Upload folder using huggingface_hub

.gitattributes

@@ -58,3 +58,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+historica_instruct.jsonl filter=lfs diff=lfs merge=lfs -text
+instruct_mix.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,108 @@
+---
+language:
+  - la
+  - grc
+  - en
+  - he
+  - cop
+  - non
+  - ang
+  - enm
+  - got
+  - cu
+  - xcl
+  - de
+  - fr
+  - it
+  - es
+  - ru
+license: cc-by-sa-4.0
+task_categories:
+  - text-generation
+  - translation
+tags:
+  - instruction-tuning
+  - ancient-languages
+  - historical-languages
+  - translation
+  - multilingual
+  - latin
+  - ancient-greek
+  - coptic
+  - old-norse
+  - old-english
+  - gothic
+size_categories:
+  - 10K<n<100K
+---
+
+# historica-instruct
+
+instruction-response pairs for fine-tuning language models on ancient and historical language tasks. 50k examples with multilingual instructions in 8 languages.
+
+## overview
+
+| | count |
+|---|---:|
+| total examples | 50,000 |
+| translation instructions | 25,000 (50%) |
+| continuation instructions | 25,000 (50%) |
+| source languages | 15 |
+| instruction languages | 8 |
+
+## composition
+
+**translation instructions (50%):** source text in an ancient language paired with its translation. the instruction asks the model to translate, in one of 8 instruction languages.
+
+**continuation instructions (50%):** a passage is split into prefix (instruction) and suffix (response). the instruction asks the model to continue writing in the source language.
+
+**church content deprioritized:** christian/patristic texts sampled at 15-30% of their natural rate. pagan mythology, secular literature, law, philosophy, poetry are proportionally overrepresented.
+
+## schema
+
+| column | type | description |
+|---|---|---|
+| `instruction` | string | the instruction prompt (in one of 8 languages) |
+| `response` | string | the expected response |
+| `source` | string | `historica_translate` or `historica_continue` |
+| `src_lang` | string | source text language (ISO 639) |
+| `tgt_lang` | string | target/response language |
+| `instr_lang` | string | language of the instruction itself |
+
+## source languages
+
+| language | code | examples |
+|---|---|---:|
+| Latin | lat | 23,136 |
+| Ancient Greek | grc | 11,699 |
+| English | eng | 3,646 |
+| Middle English | enm | 2,598 |
+| Coptic | cop | 2,517 |
+| Hebrew | heb | 2,435 |
+| Old Norse | non | 2,229 |
+| Old English | ang | 1,499 |
+| Gothic | got | 58 |
+| + others | deu, nob, xcl, fra, sme, chu | < 100 each |
+
+## instruction languages
+
+instructions are given in 8 languages with correct grammar and declension:
+
+| language | examples |
+|---|---:|
+| English | ~17k |
+| Russian | ~8k |
+| Italian | ~4k |
+| French | ~4k |
+| Latin | ~4k |
+| Modern Greek | ~4k |
+| Spanish | ~4k |
+| German | ~4k |
+
+## data sources
+
+built from [sol-r/historica-pairs](https://huggingface.co/datasets/sol-r/historica-pairs) (translation) and [sol-r/historica-corpus](https://huggingface.co/datasets/sol-r/historica-corpus) (continuation).
+
+## intended use
+
+instruction fine-tuning of encoder-decoder or decoder-only models for ancient language tasks: translation, text continuation, analysis. designed to be mixed with general instruction datasets (e.g. OpenHermes, tulu-3) for step 1.5 fine-tuning.

data/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c73f969c86257a43c11e9a55101feb92238e38183de7c07f81f42569b222c39c
+size 76314111

historica_instruct.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b397b1fb824b820007590abaf77fdf5f346bf03b7aa6cae6a94d71cc4b0a026
+size 119544424

instruct_mix.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5684916d578c2c6637da48a256d4195824ed9c7580e1a35391953282aebb1c47
+size 159346982