close stringlengths 1 1.25k |
|---|
close |
Google Account |
reethu |
reethu0001@gmail.com |
. |
close |
ename notebook |
SAWIT_AI_Hackathon25.ipynb_ |
startar notebook in Google Drive |
loud_donell changes saved |
ile |
dit |
iew |
nsert |
untime |
ools |
elp |
commentpen comments pane |
settingspen settings |
eoplehare hare notebook |
ormat_list_bulleted |
earch |
pn_key |
older |
ode |
erminal |
ddnsert code cell below trl+M Bnsert code cell below (Ctrl+M B) |
ode |
dddd text cell dd text cell |
ext |
loud_donell changes saved |
oneocus the last run cell |
onnected toPython 3 Google Compute Engine backend (GPU)RAM: 3.02 GB/12.67 GBDisk: 38.00 GB/112.64 GB onnected to |
ython 3 Google Compute Engine backend (GPU) |
AM: 3.02 GB/12.67 GB |
isk: 38.00 GB/112.64 GB |
4 |
AM |
isk |
arrow_drop_downdditional connection options |
emini |
eople |
settings |
expand_lessexpand_moreoggle header visibility |
otebook |
more_horiz |
sparkemini |
keyboard_arrow_down |
SAWiT AI Hackathon: Colloquial Dataset Creation and Model Training |
Welcome to the SAWiT AI Hackathon! In this hackathon, you'll be creating a colloquial language dataset for one of the following languages: |
Marathi |
Tamil |
Hindi |
Telugu |
Malayalam |
Bengali |
Task Overview |
1. Pick a Language: Choose one language from the list above. |
2. Create a Colloquial Dataset: Collect and curate a dataset that represents the colloquial language of your selected language. This could include informal conversations, slang, and regional variations. |
3. Train a Model: Use your dataset to train a machine learning model that understands or processes the colloquial language. You can use existing models and fine-tune them with your dataset. |
4. Push the Dataset & Model to Hugging Face: Once the model is trained, push both your dataset and model to Hugging Face. You will need to create a Hugging Face account if you don't have one already. |
5. Share the Final Links for Evaluation: After pushing the dataset and model to Hugging Face, share the final links for evaluation in Hackathon Platform. |
Helpful Resources |
Unsloth: An optimization framework for fine-tuning Large Language Models (LLMs) that makes training 2-4x faster. It provides optimized implementations of common operations like LoRA (Low-Rank Adaptation) training, specialized kernels for faster computation, and memory-efficient training methods. The framework integrate... |
Hugging Face: A platform for hosting and sharing your datasets and models. You can push your trained models and datasets to Hugging Face and share the link for evaluation. |
Steps Overview: |
1. Dataset Creation: Use Unsloth or similar approach for dataset creation (link above). |
2. Model Training: Fine-tune or train a model using your dataset. |
3. Push to Hugging Face: Upload your model and dataset to Hugging Face. |
4. Evaluation: Share your Hugging Face model URL for evaluation. |
Example Output |
`` |
Input to Translate: "What is data structure?" |
Output Expected (Tamil): "Data structure na yenna?" |
`` |
Good Luck! |
We are excited to see your contributions! Happy coding and training! 🚀 |
Below is an example that demonstrates the process of sample dataset creation, model training using Unsloth, and uploading it to Hugging Face |
] |
subdirectory_arrow_right 9 cells hidden |
add ode |
add ext |
spark emini |
1] un cell (Ctrl+Enter) |
ell executed since last change |
xecuted by reethu |
1:51 AM (50 minutes ago) |
xecuted in 259.605s |
m |
# Install required packages |
!pip install torch==2.5.1 |
!pip install transformers datasets accelerate bitsandbytes |
!pip install unsloth |
!pip install peft |
import torch |
from datasets import Dataset |
from unsloth import FastLanguageModel |
import pandas as pd |
from datetime import datetime |
from transformers import TrainingArguments, Trainer |
End of preview. Expand in Data Studio
No dataset card yet
- Downloads last month
- -