This next round of Artisanal Language Model improvements are mostly focused on efficiency rather than dramatic algorithmic improvements.
https://brandonrohrer.org/alms_somm_sparse.html
I moved the Markov models from a dense representation, using arrays, to a sparse representation, using dictionaries. That cut the storage requirements down to a few percent of their original levels.
Brandon Rohrer
npub1jh…9eag0
2026-06-21 11:17:07 UTC
Author Public Key
npub1jh4qsxnz0nhyfefjsfvcdmxxvgfe6p5vf0dvh6pq4r6ytwwxcp4sl9eag0Published at
2026-06-21 11:17:07 UTCEvent JSON
{
"id": "bc8caf4fbe6219d4cf5101b44529052aa144810ca69e15a23794c737192c1f27",
"pubkey": "95ea081a627cee44e532825986ecc662139d068c4bdacbe820a8f445b9c6c06b",
"created_at": 1782040627,
"kind": 1,
"tags": [
[
"proxy",
"https://recsys.social/@brohrer/116787814595088804",
"web"
],
[
"proxy",
"https://recsys.social/users/brohrer/statuses/116787814595088804",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://recsys.social/users/brohrer/statuses/116787814595088804",
"pink.momostr"
],
[
"-"
]
],
"content": "This next round of Artisanal Language Model improvements are mostly focused on efficiency rather than dramatic algorithmic improvements. \n\nhttps://brandonrohrer.org/alms_somm_sparse.html\n\nI moved the Markov models from a dense representation, using arrays, to a sparse representation, using dictionaries. That cut the storage requirements down to a few percent of their original levels.",
"sig": "645342acc7a6bdc9863e7a63ed35f5830e97f663327944191fe63cd009006860b85dc9ed8dc6e1b9dbd2b9948233d01ed6b1d4f8162518184746c1fc90fbbb40"
}