Dataset Card for "open-instruct"

This dataset is a combination of:

Filtered subset of OpenAssistant/oasst1 train split of Mosaic-dolly-hhrlhf (consists of Databrick's dolly-15k dataset and a filtered subset of Anthropic's HH-RLHF). Filtered subset of conceptofmind/cot_submix_original

  Dataset

The dataset consists of 6 columns:

instruction: The natural language instruction without any prompt templates (we extracted them out of the alpaca-format in… See the full description on the dataset page: https://huggingface.co/datasets/VMware/open-instruct.

NeuralNovel/Open-Instruct-v1 dataset hosted on Hugging Face and contributed by the HF Datasets community

AlignmentLab-AI/open-instruct-sharegpt dataset hosted on Hugging Face and contributed by the HF Datasets community

sgp-bench/open-instruct-gpt4o_55k_rev_sit_72k dataset hosted on Hugging Face and contributed by the HF Datasets community

fireworks-ai/four-digits-multiply-open-instruct dataset hosted on Hugging Face and contributed by the HF Datasets community

sgp-bench/open-instruct dataset hosted on Hugging Face and contributed by the HF Datasets community

Chat-Error/open-instruct-v3 dataset hosted on Hugging Face and contributed by the HF Datasets community

allenai/open_instruct: Rejection Sampling Dataset

See https://github.com/allenai/open-instruct/blob/main/docs/algorithms/rejection_sampling.md for more detail

  Configs

args: {'add_timestamp': False, 'hf_entity': 'jacobmorrison', 'hf_repo_id': 'rejection_sampling_22591', 'hf_repo_id_scores': 'scores_22591', 'input_filename': '/output/shards/22591/29.jsonl', 'max_forward_batch_size': 64, 'mode': 'judgement', 'model_names_or_paths':… See the full description on the dataset page: https://huggingface.co/datasets/jacobmorrison/rejection_sampling_22591.

Open Instruct V1 Better Uncensored

This is the open-instruct-v1 dataset processed with the Better Uncensored pipeline. About 2.5% of the dataset was removed, a quick review of the removed examples seems to point that is mostly false positives or answers with debatable moralizing content. No clear refusal was seen in the quick review. The original dataset may be safe for training uncensored models, but if you want to be extra sure you can use this one.

  Open Instruct V1 -… See the full description on the dataset page: https://huggingface.co/datasets/betteruncensored/open-instruct-v1.

userv4oo/open-instruct-uncensored-refusals-removed dataset hosted on Hugging Face and contributed by the HF Datasets community

Original dataset page from ehartford. 810,102 entries. Sourced from open-instruct-uncensored.jsonl. Converted the jsonl to a json which can be loaded into something like LLaMa-LoRA-Tuner. I've also included smaller datasets that includes less entries depending on how much memory you have to work with. Each one is randomized before being converted, so each dataset is unique in order. Count of each Dataset: code_alpaca: 19991 unnatural_instructions: 68231 baize: 166096 self_instruct: 81512… See the full description on the dataset page: https://huggingface.co/datasets/xzuyn/open-instruct-uncensored-alpaca.

Dataset Card for "open-instruct-v1_deduped"

Deduplicated version of Isotonic/open-instruct-v1 Deduplicated with min Jaccard similarity of 0.8 Uses Stablility's System Prompt

System: StableLM Tuned (Alpha version)

• StableLM is a helpful and harmless open-source AI language model developed by StabilityAI.
• StableLM is excited to be able to help the user, but will refuse to do anything that could be considered harmful to the user.
• StableLM is more than just an information… See the full description on the dataset page: https://huggingface.co/datasets/Isotonic/open-instruct-v1_deduped.

VMware/open-instruct in ChatML format, ready to use in HuggingFace TRL's SFT Trainer. Python code used for conversion: from datasets import load_dataset from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("Felladrin/Llama-160M-Chat-v1")

dataset = load_dataset("VMware/open-instruct", split="train")

def format(columns): messages = [ { "role": "user", "content": columns["instruction"].strip(), }, {… See the full description on the dataset page: https://huggingface.co/datasets/Felladrin/ChatML-open-instruct.

