Datasets:
Dataset card for 20261102-1437-e4b-k-canon
Browse files
README.md
ADDED
|
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- ca
|
| 4 |
+
license: other
|
| 5 |
+
task_categories:
|
| 6 |
+
- text-generation
|
| 7 |
+
tags:
|
| 8 |
+
- andorra
|
| 9 |
+
- catalan
|
| 10 |
+
- sft
|
| 11 |
+
- maia
|
| 12 |
+
pretty_name: maia training data
|
| 13 |
+
size_categories:
|
| 14 |
+
- 10K<n<100K
|
| 15 |
+
configs:
|
| 16 |
+
- config_name: default
|
| 17 |
+
data_files:
|
| 18 |
+
- split: train
|
| 19 |
+
path: train.jsonl
|
| 20 |
+
- split: validation
|
| 21 |
+
path: valid.jsonl
|
| 22 |
+
---
|
| 23 |
+
|
| 24 |
+
# maia training data
|
| 25 |
+
|
| 26 |
+
Conversations in Andorran Catalan used to fine-tune **maia**, a Gemma 4 model specialised in
|
| 27 |
+
Andorra. This version is the exact data of training run **`20261102-1437-e4b-k-canon`** (variant `K-canon`).
|
| 28 |
+
|
| 29 |
+
| Split | Conversations |
|
| 30 |
+
|---|---:|
|
| 31 |
+
| train | 28679 |
|
| 32 |
+
| validation | 732 |
|
| 33 |
+
|
| 34 |
+
Each row is `{"messages": [...]}` in chat format. Every conversation starts with the same
|
| 35 |
+
system message, which the model also expects at inference time:
|
| 36 |
+
|
| 37 |
+
> Ets maia, un assistent que coneix a fons Andorra: la seva història, les institucions, la geografia, la cultura i la gent. Respons en català, amb naturalitat i amb la parla d'Andorra.
|
| 38 |
+
|
| 39 |
+
## What is in it
|
| 40 |
+
|
| 41 |
+
- Knowledge conversations written from curated articles about Andorra (history, institutions,
|
| 42 |
+
territory, society, culture, economy, sports, language).
|
| 43 |
+
- An **augmented canon of facts**: verified atomic facts, each asked about in about 11 different
|
| 44 |
+
ways (direct, reverse, paraphrase, context, two-turn, false premise, abstention), plus
|
| 45 |
+
contrast examples between facts that are easy to confuse.
|
| 46 |
+
- `probe.jsonl`: 338 held-out questions (never in train) with the keywords a correct answer
|
| 47 |
+
must contain. They measure fact recall before and after training.
|
| 48 |
+
|
| 49 |
+
Questions of the maia-bench benchmark and the articles it reserves are excluded.
|
| 50 |
+
|
| 51 |
+
## Sources and licence
|
| 52 |
+
|
| 53 |
+
Every source has a provenance record (owner, licence, terms, redistribution status) in the maia
|
| 54 |
+
repository (`docs/fonts/`). Redistribution is decided by the project owner per source; see the
|
| 55 |
+
records before reusing a portion of the data. Code: [github.com/ericrisco/maia](https://github.com/ericrisco/maia).
|