mengsay commited on
Commit
73c7c04
·
verified ·
1 Parent(s): 7423a2d

Initial private draft of the Samdei Type release

Browse files
LICENSE ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ Apache License
3
+ Version 2.0, January 2004
4
+ http://www.apache.org/licenses/
5
+
6
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
7
+
8
+ 1. Definitions.
9
+
10
+ "License" shall mean the terms and conditions for use, reproduction,
11
+ and distribution as defined by Sections 1 through 9 of this document.
12
+
13
+ "Licensor" shall mean the copyright owner or entity authorized by
14
+ the copyright owner that is granting the License.
15
+
16
+ "Legal Entity" shall mean the union of the acting entity and all
17
+ other entities that control, are controlled by, or are under common
18
+ control with that entity. For the purposes of this definition,
19
+ "control" means (i) the power, direct or indirect, to cause the
20
+ direction or management of such entity, whether by contract or
21
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
22
+ outstanding shares, or (iii) beneficial ownership of such entity.
23
+
24
+ "You" (or "Your") shall mean an individual or Legal Entity
25
+ exercising permissions granted by this License.
26
+
27
+ "Source" form shall mean the preferred form for making modifications,
28
+ including but not limited to software source code, documentation
29
+ source, and configuration files.
30
+
31
+ "Object" form shall mean any form resulting from mechanical
32
+ transformation or translation of a Source form, including but
33
+ not limited to compiled object code, generated documentation,
34
+ and conversions to other media types.
35
+
36
+ "Work" shall mean the work of authorship, whether in Source or
37
+ Object form, made available under the License, as indicated by a
38
+ copyright notice that is included in or attached to the work
39
+ (an example is provided in the Appendix below).
40
+
41
+ "Derivative Works" shall mean any work, whether in Source or Object
42
+ form, that is based on (or derived from) the Work and for which the
43
+ editorial revisions, annotations, elaborations, or other modifications
44
+ represent, as a whole, an original work of authorship. For the purposes
45
+ of this License, Derivative Works shall not include works that remain
46
+ separable from, or merely link (or bind by name) to the interfaces of,
47
+ the Work and Derivative Works thereof.
48
+
49
+ "Contribution" shall mean any work of authorship, including
50
+ the original version of the Work and any modifications or additions
51
+ to that Work or Derivative Works thereof, that is intentionally
52
+ submitted to Licensor for inclusion in the Work by the copyright owner
53
+ or by an individual or Legal Entity authorized to submit on behalf of
54
+ the copyright owner. For the purposes of this definition, "submitted"
55
+ means any form of electronic, verbal, or written communication sent
56
+ to the Licensor or its representatives, including but not limited to
57
+ communication on electronic mailing lists, source code control systems,
58
+ and issue tracking systems that are managed by, or on behalf of, the
59
+ Licensor for the purpose of discussing and improving the Work, but
60
+ excluding communication that is conspicuously marked or otherwise
61
+ designated in writing by the copyright owner as "Not a Contribution."
62
+
63
+ "Contributor" shall mean Licensor and any individual or Legal Entity
64
+ on behalf of whom a Contribution has been received by Licensor and
65
+ subsequently incorporated within the Work.
66
+
67
+ 2. Grant of Copyright License. Subject to the terms and conditions of
68
+ this License, each Contributor hereby grants to You a perpetual,
69
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
70
+ copyright license to reproduce, prepare Derivative Works of,
71
+ publicly display, publicly perform, sublicense, and distribute the
72
+ Work and such Derivative Works in Source or Object form.
73
+
74
+ 3. Grant of Patent License. Subject to the terms and conditions of
75
+ this License, each Contributor hereby grants to You a perpetual,
76
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
77
+ (except as stated in this section) patent license to make, have made,
78
+ use, offer to sell, sell, import, and otherwise transfer the Work,
79
+ where such license applies only to those patent claims licensable
80
+ by such Contributor that are necessarily infringed by their
81
+ Contribution(s) alone or by combination of their Contribution(s)
82
+ with the Work to which such Contribution(s) was submitted. If You
83
+ institute patent litigation against any entity (including a
84
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
85
+ or a Contribution incorporated within the Work constitutes direct
86
+ or contributory patent infringement, then any patent licenses
87
+ granted to You under this License for that Work shall terminate
88
+ as of the date such litigation is filed.