Dataset Card for "open-instruct-uncensored-alpaca"

More Information needed

allenai/open_instruct: Generation Dataset

See https://github.com/allenai/open-instruct/blob/main/docs/algorithms/rejection_sampling.md for more detail

  Configs

args: {'add_timestamp': True, 'dataset_end_idx': 3239, 'dataset_mixer_list': ['VGraf/alpacaeval_paraphrase_questions_dev', '1.0'], 'dataset_splits': ['train', 'train'], 'dataset_start_idx': 0, 'hf_entity': 'VGraf', 'hf_repo_id': 'generation', 'mode': 'generation', 'model_name_or_path': 'gpt-3.5-turbo-0125'… See the full description on the dataset page: https://huggingface.co/datasets/VGraf/generation_1746311715.

from datasets import load_dataset, Dataset import re

Script to filter and process the OpenAssistant dataset (oasst2).

Based on the conversion script from the open-instruct repo -> https://github.com/allenai/open-instruct/blob/main/scripts/data/sft/utils.py#L1

def should_be_filtered_by_keyword(example, verbose=False): # we filter out conversations that contain some specific strings filter_strings = [ "OpenAI", "Open AI", "ChatGPT", "Chat GPT"… See the full description on the dataset page: https://huggingface.co/datasets/PRLM/oasst2.

allenai/open_instruct: Generation Dataset

See https://github.com/allenai/open-instruct/blob/main/docs/algorithms/rejection_sampling.md for more detail

  Configs

args: {'add_timestamp': False, 'hf_entity': 'jacobmorrison', 'hf_repo_id': 'generation_19367', 'mode': 'generation', 'model_name_or_path': '/model', 'push_to_hub': True, 'revision': 'main', 'save_filename': '/output/shards/19367/14.jsonl', 'skill': 'chat'}

dataset_args: {'dataset_end_idx': 29850… See the full description on the dataset page: https://huggingface.co/datasets/jacobmorrison/generation_19367.

allenai/open_instruct: Rejection Sampling Dataset

See https://github.com/allenai/open-instruct/blob/main/docs/algorithms/rejection_sampling.md for more detail

  Configs

args: {'add_timestamp': False, 'hf_entity': 'jacobmorrison', 'hf_repo_id': 'rejection_sampling_5816', 'hf_repo_id_scores': 'scores_5816', 'input_filename': '/output/shards/5816/4.jsonl', 'max_forward_batch_size': 64, 'mode': 'judgement', 'model_names_or_paths': ['Skywork/Skywork-Reward-Llama-3.1-8B']… See the full description on the dataset page: https://huggingface.co/datasets/jacobmorrison/scores_5816.

allenai/open_instruct: Rejection Sampling Dataset

See https://github.com/allenai/open-instruct/blob/main/docs/algorithms/rejection_sampling.md for more detail

  Configs

args: {'add_timestamp': False, 'hf_entity': 'vwxyzjn', 'hf_repo_id': 'rejection_sampling_26764', 'hf_repo_id_scores': 'scores_26764', 'input_filename': 'output/shards/26764/3.jsonl', 'max_forward_batch_size': 64, 'mode': 'judgement', 'model_names_or_paths': ['allenai/llama-3-tulu-2-8b-uf-mean-rm']… See the full description on the dataset page: https://huggingface.co/datasets/vwxyzjn/scores_26764.

allenai/open_instruct: Generation Dataset

See https://github.com/allenai/open-instruct/blob/main/docs/algorithms/rejection_sampling.md for more detail

  Configs

args: {'add_timestamp': False, 'hf_entity': 'vwxyzjn', 'hf_repo_id': 'generation_6296', 'mode': 'generation', 'model_name_or_path': 'allenai/open_instruct_dev', 'push_to_hub': True, 'revision': 'costa_finetune_tulu3_8b_norobot_meta-llama_Meta-Llama-3.1-8B_42_1725559869', 'save_filename':… See the full description on the dataset page: https://huggingface.co/datasets/vwxyzjn/generation_6296.