Initial private draft of the Samdei Type release
Browse files- LICENSE-chhunneng.txt +21 -0
- README.md +204 -0
- abbreviated/ft_v2_abbr3.parquet +3 -0
- abbreviated/train_gemrom2_abbr3.parquet +3 -0
- casual/ft_casual_noakk.parquet +3 -0
- generator/ft_v2.parquet +3 -0
- llm_romanized/train_gemrom2.parquet +3 -0
- mined_tables/ft_rules_chh.parquet +3 -0
- mined_tables/ft_rules_chhwd.parquet +3 -0
- recipes.json +212 -0
- validation/train_v2_s3.dev.parquet +3 -0
LICENSE-chhunneng.txt
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
MIT License
|
| 2 |
+
|
| 3 |
+
Copyright (c) 2025 Chrea Chanchhunneng
|
| 4 |
+
|
| 5 |
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
| 6 |
+
of this software and associated documentation files (the "Software"), to deal
|
| 7 |
+
in the Software without restriction, including without limitation the rights
|
| 8 |
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
| 9 |
+
copies of the Software, and to permit persons to whom the Software is
|
| 10 |
+
furnished to do so, subject to the following conditions:
|
| 11 |
+
|
| 12 |
+
The above copyright notice and this permission notice shall be included in all
|
| 13 |
+
copies or substantial portions of the Software.
|
| 14 |
+
|
| 15 |
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
| 16 |
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
| 17 |
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
| 18 |
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
| 19 |
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
| 20 |
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
| 21 |
+
SOFTWARE.
|
README.md
ADDED
|
@@ -0,0 +1,204 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
pretty_name: Samdei Type Fine-tuning Data
|
| 3 |
+
license: cc-by-sa-4.0
|
| 4 |
+
language:
|
| 5 |
+
- km
|
| 6 |
+
task_categories:
|
| 7 |
+
- text2text-generation
|
| 8 |
+
tags:
|
| 9 |
+
- transliteration
|
| 10 |
+
- romanization
|
| 11 |
+
- romanized-khmer
|
| 12 |
+
- khmer
|
| 13 |
+
- synthetic
|
| 14 |
+
- chat
|
| 15 |
+
size_categories:
|
| 16 |
+
- 1M<n<10M
|
| 17 |
+
multilinguality: monolingual
|
| 18 |
+
source_datasets:
|
| 19 |
+
- extended|HPLT/HPLT2.0_cleaned
|
| 20 |
+
- extended|HuggingFaceFW/fineweb-2
|
| 21 |
+
- extended|wikimedia/wikipedia
|
| 22 |
+
- extended # + ALT (Asian Language Treebank, NICT), Chhunneng word list, Wikidata names
|
| 23 |
+
annotations_creators:
|
| 24 |
+
- machine-generated
|
| 25 |
+
language_creators:
|
| 26 |
+
- found
|
| 27 |
+
- machine-generated
|
| 28 |
+
configs:
|
| 29 |
+
- config_name: chat_mix
|
| 30 |
+
default: true
|
| 31 |
+
data_files:
|
| 32 |
+
- split: train
|
| 33 |
+
path: generator/ft_v2.parquet
|
| 34 |
+
- config_name: llm_romanized
|
| 35 |
+
data_files:
|
| 36 |
+
- split: train
|
| 37 |
+
path: llm_romanized/train_gemrom2.parquet
|
| 38 |
+
- config_name: abbreviated
|
| 39 |
+
data_files:
|
| 40 |
+
- split: train
|
| 41 |
+
path:
|
| 42 |
+
- abbreviated/ft_v2_abbr3.parquet
|
| 43 |
+
- abbreviated/train_gemrom2_abbr3.parquet
|
| 44 |
+
- config_name: mined_tables_chh
|
| 45 |
+
data_files:
|
| 46 |
+
- split: train
|
| 47 |
+
path: mined_tables/ft_rules_chh.parquet
|
| 48 |
+
- config_name: mined_tables_chh_wd
|
| 49 |
+
data_files:
|
| 50 |
+
- split: train
|
| 51 |
+
path: mined_tables/ft_rules_chhwd.parquet
|
| 52 |
+
- config_name: casual
|
| 53 |
+
data_files:
|
| 54 |
+
- split: train
|
| 55 |
+
path: casual/ft_casual_noakk.parquet
|
| 56 |
+
- config_name: synthetic_heldout
|
| 57 |
+
data_files:
|
| 58 |
+
- split: validation
|
| 59 |
+
path: validation/train_v2_s3.dev.parquet
|
| 60 |
+
---
|
| 61 |
+
|
| 62 |
+
# Samdei Type Fine-tuning Data
|
| 63 |
+
|
| 64 |
+
These are the exact **fine-tuning (stage 2)** and **adaptation (stage 3)** pairs of the Samdei Type romanized-Khmer → Khmer-script converters, [Samdei Type Mini](https://huggingface.co/samdei/samdei-type-mini) and [Samdei Type Base](https://huggingface.co/samdei/samdei-type-base).
