🐈‍⬛

Roman Knyazhitskiy knyazer

🐈‍⬛

Petting some cats 🐈‍⬛

knyazer / distill_dataset.py

Last active May 12, 2026 15:08

interface for distillation dataset


	class PromptStore:
	path: Path
	name: str # dataset name
	hash: str # block-wise hash of the mixture or smth like that

	@classmethod
	def init(cls, path: Path):
	# load the dataset, dump into the path with some mixing
	# or not load it, but just reuse from the path + some heuristic way to check that metadata matches remote

knyazer / gist:4419899ebb4ec8b7aae12c0c610108ae

Created February 2, 2026 23:28

prefill.py

	@eqx.filter_jit
	def _make_chunks(
	self,
	token_ids: Int[Array, "batch tokens"],
	lengths_without_padding: Int[Array, " batch"] \| None,
	chunk_size: int,
	) -> Chunk:
	batch_size, sequence_length = token_ids.shape

	if lengths_without_padding is None:

knyazer / log.txt

Created January 23, 2026 21:10

blowing on 3.12.12

	Using CPython 3.12.12
	Removed virtual environment at: .venv
	Creating virtual environment at: .venv
	Installed 158 packages in 119ms
	===================================================== test session starts =====================================================
	platform linux -- Python 3.12.12, pytest-9.0.1, pluggy-1.6.0
	rootdir: /home/knyaz/workspace/lalamo
	configfile: pyproject.toml
	plugins: anyio-4.11.0, hypothesis-6.150.0, jaxtyping-0.3.3, xdist-3.8.0, env-1.2.0
	collected 163 items

knyazer / add_to_fish_config.fish

Last active January 21, 2025 21:51

Bringing AI to Git commit.

	function gcm
	# Check if llm is installed, if not, install it
	if not type -q llm
	echo "'llm' is not installed. Attempting to install it using pip..."
	if pip install llm
	echo "'llm' installed successfully."
	else
	echo "Failed to install 'llm'. Please install it manually and try again."
	return 1
	end