Three real reports
Each one went through the normal finetuna pipeline: JSONL in, QLoRA on an RTX 4090, GGUF and report.html out. Nothing is cut or touched up. The datasets are public, so you can reproduce any of them for $5.
Polish review sentiment on Bielik 11B
A Polish review in, one word out: pozytywna, negatywna, neutralna or niejednoznaczna.
- Model
- Bielik 11B v3 Instruct
- Dataset
- PolEmo 2.0 (clarin-pl, via allegro/klej-polemo2-in), CC BY-SA 4.0
- Size
- 2,000 lines, 1 epoch
- Result
- still training, the report appears here on delivery
- What to look for
- The base model explains itself in a paragraph. The tuned one answers with the label, and gets the ambiguous reviews right more often.
JSON entity extraction on Qwen3 8B
A sentence in, one JSON object out with eight fixed keys (person, organization, location, event, product, art, building, other).
- Model
- Qwen3 8B
- Dataset
- Few-NERD, supervised split (DFKI-SLT), CC BY-SA 4.0
- Size
- 3,000 lines, 2 epochs
- Result
- still training, the report appears here on delivery
- What to look for
- The base model adds thinking, prose or extra keys. The tuned one returns the schema every time, so json.loads never fails.
Yoda style rewrite on Llama 3.1 8B
A plain sentence in, the same sentence the way Yoda says it out. Nothing else.
- Model
- Llama 3.1 8B Instruct
- Dataset
- yoda_sentences (dvgodoy), Apache 2.0
- Size
- 720 lines, 3 epochs
- Result
- still training, the report appears here on delivery
- What to look for
- 720 lines is a small dataset. Watch whether the tuned model drops the preamble and keeps the inversion pattern consistent.
How to reproduce one
The script that builds these three JSONL files from the source datasets is in the repository as demos/prepare.py. Upload its output on the New job page, pick the same model and epochs. Put the system prompt from the file's first line in the system prompt field so the Modelfile carries it.