Radi Radev radiradev

Anti-hype LLM reading list

Goals: Add links that are reasonable and good explanations of how stuff works. No hype and no vendor content if possible. Practical first-hand accounts and experience preferred (super rare at this point).

My own notes from a few months back.

	"""
	Conditional Flow Matching on the Two Moons dataset.

	We train a small velocity-field network v_theta(x, t, c) to transport samples
	from a standard Gaussian (t=0) to the two-moons distribution (t=1), conditioned
	on the moon label c in {0, 1}.

	Flow matching objective (linear / OT-style interpolant):
	x_t = (1 - t) * x_0 + t * x_1, x_0 ~ N(0, I), x_1 ~ data
	target velocity: u_t = x_1 - x_0

	nodes:
	data:
	node_builder:
	_target_: anemoi.graphs.nodes.ReducedGaussianGridNodes
	grid: o48
	attributes:
	area_weight:
	_target_: anemoi.graphs.nodes.attributes.SphericalAreaWeights
	norm: unit-max
	hidden:

	FROM rradev/minkowski-engine:latest

	RUN pip install timm tensorboardx einops pytorch-lightning

	CMD ["bash"]

	# the minkowski-engine:latest uses the dockerfile defined here https://gist.github.com/radiradev/7af53500133773948c813d3e133e62c8

	FROM deeplearnphysics/larcv2:ub20.04-cuda11.6-pytorch1.13-larndsim

	# Install ninja
	RUN pip install ninja


	ENV TORCH_CUDA_ARCH_LIST="6.0 7.0 7.5 8.0"

	# Install MinkowskiEngine with DepthwiseConvolution from GitHub
	RUN pip install -U --target=/usr/local/lib/python3.8/dist-packages \

Radi Radev radiradev

Anti-hype LLM reading list

Background