89
+
90
+ 4. Redistribution. You may reproduce and distribute copies of the
91
+ Work or Derivative Works thereof in any medium, with or without
92
+ modifications, and in Source or Object form, provided that You
93
+ meet the following conditions:
94
+
95
+ (a) You must give any other recipients of the Work or
96
+ Derivative Works a copy of this License; and
97
+
98
+ (b) You must cause any modified files to carry prominent notices
99
+ stating that You changed the files; and
100
+
101
+ (c) You must retain, in the Source form of any Derivative Works
102
+ that You distribute, all copyright, patent, trademark, and
103
+ attribution notices from the Source form of the Work,
104
+ excluding those notices that do not pertain to any part of
105
+ the Derivative Works; and
106
+
107
+ (d) If the Work includes a "NOTICE" text file as part of its
108
+ distribution, then any Derivative Works that You distribute must
109
+ include a readable copy of the attribution notices contained
110
+ within such NOTICE file, excluding those notices that do not
111
+ pertain to any part of the Derivative Works, in at least one
112
+ of the following places: within a NOTICE text file distributed
113
+ as part of the Derivative Works; within the Source form or
114
+ documentation, if provided along with the Derivative Works; or,
115
+ within a display generated by the Derivative Works, if and
116
+ wherever such third-party notices normally appear. The contents
117
+ of the NOTICE file are for informational purposes only and
118
+ do not modify the License. You may add Your own attribution
119
+ notices within Derivative Works that You distribute, alongside
120
+ or as an addendum to the NOTICE text from the Work, provided
121
+ that such additional attribution notices cannot be construed
122
+ as modifying the License.
123
+
124
+ You may add Your own copyright statement to Your modifications and
125
+ may provide additional or different license terms and conditions
126
+ for use, reproduction, or distribution of Your modifications, or
127
+ for any such Derivative Works as a whole, provided Your use,
128
+ reproduction, and distribution of the Work otherwise complies with
129
+ the conditions stated in this License.
130
+
131
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
132
+ any Contribution intentionally submitted for inclusion in the Work
133
+ by You to the Licensor shall be under the terms and conditions of
134
+ this License, without any additional terms or conditions.
135
+ Notwithstanding the above, nothing herein shall supersede or modify
136
+ the terms of any separate license agreement you may have executed
137
+ with Licensor regarding such Contributions.
138
+
139
+ 6. Trademarks. This License does not grant permission to use the trade
140
+ names, trademarks, service marks, or product names of the Licensor,
141
+ except as required for reasonable and customary use in describing the
142
+ origin of the Work and reproducing the content of the NOTICE file.
143
+
144
+ 7. Disclaimer of Warranty. Unless required by applicable law or
145
+ agreed to in writing, Licensor provides the Work (and each
146
+ Contributor provides its Contributions) on an "AS IS" BASIS,
147
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
148
+ implied, including, without limitation, any warranties or conditions
149
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
150
+ PARTICULAR PURPOSE. You are solely responsible for determining the
151
+ appropriateness of using or redistributing the Work and assume any
152
+ risks associated with Your exercise of permissions under this License.
153
+
154
+ 8. Limitation of Liability. In no event and under no legal theory,
155
+ whether in tort (including negligence), contract, or otherwise,
156
+ unless required by applicable law (such as deliberate and grossly
157
+ negligent acts) or agreed to in writing, shall any Contributor be
158
+ liable to You for damages, including any direct, indirect, special,
159
+ incidental, or consequential damages of any character arising as a
160
+ result of this License or out of the use or inability to use the
161
+ Work (including but not limited to damages for loss of goodwill,
162
+ work stoppage, computer failure or malfunction, or any and all
163
+ other commercial damages or losses), even if such Contributor
164
+ has been advised of the possibility of such damages.
