tripplyons

7 ms·

91.
by tripplyons 1y ago
I was curious about what was released, and the parameters for AlphaFold V3 are only given to certain groups for non-commercial use: https://github.com/google-deepmind/alphafold3?tab=readme-ov-... However, it seems like
92.
by tripplyons 1y ago
They use Bing: https://www.forbes.com/sites/katherinehamilton/2023/05/23/ch...
93.
by tripplyons 1y ago
More competition in the space would be great for me as a consumer, but the problem is that the high fixed costs make starting an index difficult.
94.
by tripplyons 1y ago
I haven't needed to tweak mine for similar reasons, but I'm surprised to hear that the "code that triggers the searches" is slow. Are you referring to something in Open WebUI?
95.
by tripplyons 1y ago
Based on the fact that there are very few up-to-date English-language search indexes (Google, Bing, and Brave if you count it), it must be incredibly costly. I doubt they are maintaining their own.
96.
Free web search MCP using a local SearXNG instance(gist.github.com)
2 points by tripplyons 1y ago|1 comments
97.
by tripplyons 1y ago
I saw the Ollama web search MCP announcement and decided to share what I've been using as a web search MCP. It runs for free by sending requests to a local SearXNG instance.
98.
by tripplyons 1y ago
You can try removing search engines that fail or reducing their timeout setting to something faster than the default of a few seconds.
99.
by tripplyons 1y ago
I have no idea how well Ollama's works, but I haven't ran into any issues with SearXNG. The alternatives aren't worth paying for in any use case I've encountered.
100.
by tripplyons 1y ago
Just set up SearXNG locally if you want a free/local web search MCP: https://gist.github.com/tripplyons/a2f9d8bd553802f9296a7ec3b...
101.
by tripplyons 1y ago
How does this trash end up in Bloomberg? I get that it's an opinion piece, but how do they allow this kind of stuff take advantage of their reputable domain name?
102.
by tripplyons 1y ago
I have can host it on my M3 laptop somewhere around 30-40 tokens per second using mlx_lm's server command: mlx_lm.server --model mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit --trust-remote-code --port 4444 I'm not sure if t
103.
by tripplyons 1y ago
At that point it is not following a diffusion training objective. I am aware of papers that do this, but I have not seen one that shows it as a better pretraining objective than something like v-prediction or flow matching.
104.
Potential plagiarism in Hierarchical Reasoning Model paper(twitter.com)
2 points by tripplyons 1y ago|0 comments
105.
by tripplyons 1y ago
There are definitely parallels between diffusion and reasoning models, mostly being able to spend longer to get a better solution by using a more precise ODE solver for diffusion or using more tokens for reasoning. However, due to how diffu
106.
by tripplyons 1y ago
Have you explored chunkwise parallel approaches? They use the O(n log n) parallel algorithm within a subsequence and update bewteen chunks recurrently like the O(n) recurrent algorithm. These are usually the fastest kernels for these kinds
107.
by tripplyons 1y ago
I think of analogue computing as more continuous than branchy from what I have heard about it. I don't know much about it though.
108.
by tripplyons 1y ago
The output of the recurrence is still dependent on previous tokens, but it usually less expressive within the recurrence in order make parallelism possible. In MinGRU the main operation used to share information between tokens is addition (
109.
by tripplyons 1y ago
Typically the fastest approaches for associative RNNs combine the advantages of the parallel O(n log n) algorithm with a recurrent non-parallel O(n) approach by computing results for subsequence chunks in parallel and moving to the next chu
110.
by tripplyons 1y ago
I would take this time to set up earthquake alerts on your phone using an app like MyShake if you live in an area where earthquakes happen.
111.
by tripplyons 1y ago
I have clarified which definition I used.
112.
by tripplyons 1y ago
Thank you for the information! I was not aware of the FSF's definition.
113.
by tripplyons 1y ago
Open source is a more informative term for this than free software. Not all free software is open source, but all open source software is free. Edit: I was not aware of the FSF's definition. I was using a definition of free software be
114.
by tripplyons 1y ago
Is this up to date? How are there still active packages with the payload?
115.
by tripplyons 1y ago
The grok-code-fast-1 model is really impressive to me and is currently free on many platforms (GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf). It's also only $0.20 per million input tokens and has good prom
116.
by tripplyons 1y ago
If I run: pnpm config set -g minimumReleaseAge 1440 Does that work as well? I can't tell if the global settings are the same as workspace settings, and it lets me set nonsense keys this way, so I'm not sure if there is a global eq
117.
by tripplyons 1y ago
I wish Asahi worked on my M3. It is a great effort, and sadly, they don't have enough resources to focus on the newer chips yet.
118.
by tripplyons 1y ago
Is that where you approximate a partial derivative as a difference in loss over a small difference in a single parameter's value? Seems like a great way to verify results, but it has the same downsides as forward mode automatic differe
119.
by tripplyons 1y ago
I've only accounted for real numbers. I'm not sure how to cleanly account for conjugates when some of the einsums would need them and others wouldn't. For example, a matrix product would need a complex conjugate, but a Hadama
120.
by tripplyons 1y ago
Thank you!

More ›