|
| 65 |
+
|
| 66 |
+
Pretraining (stage 1, [`samdei/samdei-type-pretraining-data`](https://huggingface.co/datasets/samdei/samdei-type-pretraining-data)) teaches Khmer words and spelling from 6.5–12.8M generator pairs. These two short stages teach **how people actually type in chat**:
|
| 67 |
+
- **Stage 2:** looser LLM spellings, spelling habits mined from public word lists, and chat-like abbreviations.
|
| 68 |
+
- **Stage 3:** the spoken register and stretched letters.
|
| 69 |
+
|
| 70 |
+
Each stage is 3–6k steps (10–36 minutes on a laptop).
|
| 71 |
+
|
| 72 |
+
Every **Khmer target is real text**, except 2,483 chat sentences written by Gemma 4. Only the **romanized input** varies, by source.
|
| 73 |
+
|
| 74 |
+
Paper: `<paper-url>`. Code: `<code-repo-url>`. Test set: [`samdei/samdei-type-bench`](https://huggingface.co/datasets/samdei/samdei-type-bench).
|
| 75 |
+
|
| 76 |
+
## Configs
|
| 77 |
+
|
| 78 |
+
The data is organized by kind, so each source of spelling noise can be used, inspected or left out on its own. Stages 2 and 3 share one dataset because stage 3 replays the whole stage-2 mix; the stage-3 data is the `casual` config.
|
| 79 |
+
|
| 80 |
+
| Config | Files (rows) | What it is | Stage | Used by |
|
| 81 |
+
|---|---|---|---|---|
|
| 82 |
+
| `chat_mix` (default) | `ft_v2` (629,878) | a chat-heavy sample of generator pairs: more LLM-written chat sentences and ALT, fewer web pieces; English chat words and place names inserted | 2, 3 | Mini, Base |
|
| 83 |
+
| `llm_romanized` | `train_gemrom2` (18,273) | Gemma 4 romanizations of 6,382 real sentences, 4 style prompts, ×3 copies with random casing | 2, 3 (×4 in the mix) | Mini, Base |
|
| 84 |
+
| `abbreviated` | `ft_v2_abbr3` (139,704), `train_gemrom2_abbr3` (16,302) | chat-like abbreviated copies of `chat_mix` and `llm_romanized` pairs; only the input changes (`tov` → `tv`, `bong` → `bg`) | 2, 3 (Gemma part ×2) | Mini, Base |
|
| 85 |
+
| `mined_tables_chh` | `ft_rules_chh` (289,820) | new pairs spelled with the table mined from the Chhunneng word list (dictionary-like scheme), alone or with mixed / lazy noise habits | 2, 3 | **Mini** |
|
| 86 |
+
| `mined_tables_chh_wd` | `ft_rules_chhwd` (289,647) | made from the same sources with the same seed, spelled 70% with the Chhunneng table and 30% with the table mined from Wikidata names (official, name-style scheme) | 2, 3 | **Base** |
|
| 87 |
+
| `casual` | `ft_casual_noakk` (426,946) | 21.6k spoken-register web sentences, spelled by the generator with stretched letters (`tovvv`, `baaan`) | **3 only** | Mini, Base |
|
| 88 |
+
| `synthetic_heldout` | `train_v2_s3.dev` (12,232) | synthetic held-out pairs monitored during training (validation split) | – | – |
|
| 89 |
+
|
| 90 |
+
Mini and Base use the same data. The only difference is the mined-table file: Mini uses `mined_tables_chh`, Base `mined_tables_chh_wd`.
|
| 91 |
+
|
| 92 |
+
| Mix | Files, with repeats | Pairs |
|
| 93 |
+
|---|---|---|
|
| 94 |
+
| Stage 2, Base | `ft_v2` + `ft_v2_abbr3` + `train_gemrom2` ×4 + `train_gemrom2_abbr3` ×2 + `ft_rules_chhwd` | 1,164,925 |
|
| 95 |
+
| Stage 2, Mini | the same with `ft_rules_chh` | 1,165,098 |
|
| 96 |
+
| Stage 3 | the stage-2 mix of the model + `ft_casual_noakk` | 1,591,871 (Base), 1,592,044 (Mini) |
|
| 97 |
+
|
| 98 |
+
`recipes.json` gives the exact mix of each checkpoint (files in order, repeats, steps, learning rate, initialization), read from the training arguments stored in the checkpoints.
|
| 99 |
+
|
| 100 |
+
```
|
| 101 |
+
generator/ ft_v2.parquet
|
| 102 |
+
llm_romanized/ train_gemrom2.parquet
|
| 103 |
+
abbreviated/ ft_v2_abbr3.parquet train_gemrom2_abbr3.parquet
|
| 104 |
+
mined_tables/ ft_rules_chh.parquet ft_rules_chhwd.parquet
|
| 105 |
+
casual/ ft_casual_noakk.parquet
|
| 106 |
+
validation/ train_v2_s3.dev.parquet
|
| 107 |
+
recipes.json
|
| 108 |
+
```
|
| 109 |
+
|
| 110 |
+
## Columns
|
| 111 |
+
|
| 112 |
+
| Column | Content |
|
| 113 |
+
|---|---|
|
| 114 |
+
| `latin` | romanized input, as the model saw it |
|
| 115 |
+
| `khmer` | Khmer-script target, normalized with `khnormal`; English words, digits and brand names are copied from the input |
|
| 116 |
+
| `persona` | how the input was made (below) |
|
| 117 |
+
| `source` | where the Khmer target comes from: `hplt`, `fineweb2`, `wikipedia`, `alt` (a piece of a web sentence); `gemma` (a piece of an LLM-written chat sentence); `casual` (a piece of a spoken-register HPLT or FineWeb-2 sentence); `gazetteer`; `gemrom_<x>` (a whole sentence from `<x>`, romanized by Gemma 4) |
|
| 118 |
+
|
| 119 |
+
| `persona` | Meaning |
|
| 120 |
+
|---|---|
|
| 121 |
+
| `formal`, `careful`, `mixed`, `lazy` | the generator's typing personas, from standard spelling to heavy chat shortcuts and typos |
|
| 122 |
+
| `place` | standalone place-name pair from our gazetteer |
|
| 123 |
+
| `rules_chh`, `rules_chh+mixed`, `rules_chh+lazy` | Chhunneng-mined table, alone or with the noise habits of `mixed` / `lazy` |
|
| 124 |
+
| `rules_wd`, `rules_wd+mixed` | Wikidata-mined table |
|
| 125 |
+
| `…~stretch` | the persona also stretches letters inside the sentence (stage 3) |
|
| 126 |
+
| `gemrom0`–`gemrom3` | Gemma 4 romanization with one of four style prompts |
|
| 127 |
+
| `abbr` | abbreviated copy of another pair |
|
| 128 |
+
|
| 129 |
+
## How to load
|
| 130 |
+
|
| 131 |
+
```python
|
| 132 |
+
from datasets import load_dataset
|
| 133 |
+
casual = load_dataset("samdei/samdei-type-finetuning-data", "casual", split="train")
|
| 134 |
+
abbr = load_dataset("samdei/samdei-type-finetuning-data", "abbreviated", split="train")
|
| 135 |
+
```
|
| 136 |
+
|
| 137 |
+
To rebuild a mix exactly, read `recipes.json` and concatenate its files in order, with repeats. The training script reads the same four columns from TSV (`scripts/train.py --init <stage-1 checkpoint> --train <files...>`). The stage-1 checkpoints are in [`samdei/samdei-type-pretrained`](https://huggingface.co/samdei/samdei-type-pretrained).
|
| 138 |
+
|
| 139 |
+
## How the pairs were made
|
| 140 |
+
|
| 141 |
+
- **`chat_mix`.** The persona generator (second version) on a chat-heavy sample of sentences. The generator works in four steps:
|
| 142 |
+
1. it cuts sentences into pieces of 1–12 words;
|
| 143 |
+
2. it pronounces each word with Google's Khmer pronunciation lexicon;
|
| 144 |
+
3. it spells the piece in a persona (formal 8%, careful 32%, mixed 40%, lazy 20%);
|
| 145 |
+
4. it adds noise (glued or split words, dropped vowels, swaps, typos), English chat words and place names.
|
| 146 |
+
- **`llm_romanized`.** Gemma 4 12B, run locally, romanized real sentences "the way Cambodians type on Messenger", 10 sentences per call. The Khmer side stays the original sentence, so the target is always correct, and Gemma's looser spelling acts as human-like noise on the input. 97% of the outputs passed our filters.
|
| 147 |
+
- **`mined_tables_*`.** `scripts/mine_rules.py` learns per-slot spelling tables (onset, vowel, coda, glide, context) by Viterbi alignment and hard EM from public (Khmer word, romanization) lists: the Chhunneng word list (24.9k pairs, MIT) and 1.6k Wikidata person and place names (CC0). Words that are targets of a word-level evaluation set are removed first. The tables are in [`samdei/samdei-type-resources`](https://huggingface.co/datasets/samdei/samdei-type-resources). The pairs here are new sentences (HPLT, FineWeb-2, ALT, chat sentences; seed 31) spelled with them.
|
| 148 |
+
- **`abbreviated`** (`scripts/make_abbrev.py --short_top 1000`).
|
| 149 |
+
- It covers 40% of `chat_mix` pairs and all `llm_romanized` pairs.
|
| 150 |
+
- Operations: consonant skeleton (`bong` → `bg`), dropped vowels, first + last letter (`jong` → `jg`), dropped digraph h (`khlang` → `klang`), dropped cluster member.
|
| 151 |
+
- Vowels are dropped only from one-syllable words among the 1,000 most frequent.
|
| 152 |
+
- An abbreviation that the chat lexicon reserves for one word (`nh` for ខ្ញុំ) is never produced for another.
|
| 153 |
+
- **`casual`.** 21.6k HPLT and FineWeb-2 sentences that contain one of 13 spoken-register words (ហ្នឹង, ចឹង, អី, តោះ, …) as a token; misordered PDF text is dropped. They are spelled with stretched letters by the generator (formal 5%, careful 25%, mixed 35%, lazy 15%, Chhunneng table with mixed habits 20%; seed 34). No Akkhara-derived spelling rules are used.
|
| 154 |
+
|
| 155 |
+
**Test separation.**
|
| 156 |
+
- Every sentence of every evaluation set, including Samdei Type Bench and its chat context, is banned in the generator.
|
| 157 |
+
- The rule miner never sees a test row or a word-level test target.
|
| 158 |
+
- No Samdei Type Bench message equals a training target. The only shared strings are common single words.
|
| 159 |
+
|
| 160 |
+
**Reproducibility.** The commands are in the code repository: `run_rules.sh` (mined-table pairs), `run_abbr3.sh` (abbreviated copies), `build_distill_pairs.py` (Gemma romanizations to pairs) and `run_stage3_casual.sh` (casual-register pairs and the stage-3 runs). The `make_synth.py` command line of `ft_v2` is not kept in a script; its seed and sources are recorded. The generator has since gained stricter banned lists, so regenerating today can differ slightly. These files are the ones the released models were trained on.
