AmerAlreyahi commited on
Commit
2d145d4
·
verified ·
1 Parent(s): bd68441

Add OptimaMT th-en

Browse files
.gitattributes CHANGED
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ source.spm filter=lfs diff=lfs merge=lfs -text
37
+ target.spm filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,57 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ language:
4
+ - th
5
+ - en
6
+ pipeline_tag: translation
7
+ library_name: transformers.js
8
+ base_model: Xenova/opus-mt-th-en
9
+ tags:
10
+ - translation
11
+ - marian
12
+ - opus-mt
13
+ - onnx
14
+ ---
15
+
16
+ # OptimaMT th → en (Thai to English)
17
+
18
+ Drop-in copy of [Xenova/opus-mt-th-en](https://huggingface.co/Xenova/opus-mt-th-en) at revision `7f0406b972f7` with a lighter merged decoder: same weights, same tokens, less memory.
19
+ Part of [OptimaMT](https://github.com/AmerAlreyahi/OptimaMT) (49 pairs, `OptimaMT/<src>-<tgt>`).
20
+
21
+ | Memory after load | Linux | Windows | Identical tokens |
22
+ |---|---:|---:|---|
23
+ | `disable_quant_qdq = 1` | 524 → 280 MiB (-47 %) | 501 → 260 MiB (-48 %) | yes |
24
+ | ONNX Runtime defaults | 280 → 219 MiB (-22 %) | 199 → 167 MiB (-16 %) | yes |
25
+
26
+ ONNX Runtime 1.31.1, CPU, 4 threads (Linux; Windows in a VM). Interactive results: [OptimaMT/results](https://huggingface.co/spaces/OptimaMT/results) ·
27
+ method: [REPORT.md](https://github.com/AmerAlreyahi/OptimaMT/blob/main/REPORT.md).
28
+
29
+ ## Use
30
+
31
+ ```js
32
+ import { pipeline } from "@huggingface/transformers";
33
+
34
+ const translator = await pipeline("translation", "OptimaMT/th-en", {
35
+ dtype: "q8",
36
+ session_options: { extra: { session: { disable_quant_qdq: "1" } } },
37
+ });
38
+ const [out] = await translator("ประภาคารตั้งอยู่บนชายฝั่ง");
39
+ console.log(out.translation_text);
40
+ ```
41
+
42
+ Python ([translate.py](https://github.com/AmerAlreyahi/OptimaMT/blob/main/scripts/translate.py): onnxruntime, sentencepiece, numpy):
43
+
44
+ ```bash
45
+ python translate.py OptimaMT/th-en "ประภาคารตั้งอยู่บนชายฝั่ง"
46
+ ```
47
+
48
+ ## Files
49
+
50
+ Only `onnx/decoder_model_merged_quantized.onnx` is rebuilt; every other file is the original, byte for byte.
51
+ Vocabulary: 62,307 tokens.
52
+
53
+ ## Licence and credits
54
+
55
+ - Weights: [OPUS-MT](https://huggingface.co/Helsinki-NLP/opus-mt-th-en) by the University of Helsinki, CC BY 4.0.
56
+ - ONNX export: [Xenova/opus-mt-th-en](https://huggingface.co/Xenova/opus-mt-th-en) at revision `7f0406b972f7`.
57
+ - Modified: the decoder's output projection was moved out of its two branches (graph change only). CC BY 4.0.
config.json ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "Helsinki-NLP/opus-mt-th-en",
3
+ "activation_dropout": 0.0,
4
+ "activation_function": "swish",
5
+ "add_bias_logits": false,
6
+ "add_final_layer_norm": false,
7
+ "architectures": [
8
+ "MarianMTModel"
9
+ ],
10
+ "attention_dropout": 0.0,
11
+ "bad_words_ids": [
12
+ [
13
+ 62306
14
+ ]
15
+ ],
16
+ "bos_token_id": 0,
17
+ "classif_dropout": 0.0,
18
+ "classifier_dropout": 0.0,
19
+ "d_model": 512,
20
+ "decoder_attention_heads": 8,
21
+ "decoder_ffn_dim": 2048,
22
+ "decoder_layerdrop": 0.0,
23
+ "decoder_layers": 6,
24
+ "decoder_start_token_id": 62306,
25
+ "decoder_vocab_size": 62307,
26
+ "dropout": 0.1,
27
+ "encoder_attention_heads": 8,
28
+ "encoder_ffn_dim": 2048,
29
+ "encoder_layerdrop": 0.0,
30
+ "encoder_layers": 6,
31
+ "eos_token_id": 0,
32
+ "extra_pos_embeddings": 62307,
33
+ "forced_eos_token_id": 0,
34
+ "id2label": {
35
+ "0": "LABEL_0",
36
+ "1": "LABEL_1",
37
+ "2": "LABEL_2"
38
+ },
39
+ "init_std": 0.02,
40
+ "is_encoder_decoder": true,
41
+ "label2id": {
42
+ "LABEL_0": 0,
43
+ "LABEL_1": 1,
44
+ "LABEL_2": 2
45
+ },
46
+ "max_length": 512,
47
+ "max_position_embeddings": 512,
48
+ "model_type": "marian",
49
+ "normalize_before": false,
50
+ "normalize_embedding": false,
51
+ "num_beams": 6,
