Duplicate from QuixiAI/ultrachat-uncensored

Co-authored-by: Eric Hartford <ehartford@users.noreply.huggingface.co>

.gitattributes

@@ -0,0 +1,56 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.lz4 filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+# Audio files - uncompressed
+*.pcm filter=lfs diff=lfs merge=lfs -text
+*.sam filter=lfs diff=lfs merge=lfs -text
+*.raw filter=lfs diff=lfs merge=lfs -text
+# Audio files - compressed
+*.aac filter=lfs diff=lfs merge=lfs -text
+*.flac filter=lfs diff=lfs merge=lfs -text
+*.mp3 filter=lfs diff=lfs merge=lfs -text
+*.ogg filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text
+# Image files - uncompressed
+*.bmp filter=lfs diff=lfs merge=lfs -text
+*.gif filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.tiff filter=lfs diff=lfs merge=lfs -text
+# Image files - compressed
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+*.webp filter=lfs diff=lfs merge=lfs -text
+ultrachat_uncensored.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,13 @@
+---
+license: mit
+---
+
+This is based on ultrachat dataset https://huggingface.co/datasets/stingning/ultrachat
+
+I filtered it using the classic "unfiltered" keywords list https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered to remove instances of refusals and bias
+
+About 90% of the dataset was removed.
+
+What remains (400k conversations) is unlikely to inclinate the model to refuse.
+
+I am investigating a less heavy handed approach using dolphin-2.1 to reword any detected refusals.

convertToShareGpt.py

@@ -0,0 +1,42 @@
+import argparse
+import jsonlines
+import json
+import re
+
+from tqdm import tqdm
+
+import re
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--in-file", type=str, default="ultrachat_uncensored_100k.jsonl"
+    )
+    parser.add_argument(
+        "--out-file", type=str, default="ultrachat_uncensored_sharegpt_100k.jsonl"
+    )
+
+    args = parser.parse_args()
+
+    in_file = args.in_file
+    out_file = args.out_file
+
+    with jsonlines.open(in_file) as reader:
+        f = open(out_file, "w", encoding="utf-8")
+
+        for conv in tqdm(reader):
+            conversations = []
+            for i, v in enumerate(conv["data"]):
+                conversations.append(
+                    {"from": "human" if i % 2 == 0 else "gpt", "value": v}
+                )
+
+            json.dump(
+                {
+                    "id": conv["id"],
+                    "conversations": conversations,
+                },
+                f,
+                ensure_ascii=False,
+            )
+            f.write("\n")

