Instructions to use ssdataanalysis/gemma-4-E2B-hebrew-first with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use ssdataanalysis/gemma-4-E2B-hebrew-first with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("ssdataanalysis/gemma-4-E2B-hebrew-first", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Fix: add assistant_only_loss=False to prevent all labels being masked to -100
Browse files
train.py
CHANGED
|
@@ -122,6 +122,7 @@ training_args = SFTConfig(
|
|
| 122 |
weight_decay=0.01,
|
| 123 |
max_length=2048,
|
| 124 |
packing=False,
|
|
|
|
| 125 |
bf16=True,
|
| 126 |
logging_strategy="steps",
|
| 127 |
logging_steps=10,
|
|
|
|
| 122 |
weight_decay=0.01,
|
| 123 |
max_length=2048,
|
| 124 |
packing=False,
|
| 125 |
+
assistant_only_loss=False, # CRITICAL FIX: prevent all labels being masked to -100
|
| 126 |
bf16=True,
|
| 127 |
logging_strategy="steps",
|
| 128 |
logging_steps=10,
|