Solshine's picture
Update README.md
d79840c verified
|
raw
history blame
958 Bytes
metadata
dataset_info:
  features:
    - name: input_ids_1
      sequence: int64
    - name: attention_mask_1
      sequence: int64
    - name: labels_1
      sequence: int64
    - name: input_ids_2
      sequence: int64
    - name: attention_mask_2
      sequence: int64
    - name: labels_2
      sequence: int64
  splits:
    - name: train
      num_bytes: 145200
      num_examples: 50
  download_size: 25156
  dataset_size: 145200
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

From Differential Adaptive Merging: Created via:

!python dam/data/create_merge_dataset.py \
  --dataset_names "tatsu-lab/alpaca,microsoft/orca-math-word-problems-200k" \
  --model_ids "meta-llama/Llama-3.2-1B,meta-llama/Llama-3.2-1B-Instruct" \
  --base_model_name "meta-llama/Llama-3.2-1B" \
  --cache_dir "/content/cache" \
  --compute_logits False \
  --dataset_id "your_dataset_id" \
  --example_count 50 \
  --max_length 60 \
  --add_top_k_logits False