datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Vicuna-Uncleaned-Alpaca-Format100K_deduplicated_ner_indexes_name_country_alpaca_format_json_response_all_casesDolphin1m_gpt4_Alpaca_formatGorilla_Alpaca_FormatThis is the dataset to the model used to train gorilla 7b but in the alpaca format, for lora training.
Thank you to microsoft and uc berkly for open sourcing these datasets. As of now I do not believe this dataset works, will have to do more testing, but gorilla team plans to realease training code which might make it easer to see how this was fully done. and how it can be done with lora.
For ALPACA LORA users:
Modules you can target with lora:"gate_proj", "down_proj", "up_proj", "q_proj"… See the full description on the dataset page: https://huggingface.co/datasets/Vezora/Gorilla_Alpaca_Format.Mini_Orca_Code_Uncencored_alpaca_FormatThis is dataset is a modified version of "psmathur's" Mini orca dataset, formated in the alpaca format and uncencored.
This dataset is filtered to only feature coding instructions around 50k code examples.
For ALPACA LORA users:
Modules you can target with lora:"gate_proj", "down_proj", "up_proj", "q_proj", "v_proj", "k_proj", "o_proj"
Most lora models use:"q_proj", "v_proj", "k_proj", "o_proj"
Platypus which got terrific results: "gate_proj", "down_proj", "up_proj"
Research on targeting… See the full description on the dataset page: https://huggingface.co/datasets/Vezora/Mini_Orca_Code_Uncencored_alpaca_Format.ShareGPT-unfiltered-alpaca-lora-formatgsm8k_reasoning_paths_deepseek_alpaca_formatIntentGuard-2-alpaca-format
IntentGuard 2 Alpaca Format
This dataset is IntentGuard-2 converted to Alpaca instruction format.
Every row uses instruction: "/intentguard".
Source Files
IntentGuard-2/train.jsonl
IntentGuard-2/test.jsonl
Output Files
train.json
test.json
Both output files are JSON arrays. Each item contains instruction, input, and output.
Rebuild
Run python3 scripts/convert_to_alpaca.py from this directory.
fibonacci_alpaca_to_sharegpt_gpt_format_convert_new_dataset_release50K_deduplicated_ner_indexes_name_country_alpaca_format_json_responsenovel17_train_alpaca_formatCredit: AlexanderDoria/novel17_test
100K-ner-indexes-multiple-organizations-locations-alpaca-format-json-response-all-casesgsm8k_reasoning_paths_test_deepseek_alpaca_formatPersian_Text_Dataset_QA_Alpaca_Format_Conversations5HC3-Chinese-AlpacaFormatGSM8K_ALPACA_FORMAT500K-ner-indexes-multiple-organizations-locations-alpaca-format-json-response-all-casestaxonomies-dataset-alpaca-prompt-formatingredients-recipe-alpaca-formatdolly-and-alpaca-lora-data-formatfibonacci_alpaca_to_gemma_format_dataset_Fibonacci_ai_persian_python50K_ner_indexes_name_country_alpaca_formatalpacaformatGKChestertonEZYBill_SQL_alpaca_formatIntentGuard-1-alpaca-formatThis is the IntentGuard-1 dataset converted to alpaca instruction format.
mental-alpaca-formatAlpaca_format_datasetnews_seniment_gpt_alpacaformatThis dataset is a alpaca formatted version of "oliverwang15/news_with_gpt_instructions" (https://huggingface.co/datasets/oliverwang15/news_with_gpt_instructions)
20k examples of grading senitment using gpt (unclear which model) (used to train fingptv3).
For ALPACA LORA users:
Modules you can target with lora:"gate_proj", "down_proj", "up_proj", "q_proj", "v_proj", "k_proj", "o_proj"
Most lora models use:"q_proj", "v_proj", "k_proj", "o_proj"
Platypus which got terrific results: "gate_proj"… See the full description on the dataset page: https://huggingface.co/datasets/Vezora/news_seniment_gpt_alpacaformat.autorater-alpaca-format-50kalpaca_flan-format