Hugging Face's logo

gomyk
/
minilm-student-L4_uniform_distilled

Sentence Similarity
sentence-transformers
Safetensors
bert
multilingual
layer-pruning
vocab-pruning
knowledge-distillation
minilm-l12
text-embeddings-inference
Model card Files
xet
 Community
gomyk commited on
Commit
d5fc3ce
·
verified ·
1 Parent(s): 3587adb

Upload L4_uniform_distilled (distilled) from MiniLM-L12

Browse files
Files changed (10) hide show
  1. 1_Pooling/config.json +10 -0
  2. README.md +125 -0
  3. config.json +25 -0
  4. config_sentence_transformers.json +14 -0
  5. model.safetensors +3 -0
  6. modules.json +14 -0
  7. sentence_bert_config.json +4 -0
  8. special_tokens_map.json +51 -0
  9. tokenizer.json +0 -0
  10. tokenizer_config.json +65 -0
1_Pooling/config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "word_embedding_dimension": 384,
3
+ "pooling_mode_cls_token": false,
4
+ "pooling_mode_mean_tokens": true,
5
+ "pooling_mode_max_tokens": false,
6
+ "pooling_mode_mean_sqrt_len_tokens": false,
7
+ "pooling_mode_weightedmean_tokens": false,
8
+ "pooling_mode_lasttoken": false,
9
+ "include_prompt": true
10
+ }
README.md ADDED
@@ -0,0 +1,125 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: ["ko", "en", "ja", "zh", "es", "fr", "de", "pt", "it", "ru", "ar", "hi", "th", "vi", "id", "tr", "nl", "pl"]
3
+ tags:
4
+ - sentence-transformers
5
+ - multilingual
6
+ - layer-pruning
7
+ - vocab-pruning
8
+ - knowledge-distillation
9
+ - minilm-l12
10
+ library_name: sentence-transformers
11
+ pipeline_tag: sentence-similarity
12
+ license: apache-2.0
13
+ ---
14
+
15
+ # L4_uniform_distilled (Distilled)
16
+
17
+ Lightweight sentence encoder created from `sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2` via layer pruning + vocabulary pruning + knowledge distillation.
18
+
19
+ ## Model Details
20
+
21
+ | Property | Value |
22
+ |---|---|
23
+ | Teacher | sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2 |
24
+ | Architecture | MiniLM-L12 (pruned) |
25
+ | Hidden dim | 384 |
26
+ | Layers | 4 / 12 |
27
+ | Layer indices | [0, 4, 7, 11] |
28
+ | Strategy | 4 layers, evenly spaced (compact) |
29
+ | Parameters | 103,283,328 |
30
+ | Model size (FP32) | 84.6MB |
31
+ | Distilled | Yes |
32
+
33
+ ## Architecture
34
+
35
+ ```
36
+ ==============================================================
37
+ TEACHER: MiniLM-L12 → STUDENT: 4L / 38,755 vocab
38
+ ==============================================================
39
+
40
+ TEACHER STUDENT
41
+ ─────────────────────────── ───────────────────────────
42
+
43
+ ┌─────────────────────────┐ ┌─────────────────────────┐
44
+ │ Input Tokens │ │ Input Tokens │
45
+ └────────────┬────────────┘ └────────────┬────────────┘
46
+ │ │
47
+ ┌────────────┴────────────┐ ┌────────────┴────────────┐
48
+ │ Embeddings │ │ Embeddings (pruned) │
49
+ │ vocab: 250,002 │ │ vocab: 38,755 │
50
+ │ dim: 384 │ │ dim: 384 │
51
+ └────────────┬────────────┘ └────────────┬────────────┘
52
+ │ │
53
+ ┌─────────────────────────┐ ┌─────────────────────────┐
54
+ │ Layer 0 │ ──► │ Layer 0 ← L0 │
55
+ ├─────────────────────────┤ ├─────────────────────────┤
56
+ │ Layer 1 │ ╳ │ │
57
+ ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤ │ │
58
+ │ Layer 2 │ ╳ │ │
59
+ ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤ │ │
60
+ │ Layer 3 │ ╳ │ │
61
+ ├─────────────────────────┤ ├─────────────────────────┤
62
+ │ Layer 4 │ ──► │ Layer 1 ← L4 │
63
+ ├─────────────────────────┤ ├─────────────────────────┤
64
+ │ Layer 5 │ ╳ │ │
65
+ ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤ │ │
66
+ │ Layer 6 │ ╳ │ │
67
+ ├─────────────────────────┤ ├─────────────────────────┤
68
+ │ Layer 7 │ ──► │ Layer 2 ← L7 │
69
+ ├─────────────────────────┤ ├─────────────────────────┤
70
+ │ Layer 8 │ ╳ │ │
71
+ ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤ │ │
72
+ │ Layer 9 │ ╳ │ │
73
+ ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤ │ │
74
+ │ Layer 10 │ ╳ │ │
75
+ ├─────────────────────────┤ ├─────────────────────────┤
76
+ │ Layer 11 │ ──► │ Layer 3 ← L11 │
77
+ └─────────��──┬────────────┘ └────────────┬────────────┘
78
+ │ │
79
+ ┌────────────┴────────────┐ ┌────────────┴────────────┐
80
+ │ Mean Pooling │ │ Mean Pooling │
81
+ │ → 384d embedding │ │ → 384d embedding │
82
+ └─────────────────────────┘ └─────────────────────────┘
83
+
84
+ Size: 448.0MB (FP32) → 84.6MB (FP32)
85
+ Params: 117,451,392 → 22,164,480
86
+ Reduction: 81.1%
87
+ ==============================================================
88
+ ```
89
+
90
+ ## Quick Start
91
+
92
+ ```python
93
+ from sentence_transformers import SentenceTransformer
94
+
95
+ model = SentenceTransformer("L4_uniform_distilled", trust_remote_code=True)
96
+
97
+ sentences = [
98
+ "Hello, how are you?",
99
+ "안녕하세요",
100
+ "Bonjour, comment allez-vous?",
101
+ ]
102
+
103
+ embeddings = model.encode(sentences)
104
+ print(embeddings.shape) # (3, 384)
105
+ ```
106
+
107
+
108
+
109
+ ## Training
110
+
111
+ ### Stage 1: Layer Pruning
112
+ - Teacher: `sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2` (12 layers, 384d)
113
+ - Selected layers: `[0, 4, 7, 11]` (4 layers, evenly spaced (compact))
114
+ - Vocabulary pruning applied
115
+
116
+ ### Stage 2: Knowledge Distillation
117
+ - **Method**: MSE + Cosine Similarity loss
118
+ - **Data**: MTEB Classification/Clustering/STS task datasets
119
+ - **Optimizer**: AdamW (lr=2e-5, weight_decay=0.01)
120
+ - **Schedule**: Cosine annealing over 3 epochs
121
+
122
+
123
+ ## Supported Languages (18)
124
+
125
+ ko, en, ja, zh, es, fr, de, pt, it, ru, ar, hi, th, vi, id, tr, nl, pl
config.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "BertModel"
4
+ ],
5
+ "attention_probs_dropout_prob": 0.1,
6
+ "classifier_dropout": null,
7
+ "dtype": "float32",
8
+ "gradient_checkpointing": false,
9
+ "hidden_act": "gelu",
10
+ "hidden_dropout_prob": 0.1,
11
+ "hidden_size": 384,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 1536,
14
+ "layer_norm_eps": 1e-12,
15
+ "max_position_embeddings": 512,
16
+ "model_type": "bert",
17
+ "num_attention_heads": 12,
18
+ "num_hidden_layers": 4,
19
+ "pad_token_id": 0,
20
+ "position_embedding_type": "absolute",
21
+ "transformers_version": "4.56.2",
22
+ "type_vocab_size": 2,
23
+ "use_cache": true,
24
+ "vocab_size": 38330
25
+ }
