Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
https://huggingface.co/HuggingFaceM4/tiny-random-LlamaForCausalLM
|
| 2 |
+
|
| 3 |
+
```python
|
| 4 |
+
from sparseml.transformers import SparseAutoModelForCausalLM, SparseAutoTokenizer, oneshot
|
| 5 |
+
from sparseml.modifiers import SparseGPTModifier
|
| 6 |
+
|
| 7 |
+
model_id = "HuggingFaceM4/tiny-random-LlamaForCausalLM"
|
| 8 |
+
compressed_model_id = "mgoin/tiny-random-LlamaForCausalLM-pruned95-compressed"
|
| 9 |
+
|
| 10 |
+
# Apply SparseGPT to the model
|
| 11 |
+
oneshot(
|
| 12 |
+
model=model_id,
|
| 13 |
+
dataset="open_platypus",
|
| 14 |
+
recipe=SparseGPTModifier(sparsity=0.95),
|
| 15 |
+
output_dir="temp-output",
|
| 16 |
+
)
|
| 17 |
+
|
| 18 |
+
model = SparseAutoModelForCausalLM.from_pretrained("temp-output", torch_dtype="auto")
|
| 19 |
+
tokenizer = SparseAutoTokenizer.from_pretrained(model_id)
|
| 20 |
+
|
| 21 |
+
model.save_pretrained(compressed_model_id.split("/")[-1], save_compressed=True)
|
| 22 |
+
tokenizer.save_pretrained(compressed_model_id.split("/")[-1])
|
| 23 |
+
|
| 24 |
+
# Upload the checkpoint to Hugging Face
|
| 25 |
+
from huggingface_hub import HfApi
|
| 26 |
+
HfApi().upload_folder(
|
| 27 |
+
folder_path=compressed_model_id.split("/")[-1],
|
| 28 |
+
repo_id=compressed_model_id,
|
| 29 |
+
)
|
| 30 |
+
```
|