Hugging Face's logo

KIEFERSA
/
Sophea-Nemo-Embedding

Sentence Similarity
sentence-transformers
Safetensors
Greek
English
ministral3
greek
english
retrieval
rag
embeddings
nemotron
Eval Results (legacy)
Model card Files
xet
 Community
ayoubkirouane commited on
Commit
e5b0973
·
0 Parent(s):

Initial release

Browse files
Files changed (12) hide show
  1. .gitattributes +36 -0
  2. 1_Pooling/config.json +5 -0
  3. README.md +177 -0
  4. base_vs_ft.png +0 -0
  5. config.json +46 -0
  6. config_sentence_transformers.json +21 -0
  7. model.safetensors +3 -0
  8. modules.json +20 -0
  9. sentence_bert_config.json +10 -0
  10. stack.png +0 -0
  11. tokenizer.json +3 -0
  12. tokenizer_config.json +1014 -0
.gitattributes ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
+ *.model filter=lfs diff=lfs merge=lfs -text
13
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
14
+ *.npy filter=lfs diff=lfs merge=lfs -text
15
+ *.npz filter=lfs diff=lfs merge=lfs -text
16
+ *.onnx filter=lfs diff=lfs merge=lfs -text
17
+ *.ot filter=lfs diff=lfs merge=lfs -text
18
+ *.parquet filter=lfs diff=lfs merge=lfs -text
19
+ *.pb filter=lfs diff=lfs merge=lfs -text
20
+ *.pickle filter=lfs diff=lfs merge=lfs -text
21
+ *.pkl filter=lfs diff=lfs merge=lfs -text
22
+ *.pt filter=lfs diff=lfs merge=lfs -text
23
+ *.pth filter=lfs diff=lfs merge=lfs -text
24
+ *.rar filter=lfs diff=lfs merge=lfs -text
25
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
26
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar filter=lfs diff=lfs merge=lfs -text
29
+ *.tflite filter=lfs diff=lfs merge=lfs -text
30
+ *.tgz filter=lfs diff=lfs merge=lfs -text
31
+ *.wasm filter=lfs diff=lfs merge=lfs -text
32
+ *.xz 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
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
1_Pooling/config.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "embedding_dimension": 2048,
3
+ "pooling_mode": "mean",
4
+ "include_prompt": true
5
+ }
README.md ADDED
@@ -0,0 +1,177 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - el
4
+ - en
5
+ license: other
6
+ license_name: openmdw-1.1
7
+ base_model: nvidia/Nemotron-3-Embed-1B-BF16
8
+ pipeline_tag: sentence-similarity
9
+ library_name: sentence-transformers
10
+ tags:
11
+ - greek
12
+ - english
13
+ - retrieval
14
+ - rag
15
+ - embeddings
16
+ - sentence-transformers
17
+ - nemotron
18
+ model-index:
19
+ - name: Sophea-Nemo-Embedding
20
+ results:
21
+ - task:
22
+ type: sentence-similarity
23
+ name: Greek+English retrieval (5 domains, cosine IR)
24
+ dataset:
25
+ type: retrieval
26
+ name: Internal Greek+English retrieval benchmark
27
+ metrics:
28
+ - type: nDCG@10
29
+ name: nDCG@10 (mean of 5 domains)
30
+ value: 0.8176
31
+ - type: recall@10
32
+ name: Recall@10 (mean of 5 domains)
33
+ value: 0.9310
34
+ ---
35
+
36
+ <div style="display:flex;align-items:center;gap:18px;padding:16px 20px;margin-bottom:16px;border-radius:14px;background:linear-gradient(100deg,#e8f5ec 0%,#f7fbf8 62%);border:1px solid #c6e7d1"><img alt="KIEFERSA" style="flex:0 0 auto;height:58px;width:58px;border-radius:12px;object-fit:contain" src="data:image/webp;base64,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"/><div><div style="font-size:1.24rem;font-weight:700;color:#003713;line-height:1.2">Sophea-Nemo-Embedding</div><div style="font-size:.86rem;color:#3a5a47;margin-top:2px">Greek + English dense retriever — fine-tuned NVIDIA Nemotron-3-Embed-1B</div></div></div>
37
+
38
+ **Sophea-Nemo-Embedding** is a Greek + English **dense retrieval embedder**, fine-tuned from
39
+ **[NVIDIA Nemotron-3-Embed-1B](https://huggingface.co/nvidia/Nemotron-3-Embed-1B-BF16)** for Greek and
40
+ English retrieval. It is the first stage of a two-stage Greek RAG
41
+ stack; the matching second stage is
42
+ **[Sophea-Nemo-Reranker](https://huggingface.co/KIEFERSA/Sophea-Nemo-Reranker)**.
