Update README.md
Browse files
README.md
CHANGED
|
@@ -7,16 +7,10 @@ tags:
|
|
| 7 |
datasets:
|
| 8 |
- giulio98/xlcost-single-prompt
|
| 9 |
widget:
|
| 10 |
-
- text: "
|
| 11 |
-
example_title: "
|
| 12 |
-
- text: "def print_hello_world():\n\t"
|
| 13 |
-
example_title: "Hello World!"
|
| 14 |
-
- text: "def get_file_size(filepath):"
|
| 15 |
-
example_title: "File size"
|
| 16 |
-
- text: "import numpy as"
|
| 17 |
-
example_title: "Numpy"
|
| 18 |
model-index:
|
| 19 |
-
- name: codegen-350M-multi-xlcost
|
| 20 |
results:
|
| 21 |
- task:
|
| 22 |
name: Code Generation
|
|
@@ -43,8 +37,8 @@ You can load the CodeGen-350M-multi-xlcost model and tokenizer directly in `tran
|
|
| 43 |
|
| 44 |
```Python
|
| 45 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 46 |
-
tokenizer = AutoTokenizer.from_pretrained("giulio98/codegen-350M-multi-xlcost
|
| 47 |
-
model = AutoModelForCausalLM.from_pretrained("giulio98/codegen-350M-multi-xlcost
|
| 48 |
|
| 49 |
text = tokenizer.eos_token + "\'\'\'\n" + "function to add two numbers" + "\n\'\'\'\n" + "###\n"
|
| 50 |
input_ids = tokenizer(text, return_tensors="pt").input_ids
|
|
@@ -67,7 +61,7 @@ The model was finetuned on [XLCost-single-prompt](https://huggingface.co/dataset
|
|
| 67 |
xlcost-text-to-code](https://huggingface.co/datasets/codeparrot/xlcost-text-to-code). Below the hyperparameters.
|
| 68 |
|
| 69 |
|
| 70 |
-
|
| 71 |
|Per device train batch size| 8 |
|
| 72 |
|Context size| 1024 |
|
| 73 |
|Training steps| 258|
|
|
@@ -84,8 +78,8 @@ The training was executed on 1 x V100 (16GB) GPU for 6h 42m
|
|
| 84 |
|
| 85 |
We evaluated the model on the first 400 samples of XLCOST's [XLCost-single-prompt test split](https://huggingface.co/datasets/giulio98/xlcost-single-prompt/viewer/Python/test) and comparing the outputs of the generated codes with respect to the expected output using pass@k metric.
|
| 86 |
|
| 87 |
-
| Metric | codegen-350M-multi-xlcost
|
| 88 |
-
|
| 89 |
|pass@1 | 3.70% |
|
| 90 |
|pass@10 | 14.5% |
|
| 91 |
|
|
|
|
| 7 |
datasets:
|
| 8 |
- giulio98/xlcost-single-prompt
|
| 9 |
widget:
|
| 10 |
+
- text: "'''\nfunction to add two numbers\n'''\n###\n"
|
| 11 |
+
example_title: "add two numbers"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
model-index:
|
| 13 |
+
- name: codegen-350M-multi-xlcost
|
| 14 |
results:
|
| 15 |
- task:
|
| 16 |
name: Code Generation
|
|
|
|
| 37 |
|
| 38 |
```Python
|
| 39 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 40 |
+
tokenizer = AutoTokenizer.from_pretrained("giulio98/codegen-350M-multi-xlcost")
|
| 41 |
+
model = AutoModelForCausalLM.from_pretrained("giulio98/codegen-350M-multi-xlcost")
|
| 42 |
|
| 43 |
text = tokenizer.eos_token + "\'\'\'\n" + "function to add two numbers" + "\n\'\'\'\n" + "###\n"
|
| 44 |
input_ids = tokenizer(text, return_tensors="pt").input_ids
|
|
|
|
| 61 |
xlcost-text-to-code](https://huggingface.co/datasets/codeparrot/xlcost-text-to-code). Below the hyperparameters.
|
| 62 |
|
| 63 |
|
| 64 |
+
|------|------------------|
|
| 65 |
|Per device train batch size| 8 |
|
| 66 |
|Context size| 1024 |
|
| 67 |
|Training steps| 258|
|
|
|
|
| 78 |
|
| 79 |
We evaluated the model on the first 400 samples of XLCOST's [XLCost-single-prompt test split](https://huggingface.co/datasets/giulio98/xlcost-single-prompt/viewer/Python/test) and comparing the outputs of the generated codes with respect to the expected output using pass@k metric.
|
| 80 |
|
| 81 |
+
| Metric | codegen-350M-multi-xlcost |
|
| 82 |
+
|--------|-----|
|
| 83 |
|pass@1 | 3.70% |
|
| 84 |
|pass@10 | 14.5% |
|
| 85 |
|