config_sentence_transformers.json ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "SentenceTransformer",
3
+ "__version__": {
4
+ "sentence_transformers": "5.3.0",
5
+ "transformers": "4.56.2",
6
+ "pytorch": "2.10.0+cu128"
7
+ },
8
+ "prompts": {
9
+ "query": "",
10
+ "document": ""
11
+ },
12
+ "default_prompt_name": null,
13
+ "similarity_fn_name": "cosine"
14
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:849314db7a40fefcafc751585b9a2004108365166b922a06d5c979451da025e4
3
+ size 88658024
modules.json ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "idx": 0,
4
+ "name": "0",
5
+ "path": "",
6
+ "type": "sentence_transformers.models.Transformer"
7
+ },
8
+ {
9
+ "idx": 1,
10
+ "name": "1",
11
+ "path": "1_Pooling",
12
+ "type": "sentence_transformers.models.Pooling"
13
+ }
14
+ ]
sentence_bert_config.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "max_seq_length": 128,
3
+ "do_lower_case": false
4
+ }
special_tokens_map.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "cls_token": {
10
+ "content": "<s>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "eos_token": {
17
+ "content": "</s>",
18
+ "lstrip": false,
19
+ "normalized": false,
20
+ "rstrip": false,
21
+ "single_word": false
22
+ },
23
+ "mask_token": {
24
+ "content": "<mask>",
25
+ "lstrip": true,
26
+ "normalized": false,
27
+ "rstrip": false,
28
+ "single_word": false
29
+ },
30
+ "pad_token": {
31
+ "content": "<pad>",
32
+ "lstrip": false,
33
+ "normalized": false,
34
+ "rstrip": false,
35
+ "single_word": false
36
+ },
37
+ "sep_token": {
38
+ "content": "</s>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false
43
+ },
44
+ "unk_token": {
45
+ "content": "<unk>",
46
+ "lstrip": false,
47
+ "normalized": false,
48
+ "rstrip": false,
49
+ "single_word": false
50
+ }
51
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "added_tokens_decoder": {
3
+ "0": {
4
+ "content": "<s>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false,
9
+ "special": true
10
+ },
11
+ "1": {
12
+ "content": "<pad>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false,
17
+ "special": true
18
+ },
19
+ "2": {
20
+ "content": "</s>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false,
25
+ "special": true
26
+ },
27
+ "3": {
28
+ "content": "<unk>",
29
+ "lstrip": false,
30
+ "normalized": false,
31
+ "rstrip": false,
32
+ "single_word": false,
33
+ "special": true
34
+ },
35
+ "38329": {
36
+ "content": "<mask>",
37
+ "lstrip": true,
38
+ "normalized": false,
39
+ "rstrip": false,
40
+ "single_word": false,
41
+ "special": true
42
+ }
43
+ },
44
+ "bos_token": "<s>",
45
+ "clean_up_tokenization_spaces": false,
46
+ "cls_token": "<s>",
47
+ "do_lower_case": true,
48
+ "eos_token": "</s>",
49
+ "extra_special_tokens": {},
50
+ "mask_token": "<mask>",
51
+ "max_length": 128,
52
+ "model_max_length": 128,
53
+ "pad_to_multiple_of": null,
54
+ "pad_token": "<pad>",
55
+ "pad_token_type_id": 0,
56
+ "padding_side": "right",
57
+ "sep_token": "</s>",
58
+ "stride": 0,
59
+ "strip_accents": null,
60
+ "tokenize_chinese_chars": true,
61
+ "tokenizer_class": "PreTrainedTokenizerFast",
62
+ "truncation_side": "right",
63
+ "truncation_strategy": "longest_first",
64
+ "unk_token": "<unk>"
65
+ }