Dataset · r0b0tlab
49,772 Qwen3.8-Max distillation traces
r0b0tlab/qwen3.8-max-distillation-50k
About 49,772 traces from qwen3.8-max-preview for SFT and off-policy distillation.
What's inside
A set of 49,772 traces generated by the qwen3.8-max-preview teacher model for supervised fine-tuning and off-policy knowledge distillation. Answers are saved as returned by the API; visible <think>...</think> blocks remain in the assistant message, while some simple requests received answers without reasoning. The language is English, and modalities per metadata are text and tabular. The card warns about uncleaned conditions and data provenance.
What you can do with it
- SFT of a compact model on qwen3.8-max-preview traces
- Knowledge distillation into a smaller LLM
- Evaluating the model on QA and instructions with reasoning
- Tasks
- Text generation, Question answering
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 782
- Likes
- 123
Models trained on it
- Ma7ee7/Qwen3.8_4B_Distilled_GGUFText generation 103.5K
- detrax/Qwen3-4B-Thinking-2507-Qwen3.8-Max-Distillation-DetraxText generation 1.4K
- mradermacher/Satella-30B-A3B-i1-GGUF 847
- mradermacher/Satella-30B-A3B-GGUF 817
- Atomic-Germ/Qwen3.8-Distilled-4B-NPU2Image text → text 747
- mradermacher/Satella-0.8B-GGUF 539
- Atomic-Germ/Qwen3.8-Distilled-1.2B-NPU2Text generation 253
- Ma7ee7/Qwen3.8_4B_DistilledText generation 188
How to load it
from datasets import load_dataset
ds = load_dataset("r0b0tlab/qwen3.8-max-distillation-50k")