dnhkng

9 ms·

1.
by dnhkng 2mo ago
Totally! This with DwarfStar delivers usable local AI (I hope!)
2.
by dnhkng 2mo ago
I think the commenter means the Flash vs Terra benchmarks.
3.
by dnhkng 2mo ago
It will be fair if they release the harness though. I think now the future will be paired model-harness releases, not just weight dumps. The performance changes are so big with the right harness that is makes sense to engineer the harness a
4.
by dnhkng 2mo ago
DeepSeek V4 Flash (Preview → 2026-07-31) • Terminal Bench: 56.9 → 82.7 (+25.8) • Toolathlon: 51.8 → 70.3 (+18.5) Compared to GPT-5.6 Terra: • Terminal Bench: Flash 82.7 vs Terra 78.4 • Toolathlon: Flash 70.3 vs Terra 53.1 • DeepSWE: Flash 5
5.
DeepSeek-V4-Flash Update(api-docs.deepseek.com)
745 points by dnhkng 2mo ago|347 comments
6.
by dnhkng 2mo ago
Maybe instead of proofs, we should encourage the publishing of prompts and reasoning traces?
7.
by dnhkng 2mo ago
Maybe instead of proofs, we should encourage the publishing of prompts and reasoning traces?
8.
by dnhkng 2mo ago
Author here: Yeah, the encoder and decoder stuff is explicit, but the internal structure in generated during training. I don't think the big labs were doing this back when I did the research; no one was back in '24. I just didn&#x
9.
by dnhkng 2mo ago
Too bad they also don't give anything back to individual researchers. Oh well, wasn't expecting much.
10.
by dnhkng 3mo ago
I have added an endpoint for CLI-dwellers: curl -X POST https://quantumlever.stream/api/magic-8-ball \ -H "Content-Type: application/json" \ -d '{"question":"Will my sampler touch
11.
Show HN: I built a hardware quantum RNG and wired it into a Magic 8-Ball(dnhkng.github.io)
13 points by dnhkng 3mo ago|1 comments
12.
by dnhkng 6mo ago
Yes, dammit. Author here. I drafted it before I left for holiday, at it's not ready to publish. It wasn't supposed to be officially posted yet, but I ran out of time before my flight. My apologies!
13.
by dnhkng 6mo ago
Thanks! I have pushed basic code to GitHub ( https://github.com/dnhkng/RYS ) Some interesting areas to explore might be a combination of deleting some layers and duplicating others. i.e. reduce VRAM by dropping some laye
14.
by dnhkng 6mo ago
Author here: The code is up on GitHub. The probes I used seem to help identify good configurations, but are quite noisey. A small probe set was initially used to make the scan tractable, and then the higher ranked models were retested on a
15.
by dnhkng 6mo ago
Author here: That was done in this blog post, in the beam search. I started with the best re-layer configs, and iteratively added more blocks, including the same multiple times, during a long beam search. It turns out this does not help (so
16.
by dnhkng 6mo ago
There was some work done on this a while back, during the FrankenMerge craze of 23' I am working with TurboDerp to integrate this into the Exllama v3 format.
17.
by dnhkng 6mo ago
Author here. Another thing I want to highlight: the language-agnostic "thinking space" finding came from Evan Maunder, who read Part 1 and ran an elegant experiment — same sentence in English, Mandarin, and Base64, cosine similari
18.
by dnhkng 6mo ago
Author here. The result that surprised me most: after evaluating 3,024 beam search candidates, training a surrogate model on ~4,600 measurements, and scoring 2 million configurations — the Pareto-optimal configs were all simple contiguous b
19.
Show HN: More LLM Neuroanatomy: A Hint of a Universal Language?(dnhkng.github.io)
1 points by dnhkng 6mo ago|0 comments
20.
by dnhkng 6mo ago
I stick with models I can run on VRAM, but DeepSeek Speciale have the best reasoning capabilities of the models I can actually run ( https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Speciale ). What hardware can you acc
21.
by dnhkng 6mo ago
Yes, I have done these thype of experiments; thats for the next post
22.
by dnhkng 6mo ago
Glad to see someone replicate the results already :)
23.
by dnhkng 6mo ago
But blogging is fun! I do wish one of the big labs would sponsor with a rack of HGX Rubin NVL8's. I have lots of ideas to test, and I have probably hit the spending limit with the boss on hardware (she hasn't seen the new power bi
24.
by dnhkng 6mo ago
Yes, thats true. But that points again to the main idea: The model has learnt to transform Base64 into a form it can already use in the 'regular' thinking structures. The alternative is that there is an entire parallel structure
25.
by dnhkng 6mo ago
Hi, thanks for the praise! On the other papers, models like SOLAR or training a model that uses a single layers are probably going to hit a wall, based on the heatmaps I found. The transformer stack starts with randomised weights, (analogou
26.
by dnhkng 6mo ago
Cheers. I will go back though my other old projects (optogenetics, hacking Crispr/CAS9 etc), and put them on my blog. On your questions: 1) A few other papers have been mentioned in the thread, like Solar10.7B. They duplicated the whol
27.
by dnhkng 6mo ago
Yes, it's an amazing time to be a hacker!
28.
by dnhkng 6mo ago
It's still non-trivial, as multi-digit numbers can be constructed a huge combination of valid tokens. The code in the blog helps derive useful metrics from partial answers.
29.
by dnhkng 6mo ago
There are similar patterns in the models from all the big labs. I think the transform layer stack starts out 'undifferentiated', analogous to stem cells. Pre-training pushes the model to develop structure and this technique helps
30.
by dnhkng 6mo ago
At some point I will clean up and share the dynamic layer modification code for oobabooga Text-Generation-WebuUI. You can enter the setting, and apply new re-layering architectures. Its very weird chatting with these brain-damaged models.

More ›