Iolaum

4 ms·

31.
by Iolaum 2mo ago
It was a configuration error ...
32.
by Iolaum 2mo ago
This is something you should be asking ChatGPT, and continue the conversation in a way that you learn what you need.
33.
by Iolaum 2mo ago
I would still pay ~500 for a chip that runs 10kt/s of a ~100b model on my machine even if the half life is 6months. I m sure my employer would too. Qwwen3.5 122b was released 6 months ago and is still best in class overall 100-140 B pa
34.
by Iolaum 2mo ago
Hello!? Us consumers need it too.
35.
by Iolaum 2mo ago
Yea and we are reaching the point where this benchmaxing is visible in the model's reported overthinking.
36.
by Iolaum 2mo ago
Especially with many of those datacenters being built next to residential areas ... </s>
37.
by Iolaum 2mo ago
The result is moronic but given the confines a judge is operating what else could he do? Is it legal for you to buy a book and tear it down? The problem is the law, and the people who vote for the legislative bodies not asking more from the
38.
by Iolaum 2mo ago
Looks like gpt-5.6-luna is a better option that ds4-flash in that llm market segment with the new pricing.
39.
by Iolaum 2mo ago
TBH Taalas was a company I was existed about as a consumer. A dense model like gemma4-31b or qwen3.6-27b running at 10k t/s sounds like an awesome thing to have. Would be willing to pay GPU prices for it.
40.
by Iolaum 2mo ago
There's an old joke that goes about like: How do you solve Global Warming? Nuclear Winter ... I really hope this is not how we end up geo-engineering our way out of Climate change.
41.
by Iolaum 2mo ago
It's not even at society scale IMO. How many times do people at a company deprioritize small-to-medium fixes on things that may/will become a problem later? Sometimes you have to wait until a problem gets "big enough" be
42.
by Iolaum 2mo ago
Not sure this is a meaningful factor for normal circumstances (for example below 1km altitude and for water that's safe to drink). According to CGPT at 1km altitude the boiling temperature of water is 96.7 C.
43.
by Iolaum 2mo ago
Do think about b2b. Companies are already paying much more for AI. a new K3 (or similar model) every 6 months for a monthly rate of ~100$ per month is something MANY businesses would pay for. Then they could even sell them at half the price
44.
by Iolaum 2mo ago
Do they? With Fable/Opus duo they have a better product and their customers are paying for quality. They want to be the premium LLM provider letting others compete for the commoditized part of the market.
45.
by Iolaum 2mo ago
A really fast qwen-3.6-27B type of model could be useful. With a specialized harness and this speed I 'd expect it to find many applications. Implementing a coding plan is the minimum I can think of.
46.
by Iolaum 3mo ago
QWEN-3.6-27B P.S. Couldn't resist :p
47.
by Iolaum 3mo ago
That's like saying a sling is the same as an assault rifle. Yes both are weapons but scale and capabilities matter.
48.
by Iolaum 3mo ago
Personalized psyops against voters ...
49.
by Iolaum 3mo ago
Was the temperature 0? Cause unless I don't understand it right, any non-zero temperature implies probabilistic next token prediction. You did mention, seed, which I haven't seen available anywhere else (in my limited experience)
50.
by Iolaum 3mo ago
While I agree this is the case, I think this is a consequence of low education for the general public and low quality of the public discourse. If the quality of those two goes up, then WYSIWYG politics will emerge as a preferred choice. I o
51.
by Iolaum 3mo ago
Since they did this with their own harness I m not sure it's apples to apples comparison.
52.
by Iolaum 3mo ago
Hey if you can externalize that cost, more money left for you and the investors !!!
53.
by Iolaum 3mo ago
Own a framework desktop 128gb. GPU bandwidth limits the amount of tokens/sec that you get. I 've mostly been running QWEN-3.6-35B-A3B at Q8 and QWEN-3.5-122B-A10B at Q4 with Q8 kv-cache on llama.cpp (use vulkan version). Also use
54.
by Iolaum 3mo ago
It's not even IP. Model outputs are not protected (to the best of my understanding).
55.
by Iolaum 3mo ago
Maybe because a non public agreement was in place?
56.
by Iolaum 3mo ago
Open Source models decelerate growth of closed AI. For people who think (or want) AI = closed_AI then that argument has weight. Good luck getting them to update their priors.
57.
by Iolaum 3mo ago
OpenCode has an autoretry functionality (that progressively waits more after each failed request). I m surprised other harnesses don't have that.
58.
by Iolaum 3mo ago
There's an emerging practice of using Q4 quants and Q8 KV cache for local inference. At that point you can run both Qwen3.5-122B-A10B (my personal choice on Framework Desktop 128gb) and Laguna-S-2.1. Now whether that's good enough
59.
by Iolaum 3mo ago
As long as they are transparent about what quant they serve the model and any other optimization they do that also affects performance of inferred tokens.
60.
by Iolaum 3mo ago
I m not sure about that. It raises the cost of doing business, but US tech giants still get to dominate the EU tech landscape.

More ›