411–420 of 909 posts
Inference in Python uses harmony [1] (for request and response format) which is written in Rust with Python bindings. Another OpenAI's Rust library is tiktoken [2], used for all tokenization and detokenization. OpenAI Codex [3] is also written in Rust. It looks like OpenAI is increasingly adopting Rust (at least for inference). [1] https://github.com/openai/harmony [2] https://github.com/openai/tiktoken [3] https://g…
Open models are going to win long-term. Anthropics' own research has to use OSS models [0]. China is demonstrating how quickly companies can iterate on open models, allowing smaller teams access and augmentation to the abilities of a model without paying the training cost. My personal prediction is that the US foundational model makers will OSS something close to N-1 for the next 1-3 iterations. The CAPEX for the fou…
The lede is being missed imo. gpt-oss:20b is a top ten model (on MMLU (right behind Gemini-2.5-Pro) and I just ran it locally on my Macbook Air M3 from last year. I've been experimenting with a lot of local models, both on my laptop and on my phone (Pixel 9 Pro), and I figured we'd be here in a year or two. But no, we're here today. A basically frontier model, running for the cost of electricity (free with a rounding…
I’m still trying to understand what is the biggest group of people that uses local AI (or will)? Students who don’t want to pay but somehow have the hardware? Devs who are price conscious and want free agentic coding? Local, in my experience, can’t even pull data from an image without hallucinating (Qwen 2.5 VI in that example). Hopefully local/small models keep getting better and devices get better at running bigger…
Privacy is obvious.
AI is going to to be equivalent to all computing in the future. Imagine if only IBM, Apple and Microsoft ever built computers, and all anyone else ever had in the 1990s were terminals to the mainframe, forever.
> Best with ≥60GB VRAM or unified memory
https://cookbook.openai.com/articles/gpt-oss/run-locally-oll...
The lede is being missed imo. gpt-oss:20b is a top ten model (on MMLU (right behind Gemini-2.5-Pro) and I just ran it locally on my Macbook Air M3 from last year. I've been experimenting with a lot of local models, both on my laptop and on my phone (Pixel 9 Pro), and I figured we'd be here in a year or two. But no, we're here today. A basically frontier model, running for the cost of electricity (free with a rounding…
can you please give an estimate how much slower/faster is it on your macbook compared to comparable models running in the cloud?
This is a thinking model, so I ran it against o4-mini, here are the results:
* gpt-oss:20b
* Time-to-first-token: 2.49 seconds
* Time-to-completion: 51.47 seconds
* Tokens-per-second: 2.19
* o4-mini on ChatGPT
* Time-to-first-token: 2.50 seconds
* Time-to-completion: 5.84 seconds
* Tokens-per-second: 19.34
Time to first token was similar, but the thinking piece was _much_ faster on o4-mini. Thinking took the majority of the 51 seconds for gpt-oss:20b.
Earlier quoted context omitted.
> research results have shown that highly curated technical problem solving data is unreasonably effective at boosting smaller models’ performance. same seems to be true for humans
Wish they gave us access to learn from those grandmother models instead of distilled slop.
The lede is being missed imo. gpt-oss:20b is a top ten model (on MMLU (right behind Gemini-2.5-Pro) and I just ran it locally on my Macbook Air M3 from last year. I've been experimenting with a lot of local models, both on my laptop and on my phone (Pixel 9 Pro), and I figured we'd be here in a year or two. But no, we're here today. A basically frontier model, running for the cost of electricity (free with a rounding…
I’m still trying to understand what is the biggest group of people that uses local AI (or will)? Students who don’t want to pay but somehow have the hardware? Devs who are price conscious and want free agentic coding? Local, in my experience, can’t even pull data from an image without hallucinating (Qwen 2.5 VI in that example). Hopefully local/small models keep getting better and devices get better at running bigger…
I don’t have much experience with local vision models, but for text questions the latest local models are quite good. I’ve been using Qwen 3 Coder 30B-A3B a lot to analyze code locally and it has been great. While not as good as the latest big cloud models, it’s roughly on par with SOTA cloud models from late last year in my usage. I also run Qwen 3 235B-A22B 2507 Instruct on my home server, and it’s great, roughly on par with Claude 4 Sonnet in my usage (but slow of course running on my DDR4-equipped server with no GPU).
The lede is being missed imo. gpt-oss:20b is a top ten model (on MMLU (right behind Gemini-2.5-Pro) and I just ran it locally on my Macbook Air M3 from last year. I've been experimenting with a lot of local models, both on my laptop and on my phone (Pixel 9 Pro), and I figured we'd be here in a year or two. But no, we're here today. A basically frontier model, running for the cost of electricity (free with a rounding…
I’m still trying to understand what is the biggest group of people that uses local AI (or will)? Students who don’t want to pay but somehow have the hardware? Devs who are price conscious and want free agentic coding? Local, in my experience, can’t even pull data from an image without hallucinating (Qwen 2.5 VI in that example). Hopefully local/small models keep getting better and devices get better at running bigger…
Earlier quoted context omitted.
An A100 is probably 2-4k tokens/second on a 20B model with batched inference. Multiply the number of A100's you need as necessary. Here, you don't really need the ram. If you could accept fewer tokens/second, you could do it much cheaper with consumer graphics cards. Even with A100, the sweet-spot in batching is not going to give you 1k/process/second. Of course, you could go up to H100...
You can batch only if you have distinct chat in parallel,
> We introduce gpt-oss-120b and gpt-oss-20b, two open-weight reasoning models available under the Apache 2.0 license and our gpt-oss usage policy. [0] Is it even valid to have additional restriction on top of Apache 2.0? [0]: https://openai.com/index/gpt-oss-model-card/
you can just do things
For example, GPL has a "no-added-restrictions" clause, which allows the recipient of the software to ignore any additional restrictions added alongside the license.
> All other non-permissive additional terms are considered “further restrictions” within the meaning of section 10. If the Program as you received it, or any part of it, contains a notice stating that it is governed by this License along with a term that is a further restriction, you may remove that term. If a license document contains a further restriction but permits relicensing or conveying under this License, you may add to a covered work material governed by the terms of that license document, provided that the further restriction does not survive such relicensing or conveying.