165
+
166
+ 9. Accepting Warranty or Additional Liability. While redistributing
167
+ the Work or Derivative Works thereof, You may choose to offer,
168
+ and charge a fee for, acceptance of support, warranty, indemnity,
169
+ or other liability obligations and/or rights consistent with this
170
+ License. However, in accepting such obligations, You may act only
171
+ on Your own behalf and on Your sole responsibility, not on behalf
172
+ of any other Contributor, and only if You agree to indemnify,
173
+ defend, and hold each Contributor harmless for any liability
174
+ incurred by, or claims asserted against, such Contributor by reason
175
+ of your accepting any such warranty or additional liability.
176
+
177
+ END OF TERMS AND CONDITIONS
178
+
179
+ APPENDIX: How to apply the Apache License to your work.
180
+
181
+ To apply the Apache License to your work, attach the following
182
+ boilerplate notice, with the fields enclosed by brackets "[]"
183
+ replaced with your own identifying information. (Don't include
184
+ the brackets!) The text should be enclosed in the appropriate
185
+ comment syntax for the file format. We also recommend that a
186
+ file or class name and description of purpose be included on the
187
+ same "printed page" as the copyright notice for easier
188
+ identification within third-party archives.
189
+
190
+ Copyright [yyyy] [name of copyright owner]
191
+
192
+ Licensed under the Apache License, Version 2.0 (the "License");
193
+ you may not use this file except in compliance with the License.
194
+ You may obtain a copy of the License at
195
+
196
+ http://www.apache.org/licenses/LICENSE-2.0
197
+
198
+ Unless required by applicable law or agreed to in writing, software
199
+ distributed under the License is distributed on an "AS IS" BASIS,
200
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
201
+ See the License for the specific language governing permissions and
202
+ limitations under the License.
NOTICE ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Samdei Type Mini
2
+ Copyright 2026 The Samdei Type Authors
3
+
4
+ Licensed under the Apache License, Version 2.0 (see LICENSE). This covers every file in this
5
+ repository: the ONNX graphs, onnx/vocab.json, rerank.json and the word language model.
6
+
7
+ Word language model (lm/wordfull_nowiki_p25.json.gz)
8
+ Word unigram and bigram counts of Khmer text, built without Wikipedia, from:
9
+ - HPLT v2 (HPLT/HPLT2.0_cleaned, khm_Khmr): CC0 packaging; the Common Crawl Terms of Use
10
+ apply to the text.
11
+ - FineWeb-2 (HuggingFaceFW/fineweb-2, khm_Khmr): ODC-By 1.0; the Common Crawl Terms of Use
12
+ apply to the text.
13
+ - Asian Language Treebank (ALT), Khmer side: CC BY 4.0.
14
+ - Samdei Type chat sentences (samdei/samdei-type-resources): CC BY 4.0.
15
+
16
+ Training data
17
+ The model was trained on samdei/samdei-type-pretraining-data and
18
+ samdei/samdei-type-finetuning-data (CC BY-SA 4.0); their cards list every source. Part of the
19
+ fine-tuning pairs were spelled with a table derived from the Chhunneng romanized Khmer
20
+ dataset, MIT License, Copyright (c) 2025 Chrea Chanchhunneng.
README.md ADDED
@@ -0,0 +1,244 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ language:
4
+ - km
5
+ library_name: onnx
6
+ pipeline_tag: text2text-generation
7
+ tags:
8
+ - transliteration
9
+ - romanization
10
+ - romanized-khmer
11
+ - khmer
12
+ - character-level
13
+ - onnx
14
+ - int8
15
+ - keyboard
16
+ - on-device
17
+ base_model: samdei/samdei-type-pretrained
18
+ base_model_relation: finetune
19
+ datasets:
20
+ - samdei/samdei-type-pretraining-data
21
+ - samdei/samdei-type-finetuning-data
22
+ metrics:
23
+ - cer
24
+ - exact_match
25
+ model-index:
26
+ - name: Samdei Type Mini
27
+ results:
28
+ - task:
29
+ type: text2text-generation
30
+ name: Romanized Khmer to Khmer script
31
+ dataset:
32
+ type: samdei/samdei-type-bench
33
+ name: Samdei Type Bench
34
+ split: test
35
+ metrics:
36
+ - type: cer
37
+ value: 0.058
38
+ name: CER (beam 5, word-LM rerank, 8-word pieces)
39
+ - type: exact_match
40
+ value: 0.239
41
+ name: Exact match
42
+ ---
43
+
44
+ # Samdei Type Mini
45
+
46
+ Samdei Type Mini converts romanized Khmer, as people type it in chat, into Khmer script:
47
+
48
+ | Input | Output |
49
+ |---|---|
50
+ | `knhom jong tov phsar` | ខ្ញុំចង់ទៅផ្សារ |
51
+ | `ot mean wifi te` | អត់មាន wifi ទេ |
52
+
53
+ It is a 7.1M-parameter character-level Transformer encoder-decoder. It ships as int8 ONNX (7.0 MiB), with a 1.2 MiB word bigram LM that reranks its candidates. Mini is meant for on-device use, such as a phone keyboard that offers Khmer-script candidates as the user types.
54
+
55
+ Samdei (សំដី, "what you say") is our team, and Samdei Type is the converter. The family has three models:
56
+
57
+ | Model | Parameters | Released as | Samdei Type Bench CER / exact |
58
+ |---|---|---|---|
59
+ | **Samdei Type Mini** (this repo) | 7.1M | int8 ONNX + pruned word LM | 0.058 / 24% |
60
+ | Samdei Type Base | 18.2M | PyTorch + int8 ONNX + LMs + channel model: [`samdei/samdei-type-base`](https://huggingface.co/samdei/samdei-type-base) | 0.052 / 30% |
61
+ | Samdei Type Prime | 18.2M | not released | 0.027 / 61% |
62
+
63
+ Samdei Type Prime is Base trained further on scenario-designed, LLM-drafted, human-audited data. It is not released.
64
+
65
+ The pretrained (stage-1) checkpoints of both sizes, for fine-tuning your own converter, are in [`samdei/samdei-type-pretrained`](https://huggingface.co/samdei/samdei-type-pretrained).
66
+
67
+ Paper: *Samdei Type: Separating What Is Written from How It Is Spelled for Romanized Khmer Transliteration* (`<paper-url>`). Code: `<code-repo-url>`.
68
+
69
+ ## Files
70
+
71
+ ```
72
+ onnx/encoder_kv.int8.onnx 5.0 MiB input characters -> cross-attention K/V of every decoder layer
73
+ onnx/decoder_step.int8.onnx 2.0 MiB one decoder step with a self-attention KV cache
74
+ onnx/vocab.json vocabularies and graph shapes (layers, heads, head_dim, bos, eos, max_src_len)
75
+ lm/wordfull_nowiki_p25.json.gz 1.2 MiB word bigram LM over khmercut words, pruned to bigram count >= 25 (gzipped JSON)
76
+ rerank.json decoding and rerank settings used for every number below
77
+ ```
78
+
79
+ ## Intended use
80
+
81
+ - **Inside a keyboard:** turn what a user types in Latin letters into a list of Khmer-script candidates on every keystroke, offline.
82
+ - **Reading:** show someone else's romanized message in Khmer script.
83
+ - **Mixed text:** English words, numbers and brand names are copied through unchanged.
84
+
85
+ **Out of scope:**
86
+ - the reverse direction (Khmer → Latin);
87
+ - official romanization standards, such as names in documents;
88
+ - any setting where the user cannot see and correct the output. The model makes mistakes on homophones and rare words (see Limitations).
89
+
90
+ ## How to use
91
+
92
+ Install the `rkhmer` package from the code repository:
93
+
94
+ ```bash
95
+ pip install "rkhmer[deploy] @ git+<code-repo-url>"
96
+ ```
97
+
98
+ ```python
99
+ from huggingface_hub import snapshot_download
100
+ from rkhmer.evaluation import chunk_source
101
+ from rkhmer.onnx_infer import OnnxConverter
102
+ from rkhmer.rerank import Reranker, load_weights
103
+
104
+ d = snapshot_download("samdei/samdei-type-mini", allow_patterns=["onnx/*", "lm/*", "rerank.json"])
105
+ conv = OnnxConverter(f"{d}/onnx") # int8 graphs + vocab.json, numpy + onnxruntime only
106
+ rr = Reranker(load_weights(f"{d}/rerank.json"), wordlm=f"{d}/lm/wordfull_nowiki_p25.json.gz")
107
+
108
+ def convert(text: str) -> str:
109
+ out = []
110
+ for piece in chunk_source(text, 8): # at most 8 words per piece, break after punctuation
111
+ hyps = conv.beam(piece, k=5) # [(khmer, logprob), ...], best first
112
+ out.append(rr.rerank(piece, hyps)[0][0])
113
+ return " ".join(out)
114
+
115
+ print(convert("knhom jong tov phsar")) # ខ្ញុំចង់ទៅផ្សារ
116
+ ```
117
+
118
+ **Decoding, as in the paper.**
119
+ - Beam search with k = 5.
120
+ - The input is cut into pieces of at most eight words, with a break after any punctuation mark. Each piece is decoded and reranked on its own, and the outputs are joined with a space.
121
+ - Each hypothesis y for input x is scored as `log P(y|x) + 0.3 · log P_word(y) − 1.0 · #OOV(y)`. The weights were tuned on a closed development set, and `rerank.json` holds them.
122
+ - Inputs are truncated to 249 characters.
123
+
124
+ **On a phone.**
125
+ - The two graphs are all a runtime needs.
126
+ - `encoder_kv` runs once per input.
127
+ - `decoder_step` maps the newest token, its position, the past self-attention K/V and the cross-attention K/V to the logits and the new K/V. It has a dynamic batch axis, so all beams run in one call.
128
+ - `rkhmer/onnx_infer.py` is a reference implementation in plain numpy (no torch) to port from.
129
+ - The cached decoder reproduces full re-decoding exactly. Int8 quantization does not change CER on Dev-Sentences with beam 5 and no LMs (0.010 in fp32 and in int8).
130
+
131
+ ## Model
132
+
133
+ | | Samdei Type Mini |
134
+ |---|---|
135
+ | Architecture | pre-LN Transformer encoder-decoder, GELU, learned positions up to 256, untied output projection |
136
+ | d_model / heads | 256 / 4 |
137
+ | Encoder / decoder layers | 6 / 2 (deep encoder, shallow decoder) |
138
+ | FFN width | 1024 |
139
+ | Parameters | 7,149,540 |
140
+ | Vocabulary | 215 input symbols (Latin letters, digits, punctuation, space), 228 output symbols (Khmer code points, space, copied characters), including special tokens |
141
+ | int8 ONNX | 7.0 MiB as released (7.6 MiB in the paper, with the exporter's debug metadata) |
142
+
143
+ ## Training
144
+
145
+ Mini was trained in three stages. Each stage started from the best checkpoint of the stage before and replayed its data.
146
+
147
+ | | Stage 1: pretraining | Stage 2: fine-tuning | Stage 3: adaptation |
148
+ |---|---|---|---|
149
+ | New data | rule-based persona generator applied to 1.41M web sentences and the 2.4k LLM-written chat sentences (6.50M pairs), + Gemma 4 romanizations of real sentences (5.7k ×3) | chat-heavy generator sample 630k; Gemma 4 romanizations 18.3k ×4; pairs spelled with the table mined from the Chhunneng word list 290k; chat-like abbreviated copies 140k + 16k ×2 | casual-register pairs 427k (spoken-register web sentences, stretched letters) |
150
+ | Replayed | – | – | the stage-2 mix |
151
+ | Total pairs | 6.5M | 1.16M | 1.59M |
152
+ | Steps (best checkpoint) | 40k (35k) | 6k (4k) | 3k (2.5k) |
153
+ | Peak lr, warmup | 1e-3, 2k | 2e-4, 300 | 1e-4, 100 |
154
+ | Wall clock | 2.2 h | about 31–36 min | about 10–12 min |
155
+
156
+ **Shared settings:**
157
+ - AdamW (β 0.9 / 0.98, weight decay 0.01), cosine decay to 5% of the peak, label smoothing 0.1;
158
+ - about 12k tokens per batch, bf16 autocast;
159
+ - one laptop GPU (Apple M5 Pro).
160
+
161
+ **Checkpoint selection:**
162
+ - Stages 1–2: the lowest mean greedy CER on Dev-Sentences and Akkhara-Preview. Akkhara-Preview is the 110-row public preview of Akkhara-Sing-Khmer; it was used for this selection only and never trained on.
163
+ - Stage 3: the lowest greedy CER on the first 400 messages of a closed development set.
164
+
165
+ **Data.** The exact training files are published:
166
+ - **Stage 1:** [`samdei/samdei-type-pretraining-data`](https://huggingface.co/datasets/samdei/samdei-type-pretraining-data), configs `generator_v2` and `llm_romanized`. The pretrained checkpoint is `mini/` in [`samdei/samdei-type-pretrained`](https://huggingface.co/samdei/samdei-type-pretrained).
167
+ - **Stages 2–3:** [`samdei/samdei-type-finetuning-data`](https://huggingface.co/datasets/samdei/samdei-type-finetuning-data). Mini uses `mined_tables_chh`.
168
+
169
+ Each repo's `recipes.json` gives the exact mix.
170
+ - **Khmer targets.** Always real text, except 2,483 chat sentences written by Gemma 4. The text comes from HPLT v2, FineWeb-2, Khmer Wikipedia and ALT.
171
+ - **Romanized inputs.** These come from:
172
+ - the persona generator, which uses Google's Khmer pronunciation lexicon;
173
+ - a spelling table mined from the Chhunneng word list;
174
+ - Gemma 4 romanizations of real sentences;
175
+ - chat-like abbreviations.
176
+ - **Not used:** no evaluation set, and no data of other romanization systems (in particular, no Akkhara-Sing-Khmer data).
177
+
178
+ ## Evaluation
179
+
180
+ **Samdei Type Bench** ([`samdei/samdei-type-bench`](https://huggingface.co/datasets/samdei/samdei-type-bench)) has 2,155 everyday messages and words with human-checked Khmer. 192 of them were typed blind by native speakers. Inputs have a median of 81 characters.
181
+
182
+ - **Metric.** Character error rate (CER) after normalization (`khnormal`, spaces removed, lowercase, coeng-DA merged with coeng-TA, ឲ merged with ឱ) and exact match, on each item's main input and main target.
183
+ - **Same code for every system:** `rkhmer.evaluation.score`.
184
+
185
+ | System | Parameters | int8 size | CER | Exact |
186
+ |---|---|---|---|---|
187
+ | **Samdei Type Mini** | 7.1M | 7.6 MiB | 0.058 | 24% |
188
+ | Samdei Type Base | 18.2M | 18.2 MiB | **0.052** | **30%** |
189
+ | Samdei Type Prime (not released) | 18.2M | 18.2 MiB | 0.027 | 61% |
190
+ | Akkhara-mini, whole input | 17.8M | 17.6 MiB | 0.159 | 27% |
191
+ | Akkhara-mini, 6-word pieces | 17.8M | 17.6 MiB | 0.066 | 23% |
192
+ | AkaraAlpha, word by word | 0.08M | – | 0.281 | 2% |
193
+
194
+ - **Prime is partly in-domain.** Its training data shares the scenario design of most Samdei Type Bench items, so its lead is an upper bound.
195
+ - **Reranking.** Beam 5 alone gives 0.072 / 17%; the pruned word LM brings it to 0.058 / 24%.
196
+ - **Decode long messages in pieces.** Mini was trained on short pieces (inputs of at most 132 characters, median 26). With beam 5 and no LMs, whole-input CER is 0.218 overall and 0.558 on inputs over 132 characters, against 0.072 and 0.076 in pieces.
197
+ - **Reference sets** (CER / exact): Dev-Sentences 0.005 / 95%, Chat-Words 0.081 / 84%, Dict-Words 0.282 / 34%, Casual-Synth 0.055 / 62%.
198
+ - **Speed.** One M5 Pro CPU thread, int8 ONNX, without reranking:
199
+ - greedy: 2.2 ms p50 / 4.2 ms p95 per input;
200
+ - beam 5: 4.9 / 10.0 ms.
201
+ - Per keystroke in the live demo, with the word-LM rerank: 5.7 ms p50 / 17.2 ms p95.
202
+ - Not yet measured on a phone.
203
+
204
+ ## Limitations
205
+
206
+ - **Synthetic training inputs.** Every romanized input Mini has seen comes from rules, mined tables or an LLM; real typing appears only in the evaluation.
207
+ - **Homophones that need more context than a word bigram gives**, for example:
208
+ - ដើរ / ដែរ (`der`);
209
+ - ដឹក / ដែក / ដេក (`dek`);
210
+ - អាទិត្យ / អតីត (`atit`);
211
+ - the polite particle បាទ against បាត់ ("lost") at the end of a message.
212
+ - **Spoken-register words that share a spelling with a written word:** `nung` is ហ្នឹង in chat and និង ("and") in prose.
213
+ - **Loanwords typed in English** (`check` for ឆែក) are copied through as English.
214
+ - **Punctuation.** Mini copies a Latin comma where Khmer uses a space, and never adds a final ។ that the input lacks.
215
+ - **Heavy abbreviation and dictionary words.** Mini is weaker on input abbreviated more heavily than its training data and on rare dictionary words without context, many of them Pali-Sanskrit (Dict-Words CER 0.282).
216
+ - **Web text.** The Khmer side comes mostly from web text and carries its topics and biases.
217
+
218
+ ## License and attribution
219
+
220
+ - **Everything in this repo** (ONNX graphs, vocabulary, rerank settings, word LM): Apache-2.0. The full text is in `LICENSE`. `NOTICE` holds the attributions, and redistributions must keep it.
221
+ - **`lm/wordfull_nowiki_p25.json.gz`:** word n-gram counts built without Wikipedia, from:
222
+ - HPLT v2 (CC0 packaging; Common Crawl terms apply to the text);
223
+ - FineWeb-2 (ODC-By 1.0);
224
+ - ALT (CC BY 4.0);
225
+ - our LLM-written chat sentences (CC BY 4.0; written with Gemma 4, Apache-2.0).
226
+ - **Training data:** CC BY-SA 4.0; see the dataset cards. Its sources are:
227
+ - HPLT v2, FineWeb-2, Khmer Wikipedia (CC BY-SA) and ALT;
228
+ - Google's Khmer pronunciation lexicon (CC BY 4.0);
229
+ - the Chhunneng romanized Khmer dataset (MIT, © 2025 Chrea Chanchhunneng);
230
+ - Gemma 4 output (Apache-2.0).
231
+ - **Tools:** `khmercut` (Apache-2.0) and SIL `khnormal` (MIT).
232
+
233
+ ## Citation
234
+
235
+ ```bibtex
236
+ @misc{samdei-type-2026,
237
+ title = {Samdei Type: Separating What Is Written from How It Is Spelled for Romanized Khmer Transliteration},
238
+ author = {<authors>},
239
+ year = {2026},
240
+ url = {<paper-url>}
241
+ }
242
+ ```
243
+
244
+ Contact: `<contact>`
lm/wordfull_nowiki_p25.json.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6acd33fa23ef4b0c77318d53823212a46cdc87eb21d786603421453886ef4051
3
+ size 1213177
onnx/decoder_step.int8.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2808c03ce9482819a367e0ce933be11e8215634e3c016d9ef10c020f462a3d58
3
+ size 2093997
onnx/encoder_kv.int8.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d4c8c57b7a10bd05dd4d1292114adb44a311e3521754e181d9065f5be6f65eb3
3
+ size 5269088
