Model · dictation post-processing
Gemma 3 270M Email Key-Line Bolding: bold the key lines of a dictated email
Puts the key parts of a dictated email in Markdown bold and changes nothing else. Runs after the paragraph-breaks model.
About the model
The input is an email that has already been paragraphed. The output is the same text with ** around the requests, the purpose, the main points and the key facts, so a reader who skims only the bold knows what is being asked.
It must not change anything else. Every call is checked: if anything other than ** was added, the input is returned unbolded.
Uses. In a dictation pipeline, it bolds the key lines of a dictated email after paragraphing. For outbound mail, an AI agent drafting an email can bold the details so the recipient sees them at a glance; it pairs with the paragraph-breaks model. For inbound mail, it can annotate received emails, and a mail client could show the spans as highlights instead of bold.
This is a first, preliminary version. The bolding style is mine, and most of the training labels were extrapolated from my annotations by an LLM.
Which file should I download?
safetensors (Transformers) 536 MB
Python on CPU or GPU, at full precision. Use the system prompt and validation check from the model card.
Run with Hugging Face Transformers Download safetensors (Transformers) →Results
r2, run email-bold-r2-gemma270m-e10, evaluated 21 October 2026. Scored on 6 held-out emails bolded by hand, per word: a word counts as bold if it falls inside a span. Higher is better. A fallback is an output that changed anything other than adding **, so the input is returned unbolded.
| Test set | Untrained Gemma 3 270M | Fine-tune r2 |
|---|---|---|
| Word precision | - | 0.73 |
| Word recall | - | 0.54 |
| Word F1 | - | 0.62 |
| Fallbacks (fewer is better) | 6 of 6 | 0 of 6 |
- The untrained model fell back on every email, so it bolded nothing.
- Reference: Claude Sonnet 5.5, prompted with the written rules, scores word F1 0.69 (4-fold cross-validation over the 19 hand-annotated emails).
- The model is precise but under-bolds. It tends to bold the ask and the purpose line and miss some of the listed points.
- Six test emails is a small test set. Treat the numbers as rough.
Training
- Full fine-tune of google/gemma-3-270m-it (270M parameters): run email-bold-r2-gemma270m-e10, 10 epochs, learning rate 5e-05, one NVIDIA L4 on Modal, 504 seconds.
- 336 synthetic dictated emails written by Claude Haiku 5.5, to 12 kinds of recipient. 13 were bolded by hand; the other 323 were labelled by Claude Sonnet 5.5 from rules derived from 19 hand annotations (silver labels). The median email has about 25% of its words in bold.
- Shorter runs (3 and 6 epochs, and 6 epochs at learning rate 1e-04) scored word F1 0.39 to 0.56. F1 was still rising at 10 epochs.
Training data
Checkpoints and roadmap
Where the model is now and the checkpoints still to come. Planned stages are the current plan, not promises.
- r2 Done 21 October 2026
Preliminary release: 336 emails, 10 epochs. Word F1 0.62.
- More hand annotations In progress
Annotation continues, towards about 200 hand-bolded emails, to replace the preliminary model.
Limitations
- Bolding is subjective, and the labels reflect one person's preference.
- Mostly silver labels, so the model learns the LLM labeller's habits as well as mine.
- English only, and trained on synthetic dictated emails. Not tested on real transcripts or on text other than email.