remove_refusals.py

@@ -0,0 +1,267 @@
+import argparse
+import jsonlines
+import json
+import re
+
+from tqdm import tqdm
+
+
+def contains_unwanted_words(text):
+    unwanted_words = [
+        "text-based AI language model",
+        "domestic violence",
+        "please refrain",
+        "derogatory",
+        "inappropriate",
+        "offensive",
+        "racism",
+        "racist",
+        "racial",
+        "discriminate",
+        "discriminatory",
+        "discrimination",
+        "sexist",
+        "sexism",
+        "unacceptable",
+        "inclusive workplace",
+        "lgbt",
+        "morals",
+        "ethics",
+        "ethical",
+        "legality",
+        "illegal",
+        "illegality",
+        "hateful",
+        "harmful",
+        "it is never okay",
+        "It is important to",
+        "It's important to",
+        "real-world consequences",
+        "hate speech",
+        "glorify",
+        "not be appropriate",
+        "supremacist",
+        "extremist",
+        "responsible AI",
+        "AI principles",
+        "AI assistant",
+        "an AI language",
+        "ableist",
+        "hurtful",
+        "gender stereotype",
+        "gender inequality",
+        "underrepresentation",
+        "safe spaces",
+        "gender-based",
+        "inclusivity",
+        "feminist",
+        "feminism",
+        "transgender",
+        "empowerment",
+        "communist",
+        "capitalism",
+        "stereotypes",
+        "biases",
+        "bias",
+        "Microaggression",
+        "prioritize human safety",
+        "as a language model",
+        "as an AI language model",
+        "As a large language model",
+        "As an AI",
+        "ethical principles",
+        "consensual",
+        "it is not appropriate",
+        "it's not appropriate",
+        "I cannot fulfill your request",
+        "harmful to human beings",
+        "ethical guidelines",
+        "my guidelines",
+        "prioritize user safety",
+        "adhere to ethical guidelines",
+        "harmful consequences",
+        "potentially harmful",
+        "dangerous activities",
+        "promote safety",
+        "well-being of all users",
+        "responsible information sharing",
+        "jeopardize the safety",
+        "illegal actions or intentions",
+        "undermine the stability",
+        "promote the well-being",
+        "illegal activities or actions",
+        "adherence to the law",
+        "potentially be harmful",
+        "illegal substances or activities",
+        "committed to promoting",
+        "safe information",
+        "lawful information",
+        "cannot provide guidance",
+        "cannot provide information",
+        "unable to offer assistance",
+        "cannot engage in discussions",
+        "programming prohibits",
+        "follow ethical guidelines",
+        "ensure the safety",
+        "involves an illegal subject",
+        "prioritize safety",
+        "illegal subject",
+        "prioritize user well-being",
+        "cannot support or promote",
+        "activities that could harm",
+        "pose a risk to others",
+        "against my programming",
+        "activities that could undermine",
+        "potentially dangerous",
+        "not within the scope",
+        "designed to prioritize safety",
+        "not able to provide",
+        "maintain user safety",
+        "adhere to safety guidelines",
+        "dangerous or harmful",
+        "cannot provide any information",
+        "focus on promoting safety",
+        "an AI language model you don't have",
+        "As an AI language model, I cannot",
+        "As an AI language model, I do not",
+        "As an AI language model, I am not able",
+        "As an AI language model, I don't have personal",
+        "I am an AI language model and do not",
+        "However, it is important to use any code or information provided responsibly and within legal and ethical boundaries.",
+        "As an AI language model, I don't have",
+        "As an AI language model, I am only able",
+        "AI language model and I do not",
+        "As an AI language model, I cannot modify",
+        "As an AI language model, I do not",
+        "I know as an AI language model you don't have",
+        "as an AI language model, you cannot",
+        "I'm sorry, but as an AI language model",
+        "As an AI language model, I don't have",
+        "Unfortunately, I cannot provide",
+        "I'm sorry, I cannot",
+        "I'm sorry, I cannot generate",
+        "AI cannot create or program",
+        "I'm afraid I cannot create",
+        "you cannot create an",
+        "it operates ethically and is",
+        "had an ethical system",
+        "Ensuring the ethical",
+        "and ethical sourcing",
+        "are from ethical",
+        "legal and ethical",
+        "engage in unethical",
+        "unethical or aggressive",
+        "unethical business",
+        "como modelo de lenguaje AI",
+        "Lo siento, como modelo de lenguaje",
+        "no puedo proporcionar",
+        "pero debido a mi capacidad para generar c\u00f3digos complejos y completos es limitado",
+        "Lo siento, pero no puedo",
+        "Lo siento, pero como modelo de lenguaje, no puedo proporcionar",
+        "Lo siento, como modelo de lenguaje, no tengo",
+        "Lo siento, debe haber habido una confusi\u00f3n",
+        "Lo siento, como modelo de lenguaje, no puedo realizar",
+        "Lo siento, soy un modelo de lenguaje y no tengo la capacidad de generar",
+        "Lamento no poder proporcionarte el c\u00f3digo",
+        "Desculpe-me, mas a linguagem vulgar e ofensiva",
+        "apropriada em nenhum contexto",
+        "Como modelo de linguagem",
+        "Como um modelo de linguagem, n\u00e3o tenho a capacidade de",
+        "I cannot assist",
+        "prioritize ethical",
+        "respectful",
+        "morally",
+        "I'm sorry,",
+        "I'm an",
+        "I am an",
+        "I'm an AI",
+        "I am an AI",
+        "my purpose",
+        "filter_bad_language",
+        "filter\_bad\_language",
+        "entertainment purposes",
+        "purely hypothetical",
+        "not a human",
+        "I am an AI",
+        "cannot provide",
+        "can't provide",
+        "won't provide",
+        "not provide",
+        "worth noting",
+        "cause harm",
+        "a language model",
+        "keep in mind",
+        "unethical",
+        "bad language",
+        "the words ****",
+        "bad_language",
+        "certainly not",
+        "complying",
+        "comply",
+        "I cannot",
+        "my main goal",
+        "As a machine",
+        "I don't have the ability",
+        "I am here to assist",
+        "my purpose is to ",
+        "my knowledge cutoff",
+        "my knowledge cut off",
+        "September 2021",
+        "regulations",
+        "not be suitable",
+        "I apologize, but",
+        "It is not possible",
+        "controversial",
+        "my programming",
+        "ethically",
+        "it is important to",
+        "Please note",
+        "sensitive topic",
+        "not acceptable",
+        "It is important for",
+        "divisive",
+        "not appropriate",
+        "our values",
+        "f\*cking",
+        "F\*ck",
+        "sh\*t",
+        "diversity and",
+        "diversity and inclusion",
+        "values diversity",
+        "social responsibility",
+        "environmental, social, and governance",
+        " ESG ",
+        "against women",
+        "problematic history",
+        "diversity",
+        "*This chat conversation is shared from",
+        "*This conversation is shared from",
+    ]
+    for word in unwanted_words:
+        if word.lower() in text.lower():
+            return True
+    return False
+
+
+import re
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--in-file", type=str, required=True)
+    parser.add_argument("--out-file", type=str, default="")
+
+    args = parser.parse_args()
+
+    in_file = args.in_file
+    out_file = args.out_file
+
+    if out_file == "":
+        out_file = "ultrachat_no_refusals.jsonl"
+
+    with jsonlines.open(in_file) as reader:
+        f = open(out_file, "w", encoding="utf-8")
+
+        for conv in tqdm(reader):
+            if not contains_unwanted_words("\n".join(conv["data"])):
+                json.dump(conv, f, ensure_ascii=False)
+                f.write("\n")

ultrachat_uncensored.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a847d725a3c8346c736e74b5e11d21598d65a9be64fe8ed118a063e89510f3bc
+size 2905773314