ErdeM (@erdemwrites) header
ErdeM (@erdemwrites) profile picture

Building AI Inference & Coding tools Working on 🐧 @penginlabs.

New York, USA
Joined January 2010
Photo by ErdeM (@erdemwrites)Photo by ErdeM (@erdemwrites)Photo by ErdeM (@erdemwrites)Photo by ErdeM (@erdemwrites)Photo by ErdeM (@erdemwrites)Photo by ErdeM (@erdemwrites)Photo by ErdeM (@erdemwrites)Photo by ErdeM (@erdemwrites)Photo by ErdeM (@erdemwrites)Photo by ErdeM (@erdemwrites)
I really want to hear your comments and advices on this project: teapawt.net/erdemwrites/status/212…
When I listened to what Sam Altman said and don't want to kill potential of Ram sitting in my unused computers anymore and kiln is born. I really want to hear your comments and advices.. github.com/erdemozkan/kiln
1
79
This is a huge step for the local AI ecosystem. I’ve seen people complain about their computers’ fans getting noisy when they run these models. I guess they haven’t been near a data center—you can hear the rigs from half a mile away on a quiet day šŸ˜… teapawt.net/SamuelZengML/status/21…
Introducing Edge0-145B and Edge0-120B, @Edge0_AI’s first device-native 100B-class foundation models with open weights. Cloud-level intelligence on your laptop. Free, offline, and fast enough for real vibe coding. - Smart: beats Claude Opus 4.6 on Terminal-Bench 2.1 (86.5% vs. 78.2%) - Light: needs just 29G memory, 1/3 of a standard setup (100G+) - Fast: decodes 75+ toks/s, prefills 1050+ toks/s on RTX Spark and Mac. - Steady: handles 128K–256K tokens of context without slowing down, even on long tasks Frontier-class agentic coding. No cloud API. No token bill. Your code and data never leave your machine. Open weights are live on @huggingface. Run it locally or plug it into your favorite coding agent.
2
47
When I listened to what Sam Altman said and don't want to kill potential of Ram sitting in my unused computers anymore and kiln is born. I really want to hear your comments and advices.. github.com/erdemozkan/kiln
1
1
1
150
It's the first time Claude has said this to me. Is it an Opus 5.5 thing or just a random moment? BTW, something happened to Codex and my usage credits expire unusually quickly..
3
80
We’re collecting device compatibility data for a WebGPU alternative we’re developing. The test takes 20 seconds max; no signup, no product testing. claude.ai/code/artifact/fc7c… Especially interested in seeing results across different browsers, GPUs, and OS.. @qazwsx23310 @jun_song @andrewchen
3
59
This was a personal computer MVP from the late 1970s. A good reminder that first versions don’t need to be perfect. They just need to solve a real problem. Sometimes we’re overengineering software that should have been shipped months ago..
2
3
74
I’m trying to quantize and fine-tune small models to have ultra-compact ones with under 1B parameters that can run on browsers without need to use ollama or similar tools. Task-specific small models are the future..
1
3
58
I was 6 years old when I first learned typing on an AmstradšŸ˜€ Today, we’re talking about 1T param AI models. Our generation is unique;we experienced both the primitive days of PC era and the rise of advanced AI within a single lifetime. What computer did you learn to type on?
1
1
4
126
17M tokens/day locally is really promising.. teapawt.net/0xSero/status/20584644…
My daily average local model token burn is 17M They have become tremendously useful.
1
3
118
Does anyone remember when Gmail was invite only?
1
2
49
Business owners are actually very happy with local AI solutions. They don’t want to send their data to external servers and recurring AI API costs. They just want AI running on their own computers for a specific task.. One of them told me: ā€œAll I need is AutoCAD, not DaVinci.ā€
2
5
128
If anyone here has connections to a16z Speedrun, I’d really appreciate a referral..
8
193
This city is one of the best in the World!
1
6
80
In honor of 50 years of Apple, we're sharing - for the first time ever - Don Valentine's original 1977 memo for Sequoia's investment into Apple Computer. #Apple50
6
121
It's this easy to try using yolo-coder to start fixing your code errors. I need to hear your comments..
6
90
I’ve noticed that buying extra usage for Claude is useless. The extra credits seem to disappear much faster (like 2-3x) than the normal plan usage. Another reason I’m becoming a bigger fan of local AI models..
1
6
98