43
+
44
+ - **Creator:** Kiefer SA
45
+ - **Base model:** `nvidia/Nemotron-3-Embed-1B-BF16` (Ministral-3 backbone, **bidirectional**, mean
46
+ pooling with `include_prompt=true`, 2048-d Matryoshka-sliceable, L2-normalized)
47
+ - **Languages:** Greek + English
48
+ - **Prompts:** `query: ` on queries, `passage: ` on documents (both sides prompted)
49
+ - **Training:** full fine-tune, InfoNCE (in-batch + 7 hard negatives), 1 epoch, max-len 4096, bf16
50
+
51
+ > **Greek is out-of-distribution for the base.** Nemotron-3-Embed lists 34 languages and **Greek is
52
+ > not one of them** — which is exactly why the base retrieves poorly on Greek and why this fine-tune
53
+ > exists.
54
+
55
+ ## Why it matters — base vs fine-tuned
56
+
57
+ On our held-out 5-domain Greek+English retrieval eval (cosine IR, `query: `/`passage: ` prompts), the
58
+ fine-tune roughly **2.5×'s** the base embedder's retrieval quality — on **every** domain.
59
+
60
+ ![Greek+English retrieval — base vs fine-tuned](base_vs_ft.png)
61
+
62
+ <div style="overflow-x:auto">
63
+ <table style="width:100%;border-collapse:collapse;font-size:13px;font-variant-numeric:tabular-nums;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif">
64
+ <thead><tr>
65
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:left;">metric</th>
66
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Val</th>
67
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Energy</th>
68
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Legal</th>
69
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Finance</th>
70
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Medical</th>
71
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Mean</th>
72
+ </tr></thead><tbody>
73
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">nDCG@10 — base</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.382</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.345</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.526</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.194</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.209</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.331</td></tr>
74
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">nDCG@10 — fine-tuned</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.875</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.794</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.950</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.722</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.747</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.818</td></tr>
75
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Δ nDCG@10</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.493</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.448</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.424</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.528</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.538</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.486</td></tr>
76
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Recall@10 — base</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.465</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.464</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.599</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.283</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.314</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.425</td></tr>
77
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">Recall@10 — fine-tuned</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.963</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.898</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.984</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.903</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.906</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.931</td></tr>
78
+ </tbody></table></div>
79
+
80
+ **Read.** Mean **nDCG@10 0.331 → 0.818
81
+ (+0.486)**, with the largest gains on the hardest
82
+ domains (finance +0.53, medical +0.54). Recall@10 climbs from a mean 0.425
83
+ to **0.931** — i.e. the right passage is now almost always in the top-10.
84
+
85
+ ## Size comparison — vs base Qwen3-Embedding (off-the-shelf)
86
+
87
+ Same 5-domain retrieval eval, same corpus/prompts. Your **fine-tuned 1B** embedder beats the
88
+ **base Qwen3-Embedding-8B by a wide margin** (nDCG@10):
89
+
90
+ <div style="overflow-x:auto">
91
+ <table style="width:100%;border-collapse:collapse;font-size:13px;font-variant-numeric:tabular-nums;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif">
92
+ <thead><tr>
93
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:left;">model</th>
94
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Val</th>
95
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Energy</th>
96
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Legal</th>
97
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Finance</th>
98
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Medical</th>
99
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Mean</th>
100
+ </tr></thead><tbody>
101
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">Sophea-Nemo-Embedding (ours · 1B · fine-tuned)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.875</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.794</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.950</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.722</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.747</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.818</td></tr>
102
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Qwen3-Embedding-8B (base, off-the-shelf)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.740</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.580</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.910</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.490</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.602</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.664</td></tr>
103
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Qwen3-Embedding-4B (base, off-the-shelf)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.738</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.579</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.919</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.501</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.586</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.665</td></tr>
104
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Qwen3-Embedding-0.6B (base, off-the-shelf)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.658</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.509</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.863</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.426</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.467</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.585</td></tr>
105
+ </tbody></table></div>
106
+
107
+ **Read.** Mean nDCG@10 **0.818** (ours, 1B) vs **0.664** (8B),
108
+ **0.665** (4B), **0.585** (0.6B). Note the Qwen 8B ≈ 4B —
109
+ scaling the off-the-shelf model barely helps on Greek, whereas a 1B **fine-tuned** on-domain wins by
110
+ **+0.153** over the 8×-larger model.
