Code for an encoder-decoder LLM which utilizes recurrent "reasoning" layers to advance performance despite low param count
active 2025-06-23 → 2025-09-10 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 80 days from 2025-06-23 to 2025-09-10. Pushes: 46 total, peak 7 in a day. Pull requests: 26 total, peak 5 in a day. Issues: 15 total, peak 4 in a day. Comments: 3 total, peak 1 in a day. Stars: 0 total, peak 0 in a day.
- Pushes
- Pull requests
- Issues
- Comments
- Stars
Stars, PRs, issues and forks are under-captured in the later part of this window. GH Archive progressively stopped capturing non-push events during 2026 — −95% or worse by the end of the window. Every series here except Pushes fades for that reason, so a decline above reflects the archive, not this repository. Pushes stay reliable throughout, so read them, and the contributor counts derived from them, as the real signal. Data health has the measurements.
Top contributors
Pushes, PRs, issues, reviews and comments — stars and forks excluded, so this is contribution rather than popularity
| Contributor | Contributions | Pushes | PRs | Comments |
|---|---|---|---|---|
| ashlab11 | 88 | 46 | 26 | 1 |
| cursor[bot] | 2 | 0 | 0 | 2 |
Recent activity
Latest issues, pull requests and releases
- Pull request#27ashlab112025-09-08 01:26
- Pull request#29ashlab112025-09-08 01:26
- Pull request#28ashlab112025-09-08 00:24
- Pull request#28ashlab112025-09-08 00:22
- Pull request#27ashlab112025-09-08 00:21
- Pull request#26ashlab112025-09-07 23:19
- Pull request#26ashlab112025-09-07 23:19
- Pull request#25ashlab112025-09-07 23:18
- Issue#24ashlab112025-09-07 22:03Fix conversation collator / main model, needs to work with different inputs (not just input_ids)
- Pull request#23ashlab112025-09-07 21:54
- Pull request#23ashlab112025-09-07 21:53
- Issue comment#22ashlab112025-09-05 18:15Build a continuous latent thought model
- Issue#22ashlab112025-09-05 17:58Build a continuous latent thought model
- Pull request#20ashlab112025-09-03 21:19
- Issue#21ashlab112025-09-01 22:16Train both decoder and ED, compare perplexity on held-out SFT set
- Pull request#20ashlab112025-09-01 22:14
- Issue#19ashlab112025-09-01 17:42Update SFT so it works with batching
- Issue#18ashlab112025-09-01 17:36Fix sink tokens
- Issue#9ashlab112025-09-01 17:34Fully train xavier-initialized model on 500M tokens, look at ppl on hold-out set
- Pull request#17ashlab112025-08-29 20:46
- Issue comment#16cursor[bot]2025-08-24 21:30Consolidate attention parameters into flex_dict
- Pull request#16ashlab112025-08-24 21:29
- Pull request#13ashlab112025-08-18 05:00
- Pull request#14ashlab112025-08-18 05:00
- Pull request#12ashlab112025-08-18 05:00
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 0 stars here means stars gained during the window, not the repo's star count.