https://github.com/dongryeollee1/openai_asynchronous_use
Openai API Asynchronously Using Script
https://github.com/dongryeollee1/openai_asynchronous_use
asyncio batch-processing openai openai-api
Last synced: 16 days ago
JSON representation
Openai API Asynchronously Using Script
- Host: GitHub
- URL: https://github.com/dongryeollee1/openai_asynchronous_use
- Owner: DONGRYEOLLEE1
- Created: 2024-02-01T08:20:06.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2024-02-08T02:21:08.000Z (over 2 years ago)
- Last Synced: 2025-02-23T15:15:39.723Z (over 1 year ago)
- Topics: asyncio, batch-processing, openai, openai-api
- Language: Python
- Homepage:
- Size: 127 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Openai_asynchronous_use
Openai API Asynchronously Using Script
- In my case, Implemented an automatic labelling system.
- For training dataset, I processed a huge dataset that approximately 130k.
# Pipeline
# Asynchronous batching test
- In my case, Used a total of 201 data and checked duration by batch size
- There is no guarantee that increasing the batch size further than this will reduce time.
||sequential parse|Asynchronous parse1|Asynchronous parse2|
|---|---|---|---|
|Volume|201|201|201|
|Batch size|-|100|50|
|Duration (sec)|733|**164.9**|478.82|
# Actual processing duration
- To address TPM or RPM errors, data slicing was performed in batches of 50,000 using an asynchronous mechanism for automatic labeling.
- It is recommended to proceed with caution when labeling using APIs, as resource consumption upon encountering errors can be burdensome.
- I set the batch size to `200`.
- For reference, my openai api tier is 4.
||0 ~ 49999|50000 ~ 99999|100000 ~|
|---|---|---|---|
|Number of Errors|2|7|-|
|Duration (sec)|76498.91|68978.05|-|
# Install & Usage
```script
$ pip install -U openai
$ python asynchronous.py
```