metadata
datasets:
- mvasiliniuc/iva-kotlin-codeint-clean-train
- mvasiliniuc/iva-kotlin-codeint-clean-valid
language:
- code
tags:
- gpt2
- code
- kotlin
- mobile
- generation
widget:
- text: "/**\n\t* A function that returns the version of the current operating system.\n*/\n"
example_title: Get current device operating system
- text: "/**\n\t* A function that returns the current TimeZone.\n*/\n"
example_title: Get current timezone
- text: "/**\n\t* A data class representing a Bank Account.\n*/\n"
example_title: Data Class - BankAccount
iva-codeint-kotlin-small GPT-2 is (small version - 239.4M parameters) trained from scratch to obtain results in the text-to-code task tailored for Kotlin language used in native mobile development (Android).
Usage
from transformers import pipeline
pipe = pipeline("text-generation", model="mvasiliniuc/iva-codeint-kotlin-small")
outputs = pipe("fun printToConsole()")
Inference
API_URL = "https://api-inference.huggingface.co/models/mvasiliniuc/iva-codeint-kotlin-small"
headers = {"Authorization": "Bearer <key>"}
def query(payload):
response = requests.post(API_URL, headers=headers, json=payload)
return response.json()
output = query({
"inputs": """
/**
* A public function that returns the current version of the operating system.
*/
"""
})
pprint.pprint(output, compact=True)
Training
| Config | Value |
|---|---|
| seq length | 1024 |
| weight decay | 0.1 |
| learning rate | 0.0005 |
| max eval steps | -1 |
| shuffle buffer | 10000 |
| max train steps | 150000 |
| mixed precision | fp16 |
| num warmup steps | 2000 |
| train batch size | 5 |
| valid batch size | 5 |
| lr scheduler type | cosine |
| save checkpoint steps | 15000 |
| gradient checkpointing | false |
| gradient accumulation steps | 1 |
Resources
Resources used for research: