CoolFace
Datasetpublic

Goekdeniz-Guelmez/QAT-SFT-Demo

QAT-SFT-Demo Created with MLX-LoRA-Studio · Created with MLX LoRA Studio Overview Repository: Goekdeniz-Guelmez/QAT-SFT-Demo Asset type: synthetic dataset Created at: 2026-06-19 15:38:34 UTC Synthetic data type: SFT Generator model: Qwen3.5-0.8B-bf16 Samples: 100 Estimated tokens: ~17,975 This repository was prepared by MLX LoRA Studio from local training outputs. Dataset Details Generation type: SFT Source dataset:… See the full description on the dataset page: https://huggingface.co/datasets/Goekdeniz-Guelmez/QAT-SFT-Demo.

sourceHugging Faceotherupdated 3mo agoView on Hugging Face
0likes25downloads
Dataset Card

QAT-SFT-Demo

Created with MLX-LoRA-Studio · Created with MLX LoRA Studio

assetupload

Overview

  • —Repository: Goekdeniz-Guelmez/QAT-SFT-Demo
  • —Asset type: synthetic dataset
  • —Created at: 2026-06-19 15:38:34 UTC
  • —Synthetic data type: SFT
  • —Generator model: Qwen3.5-0.8B-bf16
  • —Samples: 100
  • —Estimated tokens: ~17,975

This repository was prepared by MLX LoRA Studio from local training outputs.

Dataset Details

  • —Generation type: SFT
  • —Source dataset: mlx-community/ultrafeedback-prompts-flat-rlhf
  • —DPO generation target: both
  • —Backend: omlx
  • —Generator model: Qwen3.5-0.8B-bf16
  • —Samples: 100
  • —Estimated tokens: ~17,975
  • —Profiled file: output_full.jsonl
  • —Columns: messages

Token count is estimated from text length and is intended as a quick dataset-scale signal, not tokenizer-exact accounting.

Samples

<details><summary>Sample 1</summary>

messages

json
[
  {
    "role": "user",
    "content": "Teacher:You are given a sentence in Persian. Your job is to translate the Farsi sentence into Arabic.\nTeacher: Now, understand the problem? Solve this instance: هر وقت با کسی مصاحبه می کنم ، می تونم بگم ، \"\" به اندازه کافی خوشبو هستی برای این مصاحبه ؟ \"\"\nStudent:"
  },
  {
    "role": "assistant",
    "content": "منظومه را کامل می‌کنم. هر وقت با کسی صحبت کنم، می‌توانم بگویم، «به اندازه کافی خوش‌طبی هستی برای این جلسه»."
  }
]

</details>

<details><summary>Sample 2</summary>

messages

json
[
  {
    "role": "user",
    "content": "Are you familiar with the SAS programming language?"
  },
  {
    "role": "assistant",
    "content": "Yes, I am familiar with **SAS programming**. SAS (Statistical Analysis System) is a powerful, widely used statistical software for data analysis, particularly in business and social sciences. It is known for its reliability, extensive libraries (like PROC, PROC CORR, and beyond), and ability to handle large datasets efficiently.\n\nIf you're looking for specific features, capabilities, or how to use SAS, feel free to ask!"
  }
]

</details>

<details><summary>Sample 3</summary>

messages

json
[
  {
    "role": "user",
    "content": "In this task, you are given a sentence or phrase in English. You must translate it to Xhosa in a way that is equivalent in terms of meaning and grammatically correct.\n\nExample input: Rope and its Usage\nExample output: Intambo nomsebenzi ewenzayo.\nExample explanation: 'Intambo' means rope, 'nomsebenzi ewenzayo' roughly translates to 'the work it does', which is a native way of phrasing it.\nQ: When the passengers are embarked, the man in charge in the raft pulls the tripping line.\nA:"
  },
  {
    "role": "assistant",
    "content": "Ukukhuluka kwamandla kakhulu kakhulu kakhulu kakhulu, ukuba ukuhlanganisa iya-ziyango ye-ramo (yona ukuhlanganisa iya-ziyango) kunye noqiniso (yona ukuhlanganisa iya-ziyango)."
  }
]

</details>

Reproducibility

The full synthetic_spec.json used to launch this run is included in the repository. Re-running the same spec on the same model(s), source dataset, and generation settings should reproduce an equivalent artifact (up to sampling and kernel-level non-determinism).

About

MLX-LoRA-Studio — Created with MLX LoRA Studio

MLX LoRA Studio is a SwiftUI desktop app for fine-tuning open language models on Apple Silicon with the mlx-lm-lora trainer. Curated by MLX-LoRA-Studio.

License

The license of the upstream base model(s), source dataset, generated dataset, and any included tokenizer files applies. Check the source model and dataset cards before redistribution or downstream training.