onnx/vocab.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"src_vocab": [" ", "!", "\"", "#", "$", "%", "&", "'", "(", ")", "*", "+", ",", "-", ".", "/", "0", "1", "2", "3", "4", "5", "6", "7", "8", "9", ":", ";", "<", "=", ">", "?", "@", "A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K", "L", "M", "N", "O", "P", "Q", "R", "S", "T", "U", "V", "W", "X", "Y", "Z", "[", "]", "^", "_", "a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "|", "~", "¡", "«", "®", "°", "·", "»", "×", "á", "ç", "è", "é", "í", "ǔ", "ɓ", "ə", "ก", "ท", "น", "ภ", "ย", "ร", "ษ", "า", "ิ", "เ", "ไ", "๒", "๕", "ລ", "ວ", "ສ", "າ", "ក", "ខ", "គ", "ឃ", "ង", "ច", "ឆ", "ជ", "ឈ", "ញ", "ដ", "ឋ", "ឌ", "ណ", "ត", "ថ", "ទ", "ធ", "ន", "ប", "ផ", "ព", "ភ", "ម", "យ", "រ", "ល", "វ", "ស", "ហ", "ឡ", "អ", "ឧ", "ឫ", "ឬ", "ឯ", "ឲ", "ា", "ិ", "ី", "ឹ", "ឺ", "ុ", "ូ", "ួ", "ើ", "ឿ", "ៀ", "េ", "ែ", "ៃ", "ោ", "ៅ", "ំ", "ះ", "ៈ", "៉", "៊", "់", "៌", "៍", "៏", "័", "្", "។", "៕", "៖", "ៗ", "៘", "៛", "‑", "–", "—", "‘", "’", "“", "”", "•", "…", "‹", "›", "→", "−", "●", "�", "😀", "😛", "🙂"], "tgt_vocab": [" ", "!", "\"", "#", "$", "%", "&", "'", "(", ")", "*", "+", ",", "-", ".", "/", "0", "1", "2", "3", "4", "5", "6", "7", "8", "9", ":", ";", "<", "=", ">", "?", "@", "A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K", "L", "M", "N", "O", "P", "Q", "R", "S", "T", "U", "V", "W", "X", "Y", "Z", "[", "]", "^", "_", "a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "|", "~", "¡", "«", "®", "°", "·", "»", "×", "à", "á", "ç", "è", "é", "í", "ǔ", "ก", "ท", "น", "ภ", "ย", "ร", "ษ", "า", "ิ", "เ", "ไ", "๒", "๕", "ລ", "ວ", "ສ", "າ", "ក", "ខ", "គ", "ឃ", "ង", "ច", "ឆ", "ជ", "ឈ", "ញ", "ដ", "ឋ", "ឌ", "ឍ", "ណ", "ត", "ថ", "ទ", "ធ", "ន", "ប", "ផ", "ព", "ភ", "ម", "យ", "រ", "ល", "វ", "ឝ", "ឞ", "ស", "ហ", "ឡ", "អ", "ឥ", "ឦ", "ឧ", "ឩ", "ឪ", "ឫ", "ឬ", "ឭ", "ឮ", "ឯ", "ឰ", "ឱ", "ឲ", "ឳ", "ា", "ិ", "ី", "ឹ", "ឺ", "ុ", "ូ", "ួ", "ើ", "ឿ", "ៀ", "េ", "ែ", "ៃ", "ោ", "ៅ", "ំ", "ះ", "ៈ", "៉", "៊", "់", "៌", "៍", "៎", "៏", "័", "៑", "្", "។", "៕", "៖", "ៗ", "៘", "៛", "‑", "–", "—", "‘", "’", "“", "”", "•", "…", "‹", "›", "→", "−", "●", "�", "😀", "😛", "🙂"], "layers": 2, "heads": 4, "head_dim": 64, "bos": 1, "eos": 2, "max_src_len": 249}
rerank.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "Samdei Type Mini",
3
+ "beam": 5,
4
+ "chunk_max_words": 8,
5
+ "chunk_rule": "rkhmer.evaluation.chunk_source: at most 8 words per piece, break after any punctuation mark; pieces are decoded and reranked on their own and joined with a space",
6
+ "rerank_weights": {"char": 0.0, "word": 0.3, "oov": 1.0, "channel": 0.0},
7
+ "score": "log P(y|x) + char*log P_char(y) + word*log P_word(y) - oov*#OOV(y) + channel*log P_ch(x|y)",
8
+ "char_lm": null,
9
+ "word_lm": "lm/wordfull_nowiki_p25.json.gz",
10
+ "channel_model": null,
11
+ "max_input_chars": 249,
12
+ "weights_tuned_on": "closed development set, whole input, 2-fold cross-validated grid search; applied unchanged to 8-word pieces"
13
+ }