|
| 161 |
+
|
| 162 |
+
## Limitations
|
| 163 |
+
|
| 164 |
+
- **Synthetic inputs.** No row is real typing. Chat habits are covered only as far as our personas, two mined tables, one LLM and our abbreviation rules reach.
|
| 165 |
+
- **Scheme bias of the mined tables.** The Chhunneng table follows a dictionary-like scheme and the Wikidata table an official name scheme; neither is chat spelling as such.
|
| 166 |
+
- **Pieces, not messages.** Inputs are short (median 26 characters).
|
| 167 |
+
- **Personal data.** Web text may contain names, phone numbers or other personal data of the original authors. Lines with URLs or e-mail-like strings were dropped, but no other personal-data filter was run. Contact `<contact>` to have a row removed.
|
| 168 |
+
- **LLM text.** The Gemma-written chat sentences and romanizations can contain mistakes.
|
| 169 |
+
|
| 170 |
+
## License and attribution
|
| 171 |
+
|
| 172 |
+
**CC BY-SA 4.0** ([license text](https://creativecommons.org/licenses/by-sa/4.0/)). The rows contain pieces of the source sentences, so the source licenses travel with them: Khmer Wikipedia's share-alike terms set the license of the whole set, and rows from the other sources also stay under their own terms below.
|
| 173 |
+
|
| 174 |
+
| Source | Side | License |
|
| 175 |
+
|---|---|---|
|
| 176 |
+
| HPLT v2 (`HPLT/HPLT2.0_cleaned`, khm_Khmr) | Khmer | CC0 packaging; Common Crawl terms of use apply to the text |
|
| 177 |
+
| FineWeb-2 (`HuggingFaceFW/fineweb-2`, khm_Khmr) | Khmer | ODC-By 1.0; Common Crawl terms of use apply to the text |
|
| 178 |
+
| Khmer Wikipedia (`wikimedia/wikipedia`, 20231101.km) | Khmer | CC BY-SA 4.0 (share-alike) |
|
| 179 |
+
| Asian Language Treebank (ALT), Khmer side | Khmer | CC BY 4.0 |
|
| 180 |
+
| LLM-written chat sentences (ours, Gemma 4) | Khmer | CC BY 4.0 (Gemma 4: Apache-2.0) |
|
| 181 |
+
| Gemma 4 romanizations (ours) | Latin | Apache-2.0 model output; Khmer side as its source |
|
| 182 |
+
| Google Khmer pronunciation lexicon | Latin (generator) | CC BY 4.0 |
|
| 183 |
+
| Chhunneng romanized Khmer dataset (© 2025 Chrea Chanchhunneng) | Latin (mined table) | MIT |
|
| 184 |
+
| Wikidata names | Latin (mined table) | CC0 |
|
| 185 |
+
| Chat lexicon, place names, English chat words (ours) | Latin (generator) | CC0 |
|
| 186 |
+
|
| 187 |
+
Rows whose `source` is `wikipedia` or `gemrom_wikipedia` carry Wikipedia's share-alike terms. They occur only in `chat_mix`, `llm_romanized` and `abbreviated`; the `mined_tables_*` and `casual` configs contain none.
|
| 188 |
+
|
| 189 |
+
The MIT notice of the Chhunneng dataset, which the `mined_tables_*` tables derive from, is included as `LICENSE-chhunneng.txt`.
|
| 190 |
+
|
| 191 |
+
**Changes to the source text.** The sentences were normalized with `khnormal`, filtered and deduplicated; the generator cut them into pieces of 1–12 words (the `llm_romanized` rows keep whole sentences). Each Khmer target is paired with a romanization in the `latin` column (see "How the pairs were made").
|
| 192 |
+
|
| 193 |
+
## Citation
|
| 194 |
+
|
| 195 |
+
```bibtex
|
| 196 |
+
@misc{samdei-type-2026,
|
| 197 |
+
title = {Samdei Type: Separating What Is Written from How It Is Spelled for Romanized Khmer Transliteration},
|
| 198 |
+
author = {<authors>},
|
| 199 |
+
year = {2026},
|
| 200 |
+
url = {<paper-url>}
|
| 201 |
+
}
|
| 202 |
+
```
|
| 203 |
+
|
| 204 |
+
Contact: `<contact>`
|
abbreviated/ft_v2_abbr3.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:04e6b7b4981028d7028a708ca6f46bcd473398eefdca918f6afb908a36afa50a
|
| 3 |
+
size 5382454
|
abbreviated/train_gemrom2_abbr3.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cc5388fd3018c08a5c477300ac4382c014364eb3452789a33a9a648b1820b7db
|
| 3 |
+
size 619680
|
casual/ft_casual_noakk.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:72d98a0a80fbd952c0838ddce22b11ed6e58691ae67cf172b14d0cd8ff27e525
|
| 3 |
+
size 12664546
|
generator/ft_v2.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4bb9ba4f297191e1d983a930ae2cf7f8f719e83c15bd73f7ccc861549c14044c
|
| 3 |
+
size 19527944
|
llm_romanized/train_gemrom2.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3e3da572ef70e7985a11a178c99a6d6af5657480cc71e16035609b365dcb3d93
|
| 3 |
+
size 520718
|
mined_tables/ft_rules_chh.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6e9974a4d17b77f93ef0d60a9473785c49d51e2716a05c4be9eb480ecf48639f
|
| 3 |
+
size 7694126
|
mined_tables/ft_rules_chhwd.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2608c00526ae8c7a669eb6dd8bdaf507cb4aca67e57338519f200d7ff303fd8a
|
| 3 |
+
size 7703781
|
recipes.json
ADDED
|
@@ -0,0 +1,212 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"about": "Exact training mixes of the Samdei Type checkpoints trained on this dataset, read from the arguments stored in each checkpoint (scripts/train.py). Files are listed in the order given to the training script; 'repeat' is how many times the file was listed. Paths are relative to this dataset unless they name another repo. 'monitor' is the synthetic held-out file watched during training; checkpoints were selected as given in 'selected_on'.",
|
| 3 |
+
"shared": {
|
| 4 |
+
"optimizer": "AdamW, betas 0.9/0.98, weight decay 0.01",
|
| 5 |
+
"schedule": "cosine decay to 5% of the peak",
|
| 6 |
+
"label_smoothing": 0.1,
|
| 7 |
+
"precision": "bf16 autocast",
|
| 8 |
+
"batching": "about 12k tokens per batch, 2-D length buckets padded to multiples of 16"
|
| 9 |
+
},
|
| 10 |
+
"runs": [
|
| 11 |
+
{
|
| 12 |
+
"model": "samdei-type-base",
|
| 13 |
+
"stage": 2,
|
| 14 |
+
"direction": "latin->khmer",
|
| 15 |
+
"init": "samdei/samdei-type-pretrained: base/",
|
| 16 |
+
"released_as": null,
|
| 17 |
+
"train": [
|
| 18 |
+
{
|
| 19 |
+
"file": "generator/ft_v2.parquet",
|
| 20 |
+
"repeat": 1
|
| 21 |
+
},
|
| 22 |
+
{
|
| 23 |
+
"file": "abbreviated/ft_v2_abbr3.parquet",
|
| 24 |
+
"repeat": 1
