tiny-recursive-model 0.0.5__tar.gz → 0.0.6__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: tiny-recursive-model
3
- Version: 0.0.5
3
+ Version: 0.0.6
4
4
  Summary: Tiny Recursive Model
5
5
  Project-URL: Homepage, https://pypi.org/project/tiny-recursive-model/
6
6
  Project-URL: Repository, https://github.com/lucidrains/tiny-recursive-model
@@ -47,7 +47,7 @@ Description-Content-Type: text/markdown
47
47
 
48
48
  <img width="300" alt="trm-fig1" src="https://github.com/user-attachments/assets/950db79e-5f9c-4fec-a4e4-7b9355b39ce8" />
49
49
 
50
- ## Tiny Recursive Model (TRM) wip
50
+ ## Tiny Recursive Model (TRM)
51
51
 
52
52
  Implementation of [Tiny Recursive Model](https://arxiv.org/abs/2510.04871) (TRM), improvement to [HRM](https://github.com/lucidrains/hrm) from Sapient AI, by [Alexia Jolicoeur-Martineau](https://ajolicoeur.wordpress.com/about/)
53
53
 
@@ -1,7 +1,7 @@
1
1
 
2
2
  <img width="300" alt="trm-fig1" src="https://github.com/user-attachments/assets/950db79e-5f9c-4fec-a4e4-7b9355b39ce8" />
3
3
 
4
- ## Tiny Recursive Model (TRM) wip
4
+ ## Tiny Recursive Model (TRM)
5
5
 
6
6
  Implementation of [Tiny Recursive Model](https://arxiv.org/abs/2510.04871) (TRM), improvement to [HRM](https://github.com/lucidrains/hrm) from Sapient AI, by [Alexia Jolicoeur-Martineau](https://ajolicoeur.wordpress.com/about/)
7
7
 
@@ -1,6 +1,6 @@
1
1
  [project]
2
2
  name = "tiny-recursive-model"
3
- version = "0.0.5"
3
+ version = "0.0.6"
4
4
  description = "Tiny Recursive Model"
5
5
  authors = [
6
6
  { name = "Phil Wang", email = "lucidrains@gmail.com" }
@@ -86,7 +86,7 @@ class Trainer(Module):
86
86
 
87
87
  self.optim = optim_klass(
88
88
  model.parameters(),
89
- lr = learning_rate,
89
+ lr = learning_rate / (batch_size * max_recurrent_steps),
90
90
  weight_decay = weight_decay
91
91
  )
92
92
 
@@ -220,4 +220,4 @@ class TinyRecursiveModel(Module):
220
220
 
221
221
  losses = (loss, halt_loss)
222
222
 
223
- return (total_loss.mean(), losses, *return_package)
223
+ return (total_loss.sum(), losses, *return_package)