Laghu: a 152M chat model trained from scratch for $146. Everything is open.
• 3
AI & ML interests
AI research from first principles.
Recent Activity
Articles
any sample or example videos?
1
#1 opened 2 days ago
by 
AsThirtyThree
jiteshkprasad
updated a
collection 1 day ago
jiteshkprasad
updated a
Space 1 day ago
jiteshkprasad
published a
Space 1 day ago
jiteshkprasad
updated a
Space 1 day ago
jiteshkprasad
updated a
collection 1 day ago
jiteshkprasad
posted an update 1 day ago
Post
41
Okay so, we at 
udgamlabs just released Lekhani, a Hindi transliterator that is 0.87 MB in total. You type Hindi in English letters, like "kaise ho", and it writes कैसे हो.
The whole thing is a 0.81M-parameter model at 4 bits, a 200,000-word list and a 5 KB JavaScript runner with no dependencies. It runs in the browser or in Node, so nothing you type leaves your device.
On the Aksharantar Hindi test it gets 65.9% of words exactly right, 7 points more than IndicXlit's model on its own (58.9%), which has 14x more parameters. IndicXlit's full package is still ahead at 68.7%, but its word list alone takes 358 MB of memory, about 400 times the size of all of Lekhani. I think for a web page or an app, that difference matters a lot.
It reads one word at a time, so "main" and "mein" can come out the same, and a few common words still come out wrong. The card has the full table, the other open models we measured and where it falls short.
It's Apache-2.0 for the model and runner, and CC0 for the word list. If you type Hindi in English letters, try it and tell me which words it gets wrong.
Model: udgamlabs/Lekhani
Try it: https://udgamlabs.com/models/lekhani
Code: https://github.com/udgamlabs/lekhani
The whole thing is a 0.81M-parameter model at 4 bits, a 200,000-word list and a 5 KB JavaScript runner with no dependencies. It runs in the browser or in Node, so nothing you type leaves your device.
On the Aksharantar Hindi test it gets 65.9% of words exactly right, 7 points more than IndicXlit's model on its own (58.9%), which has 14x more parameters. IndicXlit's full package is still ahead at 68.7%, but its word list alone takes 358 MB of memory, about 400 times the size of all of Lekhani. I think for a web page or an app, that difference matters a lot.
It reads one word at a time, so "main" and "mein" can come out the same, and a few common words still come out wrong. The card has the full table, the other open models we measured and where it falls short.
It's Apache-2.0 for the model and runner, and CC0 for the word list. If you type Hindi in English letters, try it and tell me which words it gets wrong.
Model: udgamlabs/Lekhani
Try it: https://udgamlabs.com/models/lekhani
Code: https://github.com/udgamlabs/lekhani
- 4 replies
jiteshkprasad
updated 3
Spaces 1 day ago
jiteshkprasad
published a
model 1 day ago
jiteshkprasad
updated 5
Spaces 2 days ago
Udgam-384M · Udgam Labs
◾
384M tool-calling model on a pretend phone
Gargi · Udgam Labs
◾
384M model trained from scratch that thinks step by step
Laghu · Udgam Labs
◾
1
152M chat model trained from scratch for $146
Vemana · Udgam Labs
◾
1
278M Telugu base model that continues Telugu text
Shruti · Udgam Labs
◾
1
29.5 MB offline speech-to-text for Hindi, English, Hinglish
jiteshkprasad
updated a
model 3 days ago
jiteshkprasad
updated 3
collections 3 days ago