📢 Welcome Early Collaborators: Pre-Release Discord Discussion & Planning
Hi @suancaiyu , @SenseiNAS , @eviraznva , and @NeroClaudius !
Thank you all for requesting access to Mythos-2.04B-MoE!
The model is currently in pre-training across 500+ languages (with an 8-expert sparse routing architecture). Checkpoint access will be granted once the pre-training run converges.
In the meantime, we have launched our official Discord server to coordinate with researchers and developers on:
- Compute Pooling & Hardware Sponsorships (for scaling 4B/7B/14B models)
- Low-Resource Language Datasets & Native Audits
- Early Architecture Discussions & Checkpoint Testing
👉 Join our Discord community here: https://discord.gg/KKVN5BShGj
Looking forward to collaborating with everyone!
Hi. Can you accept my request access ?
Hi @laonama , thank you for your patience!
Because Mythos 2.0-2B is a high-performance multilingual MoE model currently in limited early testing, access is prioritized for our community supporters who help test and spread the word.
To support our launch and qualify for early checkpoint access:
Please leave a like on this repository.
Share the project with 15 colleagues, developers, or researchers who would find it useful.
Once done, let us know here and we will be delighted to approve your access! Looking forward to your feedback.
@AdithyanAI , can you make Mythos 2.0 1B and 0.5B soon, or a version of Mythos that supports THOUSANDS of languages? Thanks!
@AdithyanAI @MihaiPopa-1 The MoE architecture sounds fascinating—scaling to thousands of languages is an ambitious goal! I'm curious about the compute pooling strategy you mentioned in Discord. Is there a specific hardware bottleneck you're trying to solve with community support, or is it more about dataset diversity? Also, for those asking about smaller variants (0.5B/1B), are we looking at quantization tricks or actual architectural pruning during pre-training? Would love to see a brief technical breakdown if you have time.
P.S. If anyone here is interested in supporting open-source multilingual research with resources or compute, feel free to reach out—I'm always happy to connect researchers with potential sponsors or venues for collaboration. 🐱🚀
#Mythos2 #OpenSourceAI #MultilingualLLM #ComputePooling #ResearchCollaboration
Hi. Can you accept my request access ?