mmcquade_ai_u Profile Banner
Mark McQuade Profile
Mark McQuade

@mmcquade_ai_u

Followers
497
Following
576
Media
23
Statuses
182

CEO and cofounder of @arcee_ai | @huggingface šŸ¤— alum. AI and Data Obsessed. Fitness Fanatic. Tattoo Enthusiast.

Miami, FL
Joined February 2021
Don't wanna be here? Send us removal request.
@mmcquade_ai_u
Mark McQuade
25 days
Introducing AFM - Arcee Foundation Models. Today we are extremely excited to announce AFM - @arcee_ai new family of enterprise-grade foundation models designed for secure, real-time AI infrastructure and intelligent automation. Built for:.• On-prem, private cloud, or on-device
4
8
41
@mmcquade_ai_u
Mark McQuade
2 months
I have been a huge fan of what Anthropic is doing for a long time - they are right at the top of my respect list (alongside Together AI). Last week’s launch of integrations, web search, and deep research mode are šŸ”„šŸ”„šŸ”„. Truth is, I’ve been pushing for a deeper Arcee x Anthropic.
0
0
1
@mmcquade_ai_u
Mark McQuade
2 months
The next era of GenAI won't be confined to the cloud. It's built for the edge. At Arcee, we've quietly developed something new:. Arcee Foundational Models (AFMs) - our original LLM architecture, specifically engineered for local deployment. Why? Because AI's future isn't.
0
0
2
@mmcquade_ai_u
Mark McQuade
3 months
āš”ļøDropping soon: a killer combo for anyone building with Claude Desktop or custom AI clients. 1ļøāƒ£ Open Source Library: Seamlessly convert between stdio MCP servers and Server-Sent Events (SSE) — bidirectional, clean, and optimized. Perfect for unlocking Claude Desktop with your.
0
0
0
@mmcquade_ai_u
Mark McQuade
3 months
We didn’t have 100 engineers. But we built a production-grade training stack anyway. When we started building Arcee, we knew we couldn’t compete on headcount. But we could compete on efficiency, modularity, and speed. So we built a system where every component could stand on.
0
0
1
@mmcquade_ai_u
Mark McQuade
3 months
After 34,261 models and 43.2 million downloads….Here’s what we’ve learned in our first year of shipping MergeKit. It was built to make model merging fast, reproducible, and GPU-optional. No retraining. No fine-tuning loop. Just clean fusing of models. And the response? Way
Tweet media one
0
1
8
@mmcquade_ai_u
Mark McQuade
3 months
At Arcee, we started with full-stack model customization:.Continual pretraining, supervised fine-tuning, domain adaptation - you name it. But as base models got stronger, the incremental gains of heavy customization started to look less and less compelling - especially when
Tweet media one
0
0
1
@mmcquade_ai_u
Mark McQuade
3 months
šŸš€ AI Agents: Beyond Chatbots to True Automation šŸš€. The buzz about "AI agents" is everywhere, but let’s cut to the chase: most are still just chatbots in disguise, requiring human intervention for complex tasks. We need more! šŸ› ļø. In genuine operations, a single AI "brain" isn’t
Tweet media one
0
0
1
@mmcquade_ai_u
Mark McQuade
3 months
ā€œAI Agentsā€ is one of the most overused terms in the space right now. We saw teams call something an agent….but it still needed a human in the loop at every step. Start a task → get a fancy response → go back to doing the actual work yourself. So we built something simpler -
Tweet media one
1
0
1
@mmcquade_ai_u
Mark McQuade
3 months
If your AI bill is matching your cloud bill, you’ve got a significant issue. šŸ’ø. Companies are realizing that routing every query to the largest, most expensive LLMs is depleting their budgets. At Arcee, we recognized this early on and developed a solution. ⚔ Introducing.
1
0
0
@mmcquade_ai_u
Mark McQuade
3 months
A few months ago, we asked a simple question:.Why is everyone still sending every prompt to the same giant, expensive model?. We’d seen teams spending tens of thousands per month on LLMs. with most of those tokens going toward simple queries a small model could easily handle.
Tweet media one
0
1
4
@mmcquade_ai_u
Mark McQuade
3 months
šŸš€ The GenAI playbook is rapidly evolving, and not everyone will adapt in time. Here are 5 key shifts that I believe will separate the winners from the losers:. 1ļøāƒ£ One-model mindset -> Many-model orchestration: Leaders will utilize a variety of specialized models, not just a
Tweet media one
0
0
1
@mmcquade_ai_u
Mark McQuade
3 months
Controversial take:.Multi-agent orchestration isn’t real yet. We all want a swarm of autonomous agents collaborating toward a common goal — the ā€œAI teamā€ dream. But right now? It’s mostly a maze of looping messages, broken memory, and hallucinated handoffs. There’s massive.
0
0
1
@mmcquade_ai_u
Mark McQuade
3 months
MCP isn’t just protocol — it’s a preview of what’s coming. Anthropic’s Model Context Protocol (MCP) lays the groundwork for a future where models can tell us who they are — and where routing becomes intelligent by default. For too long, GenAI systems have treated models like.
0
0
3
@mmcquade_ai_u
Mark McQuade
3 months
Bold Decision that Paid Off: A Year of MergeKit Success. Last year, I made a strategic move: we acquired MergeKit, an innovative open-source project, and welcomed its creator Charles Goddard into our team. At the time, merging AI models was a groundbreaking concept that many
Tweet media one
0
0
1
@mmcquade_ai_u
Mark McQuade
3 months
RT @abhi1thakur: If you are drowning in costs associated with Large Language Models (LLMs), use this simple trick to save up to 65% of the….
0
5
0
@mmcquade_ai_u
Mark McQuade
3 months
🚨 MergeKit now supports Llama 4. The best open-source model just became merge-ready. You can now fuse multiple variations of Llama 4 into a single model using MergeKit. Whether you’re optimizing for performance, alignment, or downstream deployment — you now have more control.
0
2
6
@mmcquade_ai_u
Mark McQuade
3 months
Llama 4 is here — and it confirms what we’ve believed all along:. šŸ‘‰ The future of GenAI isn’t about bigger models. It’s about smarter systems — built from coordinated, specialized components. Meta didn’t just drop one model. They dropped a mixture-of-experts (MoE) system. Where
Tweet media one
0
0
5
@mmcquade_ai_u
Mark McQuade
4 months
RT @LucasAtkins7: There's a very real chance that on Monday @arcee_ai is going to be giving away FAR too many API credits. It'll be a hell….
0
8
0
@mmcquade_ai_u
Mark McQuade
6 months
Dropping our first distillations of @deepseek_ai V3 šŸ”„.
@LucasAtkins7
Lucas Atkins
6 months
Since @deepseek_ai V3's December launch, @arcee_ai has captured over 5 billion tokens of raw logits. With all the buzz around Deepseek, it's the perfect time to unveil our first large-scale logit-wise distillations: Virtuoso-Lite and Virtuoso-Medium.
Tweet media one
0
0
2