Teacher capture
MiMo V2.6 Distill Qwen 9B generated one web-development completion for each selected task on Kaggle 2×T4.
Public research artifact · Separate release · Chunk 00
MiMo V2.6 Distillation
Webdev Explorer
Explore the first public filtered chunk of an independent knowledge-distillation experiment using MiMo V2.6 Distill Qwen 9B as teacher data for smaller local code-capable students.
…complete examples
0.8934mean local score
MiMo 9Bteacher model
Chunk 00separate release
Loading the public dataset from Hugging Face…
Explore the dataset
Select a record to inspect its original prompt, provenance, score, generated HTML and a sandboxed browser preview. Use the buttons to copy the prompt or download the complete filtered chunk.
Sandboxed HTML preview
The preview is untrusted generated HTML in a sandbox. It is for inspection only.
How the dataset was created
Quality filtering
Incomplete outputs were removed and the remaining HTML documents were checked with a local non-rendering grader.
Student research
The filtered records are intended for supervised fine-tuning and QLoRA experiments with smaller models.
Reproduce and reuse
Google Colab
Load the dataset, inspect rows and calculate basic statistics in a browser notebook.
Open notebook ↗Hugging Face
Use the Dataset Viewer or load it with the `datasets` library.
Open dataset ↗GitHub lab
Read the manifests, grading artifacts and local training notes.
Open repository ↗Citation and provenance
This is an independent community experiment, not an official Xiaomi dataset or model release. Please preserve the chunk ID and link to the canonical HF dataset when reusing the data.
MiMo V2.6 Distillation Lab (2026). MiMo V2.6 Distill Qwen 9B Webdev — Chunk 00. https://huggingface.co/datasets/mindchain/mimo-v2.6-distill-qwen9b-webdev-chunk00
Frequently asked questions
Is this an official Xiaomi dataset?
No. It is an independent community research artifact using the MiMo V2.6 Distill Qwen 9B teacher.
Can I use it for fine-tuning?
Yes, subject to the applicable source-model and dataset terms. Review generated code and licenses before production use.
Where are the raw and filtered files?
The filtered records are on the HF Dataset; provenance and tooling are documented in the GitHub lab.