metadata

license: mit
datasets:
  - gbharti/finance-alpaca
  - lavita/ChatDoctor-HealthCareMagic-100k
  - laion/OIG
  - openai/webgpt_comparisons
  - taskydata/GPT4Tools
  - DataProvenanceInitiative/cot_submix_original
  - 0x70DA/stackoverflow-chat-data
language:
  - en
library_name: adapter-transformers
pipeline_tag: text-classification

Attempt to reproduce Mixture-of-LoRAs classifier

Mixture-of-LoRAs: An Efficient Multitask Tuning for Large Language Models

https://arxiv.org/pdf/2403.03432

Datasets

We evenly sample about 10k training data and 2k validation data on each dataset.

From laion/OIG was taken only:

unified_merged_code_xp3.jsonl
unified_grade_school_math_instructions.jsonl
unified_mathqa_flanv2_kojma_cot.jsonl