AmerAlreyahi commited on
Commit
135a9eb
·
verified ·
1 Parent(s): 03ab450

Add OptimaMT it-fr

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
+ - it
5
+ - fr
6
+ pipeline_tag: translation
7
+ library_name: transformers.js
8
+ base_model: Xenova/opus-mt-it-fr
9
+ tags:
10
+ - translation
11
+ - marian
12
+ - opus-mt
13
+ - onnx
14
+ ---
15
+
16
+ # OptimaMT it → fr (Italian to French)
17
+
18
+ Drop-in copy of [Xenova/opus-mt-it-fr](https://huggingface.co/Xenova/opus-mt-it-fr) at revision `7843e3b7137d` 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` | 503 → 272 MiB (-46 %) | 481 → 249 MiB (-48 %) | yes |
24
+ | ONNX Runtime defaults | 270 → 212 MiB (-21 %) | 191 → 162 MiB (-15 %) | 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/it-fr", {
35
+ dtype: "q8",
36
+ session_options: { extra: { session: { disable_quant_qdq: "1" } } },
37
+ });
38
+ const [out] = await translator("Il faro si trova sulla costa.");
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/it-fr "Il faro si trova sulla costa."
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: 59,351 tokens.
52
+
53
+ ## Licence and credits
54
+
55
+ - Weights: [OPUS-MT](https://huggingface.co/Helsinki-NLP/opus-mt-it-fr) by the University of Helsinki, CC BY 4.0.
56
+ - ONNX export: [Xenova/opus-mt-it-fr](https://huggingface.co/Xenova/opus-mt-it-fr) at revision `7843e3b7137d`.
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-it-fr",
3
+ "_num_labels": 3,
4
+ "activation_dropout": 0.0,
5
+ "activation_function": "swish",
6
+ "add_bias_logits": false,
7
+ "add_final_layer_norm": false,
8
+ "architectures": [
9
+ "MarianMTModel"
10
+ ],
11
+ "attention_dropout": 0.0,
12
+ "bad_words_ids": [
13
+ [
14
+ 59350
15
+ ]
16
+ ],
17
+ "bos_token_id": 0,
18
+ "classif_dropout": 0.0,
19
+ "classifier_dropout": 0.0,
20
+ "d_model": 512,
21
+ "decoder_attention_heads": 8,
22
+ "decoder_ffn_dim": 2048,
23
+ "decoder_layerdrop": 0.0,
24
+ "decoder_layers": 6,
25
+ "decoder_start_token_id": 59350,
26
+ "decoder_vocab_size": 59351,
27
+ "dropout": 0.1,
28
+ "encoder_attention_heads": 8,
29
+ "encoder_ffn_dim": 2048,
30
+ "encoder_layerdrop": 0.0,
31
+ "encoder_layers": 6,
32
+ "eos_token_id": 0,
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": 59350,
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": 59351
60
+ }
generation_config.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bad_words_ids": [
3
+ [
4
+ 59350
5
+ ]
6
+ ],
7
+ "bos_token_id": 0,
8
+ "decoder_start_token_id": 59350,
9
+ "eos_token_id": 0,
10
+ "forced_eos_token_id": 0,
11
+ "max_length": 512,
12
+ "num_beams": 6,
13
+ "pad_token_id": 59350,
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:fe60a7e329f527bb4f0c1ac904355356a7881ba194a59b4c66c50dabac182ff1
3
+ size 57297072
onnx/encoder_model_quantized.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:02efa49882a03300453fa6816ebc965baff791b280b2d03ac93316739768dd19
3
+ size 50006942
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
+ "ConstantOfShape",
8
+ "Where",
9
+ "Shape",
10
+ "Expand",
11
+ "Pow",
12
+ "Add",
13
+ "Gather",
14
+ "Sigmoid",
15
+ "Concat",
16
+ "Sqrt",
17
+ "ReduceMean",
18
+ "Cast",
19
+ "Equal",
20
+ "Sub",
21
+ "Softmax",
22
+ "Div",
23
+ "Mul",
24
+ "Reshape",
25
+ "Constant",
26
+ "Transpose",
27
+ "Unsqueeze",
28
+ "Range",
29
+ "MatMul"
30
+ ],
31
+ "weight_type": "QInt8"
32
+ },
33
+ "decoder_with_past_model": {
34
+ "op_types": [
35
+ "ConstantOfShape",
36
+ "Where",
37
+ "Shape",
38
+ "Expand",
39
+ "Pow",
40
+ "Add",
41
+ "Gather",
42
+ "Sigmoid",
43
+ "Concat",
44
+ "Sqrt",
45
+ "ReduceMean",
46
+ "Equal",
47
+ "Cast",
48
+ "Sub",
49
+ "Softmax",
50
+ "Div",
51
+ "Mul",
52
+ "Reshape",
53
+ "Constant",
54
+ "Transpose",
55
+ "Unsqueeze",
56
+ "Range",
57
+ "MatMul"
58
+ ],
59
+ "weight_type": "QInt8"
60
+ },
61
+ "decoder_model_merged": {
62
+ "op_types": [
63
+ "Less",
64
+ "ConstantOfShape",
65
+ "Where",
66
+ "Shape",
67
+ "Expand",
68
+ "Pow",
69
+ "Add",
70
+ "Gather",
71
+ "Sigmoid",
72
+ "Concat",
73
+ "Sqrt",
74
+ "ReduceMean",
75
+ "Squeeze",
76
+ "If",
77
+ "Cast",
78
+ "Equal",
79
+ "Sub",
80
+ "Softmax",
81
+ "Div",
82
+ "Mul",
83
+ "Reshape",
84
+ "Constant",
85
+ "Slice",
86
+ "Transpose",
87
+ "Unsqueeze",
88
+ "Range",
89
+ "MatMul"
90
+ ],
91
+ "weight_type": "QInt8"
92
+ },
93
+ "decoder_model": {
94
+ "op_types": [
95
+ "Less",
96
+ "ConstantOfShape",
97
+ "Where",
98
+ "Shape",
99
+ "Expand",
100
+ "Pow",
101
+ "Add",
102
+ "Gather",
103
+ "Sigmoid",
104
+ "Concat",
105
+ "Sqrt",
106
+ "ReduceMean",
107
+ "Squeeze",
108
+ "Cast",
109
+ "Equal",
110
+ "Sub",
111
+ "Softmax",
112
+ "Div",
113
+ "Mul",
114
+ "Reshape",
115
+ "Constant",
116
+ "Slice",
117
+ "Transpose",
118
+ "Unsqueeze",
119
+ "Range",
120
+ "MatMul"
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 sha256:7bdce190b91dcd93a386b7f584dbed229fa78a89d16ed0742ad1f348d9082005
3
+ size 813967
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:d542c26fcaf398aa4333e92ac391c2a744586a1b1a19cfa9f557064d0555910f
3
+ size 813389
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": "it",
8
+ "sp_model_kwargs": {},
9
+ "target_lang": "fr",
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