Instructions to use efficiency-learning/muon-gradientMatching with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use efficiency-learning/muon-gradientMatching with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("efficiency-learning/muon-gradientMatching", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Add model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model: Qwen/Qwen2.5-0.5B
|
| 3 |
+
library_name: transformers
|
| 4 |
+
tags:
|
| 5 |
+
- qwen2
|
| 6 |
+
- pretraining
|
| 7 |
+
- mu-greats
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# muon-gradientMatching
|
| 11 |
+
|
| 12 |
+
Qwen2.5-0.5B models pretrained from scratch on `openwebtext-100k`, from the Mu-GREATS subset-selection / Muon gradient-matching experiments.
|
| 13 |
+
Each run is a subfolder; load one with:
|
| 14 |
+
|
| 15 |
+
```python
|
| 16 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 17 |
+
repo = "efficiency-learning/muon-gradientMatching"
|
| 18 |
+
sub = "full_muon_lr0.001" # any folder below
|
| 19 |
+
model = AutoModelForCausalLM.from_pretrained(repo, subfolder=sub)
|
| 20 |
+
tokenizer = AutoTokenizer.from_pretrained(repo, subfolder=sub)
|
| 21 |
+
```
|
| 22 |
+
|
| 23 |
+
| Subfolder | Optimizer | LR | Mode | Step | dtype |
|
| 24 |
+
|---|---|---|---|---|---|
|
| 25 |
+
| `full_adamw_lr0.0001` | adamw | 0.0001 | full | 6317 | ? |
|
| 26 |
+
| `full_adamw_lr0.0005` | adamw | 0.0005 | full | 6317 | ? |
|
| 27 |
+
| `full_adamw_lr0.001` | adamw | 0.001 | full | 6317 | ? |
|
| 28 |
+
| `full_adamw_lr0.005` | adamw | 0.005 | full | 6317 | ? |
|
| 29 |
+
| `full_adamw_lr5e-05` | adamw | 5e-05 | full | 6317 | ? |
|
| 30 |
+
| `full_muon_lr0.0001` | muon | 0.0001 | full | 6317 | ? |
|
| 31 |
+
| `full_muon_lr0.0002` | muon | 0.0002 | full | 6317 | ? |
|
| 32 |
+
| `full_muon_lr0.0005` | muon | 0.0005 | full | 6317 | ? |
|
| 33 |
+
| `full_muon_lr0.001` | muon | 0.001 | full | 6317 | ? |
|
| 34 |
+
| `full_muon_lr0.005` | muon | 0.005 | full | 6317 | ? |
|
| 35 |
+
| `full_muon_lr5e-05` | muon | 5e-05 | full | 6317 | ? |
|