Review and verify outputs carefully, especially for critical applications. Expert validation is recommended to avoid biased or inconsistent content. Use responsibly and ethically, complying with applicable laws and regulations to prevent misuse for malicious purposes.

Training Data

The model was fine-tuned on a proprietary dataset from OpenVoid, featuring high-quality text data related to coding, cybersecurity, and hacking. Extensive filtering and preprocessing ensured data quality and relevance.

Evaluation

HumanEval v1.0: pass@1: 0.537
EvalPlus v1.1: pass@1: 0.482

How to Use the Model

Using Transformers

Example of using Prox-Llama-3-8B with the Transformers library:

python

import transformers
import torch

model_id = "openvoid/Prox-Llama-3-8B-abliterated-orpo"

pipeline = transformers.pipeline(
    "text-generation",
    model=model_id,
    model_kwargs={"torch_dtype": torch.bfloat16},
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are Prox."},
    {"role": "user", "content": "Who are you?"},
]

terminators = [
    pipeline.tokenizer.eos_token_id,
    pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

outputs = pipeline(
    messages,
    max_new_tokens=256,
    eos_token_id=terminators,
    do_sample=True,
    temperature=0.6,
    top_p=0.9,
)
print(outputs[0]["generated_text"][-1])