52
+ "num_hidden_layers": 6,
53
+ "pad_token_id": 62306,
54
+ "scale_embedding": true,
55
+ "share_encoder_decoder_embeddings": true,
56
+ "static_position_embeddings": true,
57
+ "transformers_version": "4.34.0.dev0",
58
+ "use_cache": true,
59
+ "vocab_size": 62307
60
+ }
generation_config.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bad_words_ids": [
3
+ [
4
+ 62306
5
+ ]
6
+ ],
7
+ "bos_token_id": 0,
8
+ "decoder_start_token_id": 62306,
9
+ "eos_token_id": 0,
10
+ "forced_eos_token_id": 0,
11
+ "max_length": 512,
12
+ "num_beams": 6,
13
+ "pad_token_id": 62306,
14
+ "renormalize_logits": true,
15
+ "transformers_version": "4.34.0.dev0"
16
+ }
onnx/decoder_model_merged_quantized.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d47e46270210fdb53dcb98f14d5b9ebea3e297145f9604924de44e5bda2613fb
3
+ size 58822368
onnx/encoder_model_quantized.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:37e31edabcadfca300cf13d4dd20087aa1a920cba7b20c738cba8ff44f320bbf
3
+ size 51520414
quantize_config.json ADDED
@@ -0,0 +1,125 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "per_channel": true,
3
+ "reduce_range": true,
4
+ "per_model_config": {
5
+ "encoder_model": {
6
+ "op_types": [
7
+ "Sqrt",
8
+ "Expand",
9
+ "MatMul",
10
+ "Sub",
11
+ "Sigmoid",
12
+ "Range",
13
+ "Constant",
14
+ "Pow",
15
+ "Mul",
16
+ "Concat",
17
+ "Shape",
18
+ "Div",
19
+ "Reshape",
20
+ "Cast",
21
+ "Where",
22
+ "Equal",
23
+ "Unsqueeze",
24
+ "Transpose",
25
+ "ReduceMean",
26
+ "Softmax",
27
+ "ConstantOfShape",
28
+ "Gather",
29
+ "Add"
30
+ ],
31
+ "weight_type": "QInt8"
32
+ },
33
+ "decoder_with_past_model": {
34
+ "op_types": [
35
+ "Sqrt",
36
+ "Expand",
37
+ "MatMul",
38
+ "Sub",
39
+ "Sigmoid",
40
+ "Range",
41
+ "Constant",
42
+ "Pow",
43
+ "Mul",
44
+ "Concat",
45
+ "Shape",
46
+ "Div",
47
+ "Reshape",
48
+ "Cast",
49
+ "Where",
50
+ "Equal",
51
+ "Unsqueeze",
52
+ "Transpose",
53
+ "ReduceMean",
54
+ "Softmax",
55
+ "ConstantOfShape",
56
+ "Gather",
57
+ "Add"
58
+ ],
59
+ "weight_type": "QInt8"
60
+ },
61
+ "decoder_model_merged": {
62
+ "op_types": [
63
+ "Sqrt",
64
+ "Expand",
65
+ "Less",
66
+ "MatMul",
67
+ "Sub",
68
+ "Sigmoid",
69
+ "Range",
70
+ "Constant",
71
+ "Pow",
72
+ "Mul",
73
+ "Concat",
74
+ "Shape",
75
+ "Div",
76
+ "Reshape",
77
+ "Cast",
78
+ "If",
79
+ "Where",
80
+ "Equal",
81
+ "Unsqueeze",
82
+ "Slice",
83
+ "Squeeze",
84
+ "Transpose",
85
+ "ReduceMean",
86
+ "Softmax",
87
+ "ConstantOfShape",
88
+ "Gather",
89
+ "Add"
90
+ ],
91
+ "weight_type": "QInt8"
92
+ },
93
+ "decoder_model": {
94
+ "op_types": [
95
+ "Sqrt",
96
+ "Expand",
97
+ "Less",
98
+ "MatMul",
99
+ "Sub",
100
+ "Sigmoid",
101
+ "Range",
102
+ "Constant",
103
+ "Pow",
104
+ "Mul",
105
+ "Concat",
106
+ "Shape",
107
+ "Div",
108
+ "Reshape",
109
+ "Cast",
110
+ "Where",
111
+ "Equal",
112
+ "Unsqueeze",
113
+ "Slice",
114
+ "Squeeze",
115
+ "Transpose",
116
+ "ReduceMean",
117
+ "Softmax",
118
+ "ConstantOfShape",
119
+ "Gather",
120
+ "Add"
121
+ ],
122
+ "weight_type": "QInt8"
123
+ }
124
+ }
125
+ }
source.spm ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:01ecb88d0f92bbb20c21257b92e93e301a10971ecd60971e8e637bc3eacd9124
3
+ size 1213743
special_tokens_map.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "eos_token": "</s>",
3
+ "pad_token": "<pad>",
4
+ "unk_token": "<unk>"
5
+ }
target.spm ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc618401104178afa0e8d67d469a49f8e71cd3151c51d640a7003a1f71b16b1e
3
+ size 810346
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "clean_up_tokenization_spaces": true,
3
+ "eos_token": "</s>",
4
+ "model_max_length": 512,
5
+ "pad_token": "<pad>",
6
+ "separate_vocabs": false,
7
+ "source_lang": "tha",
8
+ "sp_model_kwargs": {},
9
+ "target_lang": "eng",
10
+ "tokenizer_class": "MarianTokenizer",
11
+ "unk_token": "<unk>"
12
+ }
vocab.json ADDED
The diff for this file is too large to render. See raw diff