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"file": "llm_romanized/train_gemrom2.parquet",
|
| 28 |
+
"repeat": 4
|
| 29 |
+
},
|
| 30 |
+
{
|
| 31 |
+
"file": "abbreviated/train_gemrom2_abbr3.parquet",
|
| 32 |
+
"repeat": 2
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"file": "mined_tables/ft_rules_chhwd.parquet",
|
| 36 |
+
"repeat": 1
|
| 37 |
+
}
|
| 38 |
+
],
|
| 39 |
+
"monitor": "validation/train_v2_s3.dev.parquet",
|
| 40 |
+
"steps": 6000,
|
| 41 |
+
"selected_step": 4000,
|
| 42 |
+
"warmup": 300,
|
| 43 |
+
"peak_lr": 0.0002,
|
| 44 |
+
"max_tokens_per_batch": 12000,
|
| 45 |
+
"seed": 1,
|
| 46 |
+
"selected_on": "lowest mean greedy CER on Dev-Sentences and Akkhara-Preview (selection only; never trained on)",
|
| 47 |
+
"model_config": {
|
| 48 |
+
"d_model": 384,
|
| 49 |
+
"nhead": 6,
|
| 50 |
+
"enc_layers": 6,
|
| 51 |
+
"dec_layers": 3,
|
| 52 |
+
"ff": 1536,
|
| 53 |
+
"dropout": 0.1,
|
| 54 |
+
"max_len": 256,
|
| 55 |
+
"src_vocab": 216,
|
| 56 |
+
"tgt_vocab": 228
|
| 57 |
+
}
|
| 58 |
+
},
|
| 59 |
+
{
|
| 60 |
+
"model": "samdei-type-base",
|
| 61 |
+
"stage": 3,
|
| 62 |
+
"direction": "latin->khmer",
|
| 63 |
+
"init": "Base stage 2",
|
| 64 |
+
"released_as": "samdei/samdei-type-base: model.safetensors (and onnx/)",
|
| 65 |
+
"train": [
|
| 66 |
+
{
|
| 67 |
+
"file": "generator/ft_v2.parquet",
|
| 68 |
+
"repeat": 1
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"file": "abbreviated/ft_v2_abbr3.parquet",
|
| 72 |
+
"repeat": 1
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"file": "llm_romanized/train_gemrom2.parquet",
|
| 76 |
+
"repeat": 4
|
| 77 |
+
},
|
| 78 |
+
{
|
| 79 |
+
"file": "abbreviated/train_gemrom2_abbr3.parquet",
|
| 80 |
+
"repeat": 2
|
| 81 |
+
},
|
| 82 |
+
{
|
| 83 |
+
"file": "mined_tables/ft_rules_chhwd.parquet",
|
| 84 |
+
"repeat": 1
|
| 85 |
+
},
|
| 86 |
+
{
|
| 87 |
+
"file": "casual/ft_casual_noakk.parquet",
|
| 88 |
+
"repeat": 1
|
| 89 |
+
}
|
| 90 |
+
],
|
| 91 |
+
"monitor": "validation/train_v2_s3.dev.parquet",
|
| 92 |
+
"steps": 3000,
|
| 93 |
+
"selected_step": 3000,
|
| 94 |
+
"warmup": 100,
|
| 95 |
+
"peak_lr": 0.0001,
|
| 96 |
+
"max_tokens_per_batch": 12000,
|
| 97 |
+
"seed": 1,
|
| 98 |
+
"selected_on": "lowest greedy CER on the first 400 messages of the closed development set",
|
| 99 |
+
"model_config": {
|
| 100 |
+
"d_model": 384,
|
| 101 |
+
"nhead": 6,
|
| 102 |
+
"enc_layers": 6,
|
| 103 |
+
"dec_layers": 3,
|
| 104 |
+
"ff": 1536,
|
| 105 |
+
"dropout": 0.1,
|
| 106 |
+
"max_len": 256,
|
| 107 |
+
"src_vocab": 216,
|
| 108 |
+
"tgt_vocab": 228
|
| 109 |
+
}
|
| 110 |
+
},
|
| 111 |
+
{
|
| 112 |
+
"model": "samdei-type-mini",
|
| 113 |
+
"stage": 2,
|
| 114 |
+
"direction": "latin->khmer",
|
| 115 |
+
"init": "samdei/samdei-type-pretrained: mini/",
|
| 116 |
+
"released_as": null,
|
| 117 |
+
"train": [
|
| 118 |
+
{
|
| 119 |
+
"file": "generator/ft_v2.parquet",
|
| 120 |
+
"repeat": 1
|
| 121 |
+
},
|
| 122 |
