaily.io/wp-content/uploads/2026/11/image-116.png 1883w" sizes="(max-width: 1024px) 100vw, 1024px" /><figcaption class="wp-element-caption"><a href="https://bench.trueperformancenetwork.com/">How TPN Quantize Models</a></figcaption></figure> <p class="wp-block-paragraph"><a href="https://taodaily.io/tpn-sn65-launches-gratis-to-run-models-privately-on-your-own-device/">Gratis AI</a>, a product of TPN, gives users a way to try the results. The demo checks the visitor’s hardware and loads a model that runs locally in the browser through WebGPU, with no app to install.</p> <p class="wp-block-paragraph">Users can choose a faster model for simple tasks or wait a little longer for a more capable one. Mentor compared the current demos’ capabilities to those of early ChatGPT.</p> <figure class="wp-block-image size-large is-resized"><img loading="lazy" decoding="async" width="1024" height="434" src="https://taodaily.io/wp-content/uploads/2026/11/image-117-1024x434.png" alt="" class="wp-image-26365" style="aspect-ratio:2.3636363636363638;width:624px;height:auto" srcset="https://taodaily.io/wp-content/uploads/2026/11/image-117-1024x434.png 1024w, https://taodaily.io/wp-content/uploads/2026/11/image-117-300x127.png 300w, https://taodaily.io/wp-content/uploads/2026/11/image-117-766x325.png 766w, https://taodaily.io/wp-content/uploads/2026/11/image-117-1030x438.png 1030w, https://taodaily.io/wp-content/uploads/2026/11/image-117.png 1891w" sizes="(max-width: 1024px) 100vw, 1024px" /><figcaption class="wp-element-caption"><a href="https://gratis-ai.com/">Gratis’ Website</a></figcaption></figure> <p class="wp-block-paragraph">At the time of the interview, TPN was running a competition to see how far miners could shrink a larger Gemma model before its performance dropped sharply.</p> <p class="wp-block-paragraph">The team also plans to publish models compressed using its own workbench and put them into more public demos. Mentor said its versions outperformed the original developers’ quantized releases in TPN’s side-by-side tests.</p> <figure class="wp-block-embed is-type-rich is-provider-x wp-block-embed-x"><div class="wp-block-embed__wrapper"> <blockquote class="twitter-tweet" data-width="500" data-dnt="true"><p lang="en" dir="ltr">TPN&#39;s 8th model is under production!<br><br>We shipped 7 optimized models on <a href="https://x.com/huggingface?ref_src=twsrc%5Etfw">@huggingface</a>, coming from different labs like <a href="https://x.com/googlegemma?ref_src=twsrc%5Etfw">@googlegemma</a>, <a href="https://x.com/Alibaba_Qwen?ref_src=twsrc%5Etfw">@Alibaba_Qwen</a> and <a href="https://x.com/MistralAI?ref_src=twsrc%5Etfw">@MistralAI</a>. <br><br>Competition n8 is the largest one that TPN has ever hosted, with the scope to take <a href="https://x.com/Google?ref_src=twsrc%5Etfw">@Google</a>&#39;s Gemma-4-31B under a… <a href="https://t.co/Aoqj14SjPI">pic.twitter.com/Aoqj14SjPI</a></p>&mdash; True Performance Network (@TPN_Labs) <a href="https://x.com/TPN_Labs/status/2119055452998598823?ref_src=twsrc%5Etfw">November 7, 2026</a></blockquote><script async src="https://platform.x.com/widgets.js" charset="utf-8"></script> </div></figure> <p class="wp-block-paragraph">TPN began as a decentralised VPN before the team pivoted to model compression. Mentor said their global miner base lets the team tap computing capacity and test more approaches than it could easily manage in-house.</p> <p class="wp-block-paragraph">The immediate priority is refining the scoring system so that models that win the competition also hold up in everyday use.</p> <p class="wp-block-paragraph">Longer term, TPN wants to sell compression as a service to customers who already have a model and need it to run on specific hardware.</p> <p class="wp-block-paragraph">Within Bittensor, the same service could connect subnets that train models with providers that serve them. Compressing models between those two stages could reduce the memory needed to run them and lower the cost of inference.</p> <figure class="wp-block-embed is-type-wp-embed is-provider-the-tao-daily wp-block-embed-the-tao-daily"><div class="wp-block-embed__wrapper"> <blockquote class="wp-embedded-content" data-secret="gmbOZlS7Jv"><a href="https://taodaily.io/tpn-sn65-launches-gratis-to-run-models-privately-on-your-own-device/">TPN (SN65) Launches Gratis to Run Models Privately on Your Own Device</a></blockquote><iframe class="wp-embedded-content" sandbox="allow-scripts" security="restricted" style="position: absolute; visibility: hidden;" title="“TPN (SN65) Launches Gratis to Run Models Privately on Your Own Device” — The TAO Daily" src="https://taodaily.io/tpn-sn65-launches-gratis-to-run-models-privately-on-your-own-device/embed/#?secret=09EsifLjXR#?secret=gmbOZlS7Jv" data-secret="gmbOZlS7Jv" width="500" height="282" frameborder="0" marginwidth="0" marginheight="0" scrolling="no"></iframe> </div></figure>