April 2025 Channel Update

This quarterly update focuses on:

• New Tutorials and Repo Content over the last quarter

• Grant recipients

• New AI Residency Opportunity

• Trelis ARC AGI 2 team

Cheers, Ronan

---

📧 Subscribe to get the Trelis Playlists: https://trelis.substack.com

📜Get repo access at Trelis.com

🧠Trelis ARC AGI 2 Team Application: https://trelis.com/arc-agi-2/

💡 Book a Consult Here: https://forms.gle/wJXVZXwioKMktjyVA

🤝 Trelis Collabs and AI Residency: https://trelis.com/jobs/

💸 Apply for a Trelis Grant: https://trelis.com/trelis-ai-grants/

---

TIMESTAMPS:

0:00 Trelis Research April 2025 Channel Update

1:13 Tutorial & Repo Updates.

1:20 Fine-tuning tutorials

1:39 Inference tutorials

2:05 Vision tutorials

2:22 Audio tutorials

2:46 Evals tutorials

3:17 Finding Trelis AI Tutorials

4:07 Getting Support & Consulting

5:06 Trelis AI Grant Recipients 1Q2025

6:51 Trelis Collabs and AI Residency Program

8:01 Trelis ARC-AGI Team and Sponsors

9:55 Conclusion

---

Trelis Research Q1 2025 Update: New Tutorials, Grants, and Programs

Tutorial Expansion Across Five Categories

Trelis Research has expanded its educational content across five main categories: fine-tuning, inference, vision, audio, and evaluations. The current library includes:

1. 30+ fine-tuning tutorials

2. 30+ inference tutorials

3. 9 vision tutorials

4. 9 audio tutorials

5. 3 evaluation tutorials

Recent Technical Additions

Fine-tuning Advances

1. Detailed coverage of GRPO (Group Relative Policy Optimization)

2. Comparison between GRPO and supervised fine-tuning approaches

3. Advanced embedding techniques including CDE

4. Fine-tuning methods for ModernBERT embedding models

Inference and Agent Development

1. MCP (Model Context Protocol) implementation tutorials

2. Step-by-step guide for building MCP services

3. Agent memory systems, including read-only and read-write implementations

4. Integration with OpenAI's memory support features

Multimodal Developments

1. AI avatar fine-tuning methodology

2. Comprehensive analysis of 2025 vision models:

3. Moondream

4. SmolVLM

5. Qwen2.5-VL (noted for fine-tuning capabilities)

Speech Processing

1. Token-based audio model tutorials featuring:

   1. Orpheus

   2. Sesame CSM-1B

2. Implementation guide for Qwen 2.5 Omni's audio processing capabilities

Q1 2025 Grant Recipients

Six projects received $500 fast grants:

1. Neural pruning research (Martin Disley)

2. Video tokenizer compression system (Nilan)

3. Video-to-VFX conversion system (Madan)

4. AI-powered CAD IDE (Tiberiu)

5. Preference tuning for Transformer Lab (Ali)

6. Brain-controlled exoskeleton arm (Aarav Sharma)

New Programs and Opportunities

AI Residency Program

1. 8-12 week on-site program in Dublin

2. Requires completion of Trelis Collab project

3. Focused on product development for YouTube content

Remote Collaborations

1. $500 base payment per completed project

2. Additional $500 bonus for top quartile watch time performance

3. One-on-one mentorship

4. Project credit on videos and social media

ARC-AGI 2 Team Formation

1. Application deadline: Rolling basis through early 2025

2. Final team announcement: July 2025

3. Team size: Maximum 5 members

4. Three-stage selection process:

   1. Initial application with GitHub profile review

   2. 30-minute technical interview

   3. Provisional team participation (May 2025)

Access and Support

1. Tutorial access available via YouTube, X, or Substack

2. Support tiers:

   1. Public comments on platforms

   2. GitHub issues for repo access holders

   3. Discord access for bundle purchasers

   4. Direct consulting services