mgoin commited on
Commit
b40ac55
·
verified ·
1 Parent(s): d1837e8

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +36 -0
README.md ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ```python
2
+ model_id = "mgoin/starcoderbase-1b-pruned50-quant"
3
+
4
+ # Load model with SparseAutoModel
5
+ from sparseml.transformers.utils import SparseAutoModel
6
+ from transformers import AutoConfig
7
+
8
+ config = AutoConfig.from_pretrained(model_id) # Why does SparseAutoModel need config?
9
+ model = SparseAutoModel.text_generation_from_pretrained(model_id, config=config)
10
+
11
+ # Apply recipe to model
12
+ # Note: Really annoying we can't grab the recipe.yaml present in the uploaded model
13
+ # and you need this separate apply_recipe_structure_to_model function
14
+ from sparseml.pytorch.model_load.helpers import apply_recipe_structure_to_model
15
+ from huggingface_hub import hf_hub_download
16
+ import os
17
+
18
+ recipe_path = hf_hub_download(repo_id=model_id, filename="recipe.yaml")
19
+ apply_recipe_structure_to_model(
20
+ model=model, recipe_path=recipe_path, model_path=os.path.dirname(recipe_path)
21
+ )
22
+
23
+ # Regular HF inference
24
+ from transformers import AutoTokenizer
25
+
26
+ tokenizer = AutoTokenizer.from_pretrained(model_id)
27
+ inputs = tokenizer.encode("def print_hello_world():", return_tensors="pt")
28
+ outputs = model.generate(inputs)
29
+ print(tokenizer.decode(outputs[0]))
30
+ """
31
+ def print_hello_world():
32
+ print("Hello World!")
33
+
34
+ print_hello_world
35
+ """
36
+ ```