111
+
112
+ ## End-to-end two-stage pipeline
113
+
114
+ The two models are designed to run **together**: the embedder retrieves the top-50 candidates, the
115
+ reranker re-orders them. Measured end-to-end on the same 5-domain Greek+English eval (150 queries/set,
116
+ correct `query:`/`passage:` prompts):
117
+
118
+ ![End-to-end two-stage stack — embedder alone vs + reranker](stack.png)
119
+
120
+ <div style="overflow-x:auto">
121
+ <table style="width:100%;border-collapse:collapse;font-size:13px;font-variant-numeric:tabular-nums;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif">
122
+ <thead><tr>
123
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:left;">metric</th>
124
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Val</th>
125
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Energy</th>
126
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Legal</th>
127
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Finance</th>
128
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Medical</th>
129
+ <th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Mean</th>
130
+ </tr></thead><tbody>
131
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">nDCG@10 — embedder alone</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.863</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.802</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.939</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.702</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.689</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.799</td></tr>
132
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">nDCG@10 — + reranker (stack)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.867</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.780</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.976</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.763</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.703</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.818</td></tr>
133
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Δ from reranking</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.004</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">-0.022</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.037</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.061</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.013</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.019</td></tr>
134
+ <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">Recall@10 — full stack</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.987</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.873</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.993</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.920</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.860</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.927</td></tr>
135
+ </tbody></table></div>
136
+
137
+ **Read.** The full stack reaches **mean nDCG@10 0.818** (Recall@10 0.927) — the
138
+ strongest configuration measured. The reranker adds the most where the embedder is weakest
139
+ (**finance +0.061, legal +0.037**), i.e. it sharpens precision on the hard
140
+ cases. On **energy** it slightly regresses (-0.022) and on `val` the gain is marginal,
141
+ because the embedder alone is already very strong there and leaves little headroom — for the easiest,
142
+ already-high-recall domains you can skip the rerank stage if latency matters.
143
+
144
+
145
+ ## Usage (sentence-transformers)
146
+
147
+ ```python
148
+ from sentence_transformers import SentenceTransformer
149
+ m = SentenceTransformer("KIEFERSA/Sophea-Nemo-Embedding")
150
+ q = m.encode(["Ποια είναι η πρωτεύουσα της Ελλάδας;"], prompt="query: ", normalize_embeddings=True)
151
+ d = m.encode(["Η Αθήνα είναι η πρωτεύουσα της Ελλάδας."], prompt="passage: ", normalize_embeddings=True)
152
+ print((q @ d.T)) # cosine similarity
153
+ ```
154
+
155
+ Both sides **must** be prompted (`query: ` / `passage: `) — the base uses `include_prompt=true`, so
156
+ dropping the prefix is a distribution mismatch. Embeddings are L2-normalized (use cosine / dot).
157
+
158
+ ## Serving with vLLM
159
+
160
+ ```bash
161
+ vllm serve KIEFERSA/Sophea-Nemo-Embedding --runner pooling
162
+ ```
163
+
164
+ Exposes an OpenAI-compatible `/v1/embeddings` endpoint. vLLM does **not** auto-apply the model's ST
165
+ prompts, so prepend them **client-side** — `query: ` on queries, `passage: ` on documents:
166
+
167
+ ```python
168
+ from openai import OpenAI
169
+ c = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
170
+ name = "KIEFERSA/Sophea-Nemo-Embedding"
171
+ q = c.embeddings.create(model=name, input=["query: Ποια είναι η πρωτεύουσα της Ελλάδας;"]).data[0].embedding
172
+ d = c.embeddings.create(model=name, input=["passage: Η Αθήνα είναι η πρωτεύουσα της Ελλάδας."]).data[0].embedding
173
+ ```
174
+
175
+ ## License
176
+ Fine-tune of `nvidia/Nemotron-3-Embed-1B-BF16` — governed by **OpenMDW-1.1** (base Ministral-3 is
177
+ Apache-2.0); commercial use permitted under those terms. Review the base model's license before use.
