Thomas Dybdahl Ahle

Results 68 issues of Thomas Dybdahl Ahle

Most of the example in the README is intuitive, but the output of `autgrp` isn't. The source code has: ``` Compute the automorphism group of a graph. *g* A Graph...

**Is your feature request related to a problem? Please describe.** I'm trying to do something like [Getting 50% (SoTA) on ARC-AGI with GPT-4o](https://www.lesswrong.com/posts/Rdwui3wHxCeKb7feK/getting-50-sota-on-arc-agi-with-gpt-4o), and it requires drawing thousands of independent...

Consider this DSPy program: ```python demos = [dspy.Example(input="What is the speed of light?", output="3e8")] program = LabeledFewShot(k=len(demos)).compile( student=dspy.TypedPredictor("input -> thoughts, output"), trainset=[ex.with_inputs("input") for ex in demos], ) dspy.settings.configure(lm=DummyLM(["My thoughts", "Paris"]))...

Behavior 2.5

`dsp.modules.lm` is nearly all special cases: We should think about how to move those decisions into the LM subclasses instead. Maybe this is already happening in the backend refactor @KCaverly...

enhancement
LM Issues

Since 2df6a428cf6bcc4e4a08f15e3f7caef9ce5f4f61 it is possible to store every datapoint in `n` lists by building with `ivf.build(n_probes=n)`. This increases performance recall/qps quite a lot, but only when going from `n=1` to...

enhancement

AVX-512 has some nice features, such as support for fast float16 operations. This might allow us to do rescoring very fast. The Quicker ADC paper also mentions some uses of...

enhancement

Currently we are not using python's typing functionality.

enhancement
good first issue

Often we use PQ to estimate the distance from a full precision vector to a bunch of compressed points. However, we can also try to compute the distance between all...

enhancement
good first issue

Currently `IVF.fit(...)` uses brute force nearest neighbours to find which clusters to insert the points into. Instead we could use the same `PQ.top(...)` method that we use to do queries...

enhancement

A classical way to make building the index faster, cheaper memory wise, and potentially better (bigger, but lower quality) is to use a top level product code. Instead of just...

enhancement