Qwen3.8-27B-Distillation (40K Traces)

Qwen3.8-27B-Distillation is a dataset containing 40,000 reasoning traces distilled from Qwen's latest model — Qwen3.8-27B. We generated this dataset locally by running the model on our own infrastructure. It covers 4 domains with prompts sourced from 12 diverse open-source datasets.

Dataset Overview

Note: The dataset underwent basic deduplication: removal of duplicate and invalid examples.

Domains

Sources

Prompts were sampled from 12 open-source datasets:

Structure

Each example is a single JSONL line with the following schema:

{
  "messages": [
    {"role": "user", "content": "<prompt>"},
    {"role": "assistant", "content": "<think>...</think>\n\n<response>"}
  ],
  "domain": "code | math | science | logic",
  "category": "python | physics | logic_puzzle | ...",
  "source": "ianncity/Hunter-Alpha-Programming-160000x | ...",
  "tokens_total": 5749,
  "tokens_think": 3200
}

---

Usage

With Hugging Face Datasets

from datasets import load_dataset

dataset = load_dataset("faunix/Qwen3.8-27B-Distillation-40K", split="train")

print(dataset[0]["messages"])   # full chat with <think> + response
print(dataset[0]["domain"])     # code / math / science / logic
print(dataset[0]["tokens_think"])  # thinking tokens count

With Axolotl

datasets:
  - path: faunix/Qwen3.8-27B-Distillation-40K
    type: chat_template
    chat_template: qwen3_5
    split: train

With Unsloth

from unsloth import FastLanguageModel
from datasets import load_dataset

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="unsloth/Qwen3.5-9B",
    max_seq_length=8192,
)

dataset = load_dataset("faunix/Qwen3.8-27B-Distillation-40K", split="train")

---

Sponsor

This dataset would not exist without the help of lium.io! By granting us GPU access, we were able to create this dataset, and we will create more, and train many more interesting models! Therefore, as a sign of gratitude, we mention them here! :)

---

We are on...

Follow us on X:

And also subscribe to our HF organization! This way you won't miss new models and other projects we will be publishing! :)

---

☕ Support Faunix :)

We are two AI developers: Didiblud (12 years old) and Limen4ik (13 years old). Your support is very important to us — just like everyone else, we want to eat, so we are leaving our donation links below!

• 💳 Card / Apple Pay / PayPal: ⭐ Recommended! — app.lava.top/faunix_ai
• 💎 TON: UQArgd4zYOA0I5QSK7ylSlvgyaE5Md8pm-O-RyvSnTTk8Me0
• 💵 USDT (TRC-20): TVCgvfbrUQC1nKD6cHcSP7JGirEEx5f4Te

(Please note: 60% goes into our personal pockets fund, and 40% goes directly to Faunix development — buying domains, GPU hours, etc.)

---

Note

This dataset is open-source and distributed under the Apache 2.0 license, so you can use it in experiments and fine-tuning of your own models! We look forward to seeing what models you can fine-tune based on our dataset! :)

---

Creator: Faunix

Release Date: 22.08.26

Dataset Name: Qwen3.8-27B-Distillation-40K

:)