CoolFace
Datasetpublic

emirhanboge/sst2_mnli_qqp_llama1b_modified

Multi-Task Dataset: SST-2 + MNLI + QQP (Modified for LLaMA 1B) This dataset is a combination of SST-2, MNLI, and QQP for multi-task learning. It is preprocessed and tokenized specifically for training with the LLaMA-1B model. Modifications: Each example includes a task prefix: SST-2: "Task: SST2 | Sentence: ..." MNLI: "Task: MNLI | Premise: ... Hypothesis: ..." QQP: "Task: QQP | Q1: ... Q2: ..." Labels are standardized to integer format. Tokenized using the… See the full description on the dataset page: https://huggingface.co/datasets/emirhanboge/sst2_mnli_qqp_llama1b_modified.

sourceHugging Faceupdated 2y agoView on Hugging Face
0likes18downloads
Dataset Card

Multi-Task Dataset: SST-2 + MNLI + QQP (Modified for LLaMA 1B)

This dataset is a combination of SST-2, MNLI, and QQP for multi-task learning.

It is preprocessed and tokenized specifically for training with the LLaMA-1B model.

Modifications:

  • —Each example includes a task prefix:
  • —SST-2: "Task: SST2 | Sentence: ..."
  • —MNLI: "Task: MNLI | Premise: ... Hypothesis: ..."
  • —QQP: "Task: QQP | Q1: ... Q2: ..."
  • —Labels are standardized to integer format.
  • —Tokenized using the LLaMA-1B tokenizer.
  • —Maximum sequence length is 128 tokens.

Dataset Usage:

python
from datasets import load_dataset
dataset = load_dataset("emirhanboge/sst2_mnli_qqp_llama1b_modified")