LNlnyan·Nov 20, 2024Submission
Levanter: Scalable, Reproducible Foundation Models with Named Tensors and Jax
2 points · Open discussionmarin-community
levanter
Python2024–2024
marin-community / levanter
levanter
Legible, Scalable, Reproducible Foundation Models with Named Tensors and Jax
2Distinct people
2Mentions
0Explicit recommenders
720GitHub stars
What to know before exploring
What it is
Legible, Scalable, Reproducible Foundation Models with Named Tensors and Jax
GitHub
Where it is used
No documented product use has been verified yet.
What people are saying
No recommendation phrase was found in the indexed comments.
No classified warning is attached to this page. Read the discussions before treating silence as approval.
Also in these discussions
No other repository is repeatedly linked in the same discussions.
This week on Hacker News
Nov 7–Nov 14 UTC · Direct repository links and separately labeled related project stories. Indexed through 11/14/2026, 5:04:57 PM UTC.
No direct repository links were found in the indexed HN discussions over the past seven days. Older discussions are available below.
Mentions over time
HN mentions by week2 dated mentions
Observed mentions: 2024-04-24 05:05:15 UTC to 2024-11-20 14:50:35 UTC. Weeks start Monday, UTC.
Counts come from the repository’s dated source evidence. Index cutoff: 2026-11-02 08:16:59 UTC.
View mention counts and date ranges
| Period | Start (UTC) | End (exclusive, UTC) | Mentions | Submissions | Comments |
|---|---|---|---|---|---|
| Apr 22 | 2024-04-22 00:00:00 UTC | 2024-04-29 00:00:00 UTC | 1 | 0 | 1 |
| Apr 29 | 2024-04-29 00:00:00 UTC | 2024-05-06 00:00:00 UTC | 0 | 0 | 0 |
| May 6 | 2024-05-06 00:00:00 UTC | 2024-05-13 00:00:00 UTC | 0 | 0 | 0 |
| May 13 | 2024-05-13 00:00:00 UTC | 2024-05-20 00:00:00 UTC | 0 | 0 | 0 |
| May 20 | 2024-05-20 00:00:00 UTC | 2024-05-27 00:00:00 UTC | 0 | 0 | 0 |
| May 27 | 2024-05-27 00:00:00 UTC | 2024-06-03 00:00:00 UTC | 0 | 0 | 0 |
| Jun 3 | 2024-06-03 00:00:00 UTC | 2024-06-10 00:00:00 UTC | 0 | 0 | 0 |
| Jun 10 | 2024-06-10 00:00:00 UTC | 2024-06-17 00:00:00 UTC | 0 | 0 | 0 |
| Jun 17 | 2024-06-17 00:00:00 UTC | 2024-06-24 00:00:00 UTC | 0 | 0 | 0 |
| Jun 24 | 2024-06-24 00:00:00 UTC | 2024-07-01 00:00:00 UTC | 0 | 0 | 0 |
| Jul 1 | 2024-07-01 00:00:00 UTC | 2024-07-08 00:00:00 UTC | 0 | 0 | 0 |
| Jul 8 | 2024-07-08 00:00:00 UTC | 2024-07-15 00:00:00 UTC | 0 | 0 | 0 |
| Jul 15 | 2024-07-15 00:00:00 UTC | 2024-07-22 00:00:00 UTC | 0 | 0 | 0 |
| Jul 22 | 2024-07-22 00:00:00 UTC | 2024-07-29 00:00:00 UTC | 0 | 0 | 0 |
| Jul 29 | 2024-07-29 00:00:00 UTC | 2024-08-05 00:00:00 UTC | 0 | 0 | 0 |
| Aug 5 | 2024-08-05 00:00:00 UTC | 2024-08-12 00:00:00 UTC | 0 | 0 | 0 |
| Aug 12 | 2024-08-12 00:00:00 UTC | 2024-08-19 00:00:00 UTC | 0 | 0 | 0 |
| Aug 19 | 2024-08-19 00:00:00 UTC | 2024-08-26 00:00:00 UTC | 0 | 0 | 0 |
| Aug 26 | 2024-08-26 00:00:00 UTC | 2024-09-02 00:00:00 UTC | 0 | 0 | 0 |
| Sep 2 | 2024-09-02 00:00:00 UTC | 2024-09-09 00:00:00 UTC | 0 | 0 | 0 |
| Sep 9 | 2024-09-09 00:00:00 UTC | 2024-09-16 00:00:00 UTC | 0 | 0 | 0 |
| Sep 16 | 2024-09-16 00:00:00 UTC | 2024-09-23 00:00:00 UTC | 0 | 0 | 0 |
| Sep 23 | 2024-09-23 00:00:00 UTC | 2024-09-30 00:00:00 UTC | 0 | 0 | 0 |
| Sep 30 | 2024-09-30 00:00:00 UTC | 2024-10-07 00:00:00 UTC | 0 | 0 | 0 |
| Oct 7 | 2024-10-07 00:00:00 UTC | 2024-10-14 00:00:00 UTC | 0 | 0 | 0 |
| Oct 14 | 2024-10-14 00:00:00 UTC | 2024-10-21 00:00:00 UTC | 0 | 0 | 0 |
| Oct 21 | 2024-10-21 00:00:00 UTC | 2024-10-28 00:00:00 UTC | 0 | 0 | 0 |
| Oct 28 | 2024-10-28 00:00:00 UTC | 2024-11-04 00:00:00 UTC | 0 | 0 | 0 |
| Nov 4 | 2024-11-04 00:00:00 UTC | 2024-11-11 00:00:00 UTC | 0 | 0 | 0 |
| Nov 11 | 2024-11-11 00:00:00 UTC | 2024-11-18 00:00:00 UTC | 0 | 0 | 0 |
| Nov 18 | 2024-11-18 00:00:00 UTC | 2024-11-25 00:00:00 UTC | 1 | 1 | 0 |
What Hacker News says
2 sourcesWhat are people's thoughts on how this compares to:
- EasyLM [1]
- Levanter [2]
- T5X [3]
- and more?
[1]: https://github.com/young-geng/EasyLM
[2]: https://github.com/stanford-crfm/levanter
[3]: https://github.com/google-research/t5x
Asking because I have worked extensively on training a large model on a TPU cluster, and started with Levanter, then tried MaxText, and finally ended up on EasyLM. My thoughts are:
- Levanter is well intentioned but is unproven and lacking in features. For instance, their sharding is odd in that it requires embedding dimension to be a multiple of the number of devices, so I can't test using a model with embedding dimension 768 on a 512-device pod. Lost confidence in Levanter after finding some glaring correctness bugs (and helping get them fixed). Also, while I'm a huge fan of Equinox's approach, it's sadly
Read the rest
underdeveloped (for instance, there's no way to specify non-default weight initialization strategies without manually doing model surgery to set weights).
- MaxText was just very difficult to work with. We felt like we were fighting against it every time we needed to change something because we would be digging through numerous needless layers of abstraction. My favorite was after one long day of debugging, I found a function who's only purpose was to pass its arguments to another function untouched; this function's only purpose was to pass its arguments untouched to a new, third function, that then slightly changed them and passed them to a fourth function that did the work.
- EasyLM is, as the name says, easy. But on a deeper dive, the sharding functionality seems to be underdeveloped. What they call "FSDP" is not necessarily true FSDP, it's literally just a certain axis that the JAX mesh is being sharded around that happens to shard some data axes and some model weight axes.
I'm still searching for a "perfect" JAX LLM codebase - any pointers?