Update README.md
Browse files
README.md
CHANGED
|
@@ -1,12 +1,14 @@
|
|
| 1 |
-
---
|
| 2 |
-
library_name: transformers
|
| 3 |
-
tags:
|
| 4 |
-
- code
|
| 5 |
-
license: mit
|
| 6 |
-
datasets:
|
| 7 |
-
- iamtarun/python_code_instructions_18k_alpaca
|
| 8 |
-
pipeline_tag: text-generation
|
| 9 |
-
|
|
|
|
|
|
|
| 10 |
|
| 11 |
# PyCodeGen 350M
|
| 12 |
|
|
@@ -32,6 +34,10 @@ Finally model has been adapted to the Python language by training on the BigPyth
|
|
| 32 |
The dataset contains problem descriptions and code in python language.
|
| 33 |
This dataset is taken from sahil2801/code_instructions_120k, which adds a prompt column in alpaca style.
|
| 34 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
|
| 36 |
## Example of usage
|
| 37 |
|
|
@@ -68,25 +74,25 @@ print(tokenizer.decode(output[0], skip_special_tokens=True))
|
|
| 68 |
## Training parameters
|
| 69 |
|
| 70 |
BitsAndBytes:
|
| 71 |
-
- load_in_4bit
|
| 72 |
-
- bnb_4bit_quant_type
|
| 73 |
-
- bnb_4bit_use_double_quant
|
| 74 |
-
- bnb_4bit_compute_dtype
|
| 75 |
|
| 76 |
LoraConfig:
|
| 77 |
-
- r
|
| 78 |
-
- lora_alpha
|
| 79 |
-
- target_modules
|
| 80 |
-
- lora_dropout
|
| 81 |
-
- bias
|
| 82 |
-
- task_type
|
| 83 |
|
| 84 |
Finetuning:
|
| 85 |
-
- num_epochs
|
| 86 |
-
- train_batch_size
|
| 87 |
-
- eval_batch_size
|
| 88 |
-
- gradient_accumulation_steps
|
| 89 |
-
- learning_rate
|
| 90 |
-
- weight_decay
|
| 91 |
-
- lr_scheduler_name
|
| 92 |
-
- num_warmup_steps
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: transformers
|
| 3 |
+
tags:
|
| 4 |
+
- code
|
| 5 |
+
license: mit
|
| 6 |
+
datasets:
|
| 7 |
+
- iamtarun/python_code_instructions_18k_alpaca
|
| 8 |
+
pipeline_tag: text-generation
|
| 9 |
+
language:
|
| 10 |
+
- en
|
| 11 |
+
---
|
| 12 |
|
| 13 |
# PyCodeGen 350M
|
| 14 |
|
|
|
|
| 34 |
The dataset contains problem descriptions and code in python language.
|
| 35 |
This dataset is taken from sahil2801/code_instructions_120k, which adds a prompt column in alpaca style.
|
| 36 |
|
| 37 |
+
## Intended uses
|
| 38 |
+
|
| 39 |
+
The model can be used to generate python code that solves task with optionally given input data.
|
| 40 |
+
|
| 41 |
|
| 42 |
## Example of usage
|
| 43 |
|
|
|
|
| 74 |
## Training parameters
|
| 75 |
|
| 76 |
BitsAndBytes:
|
| 77 |
+
- load_in_4bit: True,
|
| 78 |
+
- bnb_4bit_quant_type: nf4,
|
| 79 |
+
- bnb_4bit_use_double_quant: True,
|
| 80 |
+
- bnb_4bit_compute_dtype: torch.bfloat16
|
| 81 |
|
| 82 |
LoraConfig:
|
| 83 |
+
- r: 32,
|
| 84 |
+
- lora_alpha: 16,
|
| 85 |
+
- target_modules: all-linear,
|
| 86 |
+
- lora_dropout: 0.1,
|
| 87 |
+
- bias: none,
|
| 88 |
+
- task_type: CASUAL_LM
|
| 89 |
|
| 90 |
Finetuning:
|
| 91 |
+
- num_epochs: 15
|
| 92 |
+
- train_batch_size: 4
|
| 93 |
+
- eval_batch_size: 8
|
| 94 |
+
- gradient_accumulation_steps: 8
|
| 95 |
+
- learning_rate: 3e-4
|
| 96 |
+
- weight_decay: 0.01
|
| 97 |
+
- lr_scheduler_name: cosine
|
| 98 |
+
- num_warmup_steps: 190
|