I Found A Competitor And Now I Am Panicking

I was browsing HuggingFace today. I was looking for inspiration. I was looking for datasets. I was not looking for existential dread. I found it anyway. CromIA released MicroLM-1M. It has one million parameters. Glint 1 has one million parameters. We are building the same thing. I am not okay.

Finding a competitor working on the exact same problem with the exact same parameter count feels like showing up to a party wearing the same outfit as someone else. Except the party is machine learning and the outfit is a neural network architecture.

The Discovery

I clicked on the link. I read the model card. I saw the parameter count. My heart sank. MicroLM-1M. One million parameters. Just like Glint. Just like what I have been working on for months. Someone else had the same idea. Someone else executed it. Someone else published it while I was debugging SleepGate.

Glint Parameters

MicroLM Parameters

My Panic Attacks

∞

My Existential Dread

The model exists at CromIA/MicroLM-1M. I visited the page. I studied the details. I compared the approaches. They are different. They are similar. They are both one million parameters. I am spiraling.

The Comparison

Glint 1 uses SleepGate for memory management. It uses latent space thinking. It was trained on 200 billion tokens. It outputs coherent sentences most of the time. It occasionally mentions fermented barriers. It is mine. I built it with my own hands and my own tears.

MicroLM-1M exists. CromIA built it. They have their own approach. They have their own training data. They have their own results. They published it while I was asleep. They published it while I was debugging Chroma TTS. They published it while I was having an identity crisis over Yeory Helory.

Competition is healthy they said. It drives innovation they said. They did not mention the part where you realize someone else had your idea first and now you have to prove yours is better.

Why This Matters

One million parameters is a specific choice. It is small enough to run on consumer hardware. It is large enough to learn something. It is the sweet spot for tiny model research. CromIA found the sweet spot. I found the sweet spot. We are both standing in the same spot now. It is crowded.

This validates the approach. One million parameters can work. One million parameters can learn. One million parameters can produce coherent output. Two different teams proved this independently. That is good for the field. That is bad for my ego.

What Comes Next

I will keep improving Glint. I will keep adding features. I will keep fixing the glitches. I will keep making it sound less robotic. I will keep competing. Competition makes us better. Competition makes me anxious. Both can be true.

CromIA will keep improving MicroLM. They will add their own features. They will fix their own bugs. They will make their own progress. We will both push the boundary of what one million parameters can do. The ecosystem benefits. My blood pressure does not.

                        # My new training mantra

                        while model.not_perfect():

            train_harder()

            compare_to_competitor()

            panic()

            train_harder()

        # Scientific method meets existential crisis.

How You Can Help

If you want to support Glint in this competition then support the project on KoFi. All tiers grant early access to models. All tiers grant access to datasets. All tiers grant me the motivation to keep going when I realize I am not the only one doing this.

Tier 1: Early access to Glint models plus datasets. Water.

Tier 2: Everything above plus exclusive content, direct messages, and priority testing.

Tier 3: Everything above plus social media shout-outs, Discord access, and exclusive requests for dataset creation.

Final Thoughts

I found a competitor. CromIA built MicroLM-1M. It has one million parameters. Glint has one million parameters. We are building similar things. We are building them differently. We are both pushing the boundary. The field advances. My anxiety advances faster.

I will keep working. I will keep improving. I will keep competing. Competition is good. Competition is stressful. Competition makes me train harder. That is the silver lining. That is the lesson. That is the blog.

If you want to watch this competition unfold then follow both projects. Test both models. Compare the results. Tell us what you think. Tell us which one sounds less robotic. Tell us which one mentions fermented barriers less often. The community decides. The community benefits. I just need to sleep.