+
{
|
| 123 |
+
"file": "abbreviated/ft_v2_abbr3.parquet",
|
| 124 |
+
"repeat": 1
|
| 125 |
+
},
|
| 126 |
+
{
|
| 127 |
+
"file": "llm_romanized/train_gemrom2.parquet",
|
| 128 |
+
"repeat": 4
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"file": "abbreviated/train_gemrom2_abbr3.parquet",
|
| 132 |
+
"repeat": 2
|
| 133 |
+
},
|
| 134 |
+
{
|
| 135 |
+
"file": "mined_tables/ft_rules_chh.parquet",
|
| 136 |
+
"repeat": 1
|
| 137 |
+
}
|
| 138 |
+
],
|
| 139 |
+
"monitor": "validation/train_v2_s3.dev.parquet",
|
| 140 |
+
"steps": 6000,
|
| 141 |
+
"selected_step": 4000,
|
| 142 |
+
"warmup": 300,
|
| 143 |
+
"peak_lr": 0.0002,
|
| 144 |
+
"max_tokens_per_batch": 12000,
|
| 145 |
+
"seed": 1,
|
| 146 |
+
"selected_on": "lowest mean greedy CER on Dev-Sentences and Akkhara-Preview (selection only; never trained on)",
|
| 147 |
+
"model_config": {
|
| 148 |
+
"d_model": 256,
|
| 149 |
+
"nhead": 4,
|
| 150 |
+
"enc_layers": 6,
|
| 151 |
+
"dec_layers": 2,
|
| 152 |
+
"ff": 1024,
|
| 153 |
+
"dropout": 0.1,
|
| 154 |
+
"max_len": 256,
|
| 155 |
+
"src_vocab": 215,
|
| 156 |
+
"tgt_vocab": 228
|
| 157 |
+
}
|
| 158 |
+
},
|
| 159 |
+
{
|
| 160 |
+
"model": "samdei-type-mini",
|
| 161 |
+
"stage": 3,
|
| 162 |
+
"direction": "latin->khmer",
|
| 163 |
+
"init": "Mini stage 2",
|
| 164 |
+
"released_as": "samdei/samdei-type-mini: onnx/",
|
| 165 |
+
"train": [
|
| 166 |
+
{
|
| 167 |
+
"file": "generator/ft_v2.parquet",
|
| 168 |
+
"repeat": 1
|
| 169 |
+
},
|
| 170 |
+
{
|
| 171 |
+
"file": "abbreviated/ft_v2_abbr3.parquet",
|
| 172 |
+
"repeat": 1
|
| 173 |
+
},
|
| 174 |
+
{
|
| 175 |
+
"file": "llm_romanized/train_gemrom2.parquet",
|
| 176 |
+
"repeat": 4
|
| 177 |
+
},
|
| 178 |
+
{
|
| 179 |
+
"file": "abbreviated/train_gemrom2_abbr3.parquet",
|
| 180 |
+
"repeat": 2
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"file": "mined_tables/ft_rules_chh.parquet",
|
| 184 |
+
"repeat": 1
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"file": "casual/ft_casual_noakk.parquet",
|
| 188 |
+
"repeat": 1
|
| 189 |
+
}
|
| 190 |
+
],
|
| 191 |
+
"monitor": "validation/train_v2_s3.dev.parquet",
|
| 192 |
+
"steps": 3000,
|
| 193 |
+
"selected_step": 2500,
|
| 194 |
+
"warmup": 100,
|
| 195 |
+
"peak_lr": 0.0001,
|
| 196 |
+
"max_tokens_per_batch": 12000,
|
| 197 |
+
"seed": 1,
|
| 198 |
+
"selected_on": "lowest greedy CER on the first 400 messages of the closed development set",
|
| 199 |
+
"model_config": {
|
| 200 |
+
"d_model": 256,
|
| 201 |
+
"nhead": 4,
|
| 202 |
+
"enc_layers": 6,
|
| 203 |
+
"dec_layers": 2,
|
| 204 |
+
"ff": 1024,
|
| 205 |
+
"dropout": 0.1,
|
| 206 |
+
"max_len": 256,
|
| 207 |
+
"src_vocab": 215,
|
| 208 |
+
"tgt_vocab": 228
|
| 209 |
+
}
|
| 210 |
+
}
|
| 211 |
+
]
|
| 212 |
+
}
|
validation/train_v2_s3.dev.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9b4542d22a9847a859badbc27551f4589999cb8fd0ad0e7b5f79e421daef207d
|
| 3 |
+
size 573860
|