From a2aac5a63a36d7c05123e4aa5203d6026c2d4e68 Mon Sep 17 00:00:00 2001 From: Ramon Perez Date: Wed, 30 Jul 2025 12:00:31 +1000 Subject: [PATCH] vision update --- website/src/content/products/index.mdx | 21 +++++++++++++++- .../content/products/platforms/desktop.mdx | 25 ++++++++++++------- 2 files changed, 36 insertions(+), 10 deletions(-) diff --git a/website/src/content/products/index.mdx b/website/src/content/products/index.mdx index 5c34c1165..3dab3e086 100644 --- a/website/src/content/products/index.mdx +++ b/website/src/content/products/index.mdx @@ -71,7 +71,7 @@ We train our models in public. Check the [models page](./models/jan-v1) to see: - Failed runs and what went wrong - Models in testing before release -No black box. No "trust us, it's good." Watch the entire process from dataset to deployment. +No "trust us, it's good." Watch the entire process from dataset to deployment. ### Help Evaluate Our Models Every model needs real-world testing. Join our open evaluation platform where you can: @@ -84,6 +84,25 @@ Think LMArena, but you can see all the data, run your own evals, and directly in [Test/evaluate our models here](link) +### Our Models Training Right Now + +We don't just talk about open development. Here's what's actually happening: + +| Model | Progress | Status | Details | +|:------|:---------|:-------|:--------| +| **Jan-Search-7B** | ████████░░ 82% | Testing | [View run](/) • 2.1M steps • ETA 3 days | +| **Jan-Write-13B** | ████░░░░░░ 41% | Training | [View run](/) • 980K steps • On track | +| **Jan-Analyze-13B** | ████████░░ ~~67%~~ | Failed | [View logs](/) • OOM at step 1.5M • Restarting | + +These are our actual models training on our hardware in our Singapore office. Click any run to see: +- Live loss curves +- Training datasets +- Evaluation metrics +- Even our failures + +[Watch live training →](/train) + + ## Get Involved We build in public. Everything from our model training to our product roadmap is open. diff --git a/website/src/content/products/platforms/desktop.mdx b/website/src/content/products/platforms/desktop.mdx index b77cf9c72..31e7b059f 100644 --- a/website/src/content/products/platforms/desktop.mdx +++ b/website/src/content/products/platforms/desktop.mdx @@ -6,12 +6,14 @@ sidebar: --- import { Aside, Card, CardGrid, Tabs, TabItem } from '@astrojs/starlight/components'; -Jan Desktop is your local AI workstation. Download it, run your own models, or connect to cloud providers. Your computer, your choice. +Jan Desktop is your local AI workstation. Download it, run your own models, or connect to +cloud providers. Your computer, your choice. ## How It Works ### Default: Local Mode -Open Jan. Start chatting with Jan Nano. No internet, no account, no API keys. Your conversations never leave your machine. +Open Jan. Start chatting with Jan Nano. No internet, no account, no API keys. Your conversations +never leave your machine. ### Optional: Cloud Mode Need more power? Connect to: @@ -20,23 +22,28 @@ Need more power? Connect to: - Any OpenAI-compatible API ## Why Desktop First -Your desktop has the GPU, storage, and memory to run real AI models. Not toy versions. Not demos. The same models that power ChatGPT-scale applications. +Your desktop has the GPU, storage, and memory to run real AI models. Not toy versions. Not +demos. The same models that power ChatGPT-scale applications. -More importantly: it becomes the hub for your other devices. Your phone connects to your desktop. Your team connects to your desktop. Everything stays in your control. +More importantly: it becomes the hub for your other devices. Your phone connects to your +desktop. Your team connects to your desktop. Everything stays in your control. ## Specifications - - Everything in `~/jan`. Your data, your models, your configuration. Back it up, move it, delete it - it's just files. + + Everything in `~/.local/share/jan`. Your data, your models, your configuration. Back it up, move it, delete + it - it's just files. - - OpenAI-compatible API at `localhost:1337`. Any tool that works with OpenAI works with Jan. No code changes. + + OpenAI-compatible API at `localhost:1337`. Any tool that works with OpenAI works with Jan. No + code changes. NVIDIA CUDA acceleration out of the box. Automatically detects and uses available GPUs. CPU fallback always works.