base_vs_ft.png ADDED
config.json ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Ministral3Model"
4
+ ],
5
+ "attention_dropout": 0.0,
6
+ "bos_token_id": 1,
7
+ "dtype": "bfloat16",
8
+ "eos_token_id": 2,
9
+ "head_dim": 128,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 2048,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 6144,
14
+ "is_causal": false,
15
+ "llama_4_scaling": {
16
+ "beta": 0.1,
17
+ "original_max_position_embeddings": 16384
18
+ },
19
+ "max_position_embeddings": 262144,
20
+ "model_type": "ministral3",
21
+ "nemo_version": "0.3.0rc0",
22
+ "num_attention_heads": 24,
23
+ "num_hidden_layers": 16,
24
+ "num_key_value_heads": 8,
25
+ "pad_token_id": 11,
26
+ "pooling": "avg",
27
+ "rms_norm_eps": 1e-05,
28
+ "rope_parameters": {
29
+ "apply_yarn_scaling": false,
30
+ "beta_fast": 32.0,
31
+ "beta_slow": 1.0,
32
+ "factor": 16.0,
33
+ "llama_4_scaling_beta": 0.1,
34
+ "mscale": 1.0,
35
+ "mscale_all_dim": 1.0,
36
+ "original_max_position_embeddings": 16384,
37
+ "rope_theta": 1000000.0,
38
+ "rope_type": "yarn",
39
+ "type": "yarn"
40
+ },
41
+ "sliding_window": null,
42
+ "tie_word_embeddings": true,
43
+ "transformers_version": "5.13.0",
44
+ "use_cache": false,
45
+ "vocab_size": 131072
46
+ }
config_sentence_transformers.json ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "__version__": {
3
+ "pytorch": "2.11.0+cu128",
4
+ "sentence_transformers": "5.6.0",
5
+ "transformers": "5.13.0"
6
+ },
7
+ "default_prompt_name": null,
8
+ "model_type": "SentenceTransformer",
9
+ "prompts": {
10
+ "anchor": "query: ",
11
+ "negative_1": "passage: ",
12
+ "negative_2": "passage: ",
13
+ "negative_3": "passage: ",
14
+ "negative_4": "passage: ",
15
+ "negative_5": "passage: ",
16
+ "negative_6": "passage: ",
17
+ "negative_7": "passage: ",
18
+ "positive": "passage: "
19
+ },
20
+ "similarity_fn_name": "cosine"
21
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e90f05952a4cc59a5cf75ad7d5a9a24c2cd9585ce4f80a8a648486b7834eeb29
3
+ size 2281852496
modules.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "idx": 0,
4
+ "name": "0",
5
+ "path": "",
6
+ "type": "sentence_transformers.base.modules.transformer.Transformer"
7
+ },
8
+ {
9
+ "idx": 1,
10
+ "name": "1",
11
+ "path": "1_Pooling",
12
+ "type": "sentence_transformers.sentence_transformer.modules.pooling.Pooling"
13
+ },
14
+ {
15
+ "idx": 2,
16
+ "name": "2",
17
+ "path": "2_Normalize",
18
+ "type": "sentence_transformers.sentence_transformer.modules.normalize.Normalize"
19
+ }
20
+ ]
sentence_bert_config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "transformer_task": "feature-extraction",
3
+ "modality_config": {
4
+ "text": {
5
+ "method": "forward",
6
+ "method_output_name": "last_hidden_state"
7
+ }
8
+ },
9
+ "module_output_name": "token_embeddings"
10
+ }
stack.png ADDED
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:39973f835cec6b4b5c9bd5b49a5ee121a5c51614b16206184ac0219b15c3fe91
3
+ size 17077932
tokenizer_config.json ADDED
@@ -0,0 +1,1014 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "bos_token": "<s>",
4
+ "eos_token": "</s>",
5
+ "extra_special_tokens": [
6
+ "<unk>",
7
+ "<s>",
8
+ "</s>",
9
+ "[INST]",
10
+ "[/INST]",
11
+ "[AVAILABLE_TOOLS]",
12
+ "[/AVAILABLE_TOOLS]",
13
+ "[TOOL_RESULTS]",
14
+ "[/TOOL_RESULTS]",
15
+ "[TOOL_CALLS]",
16
+ "[IMG]",
17
+ "<pad>",
18
+ "[IMG_BREAK]",
19
+ "[IMG_END]",
20
+ "[PREFIX]",
21
+ "[MIDDLE]",
22
+ "[SUFFIX]",
23
+ "[SYSTEM_PROMPT]",
24
+ "[/SYSTEM_PROMPT]",
25
+ "[TOOL_CONTENT]",
26
+ "<SPECIAL_20>",
27
+ "<SPECIAL_21>",
28
+ "<SPECIAL_22>",
29
+ "<SPECIAL_23>",
30
+ "[AUDIO]",
31
+ "[BEGIN_AUDIO]",
32
+ "<SPECIAL_26>",
33
+ "<SPECIAL_27>",
34
+ "<SPECIAL_28>",
35
+ "<SPECIAL_29>",
36
+ "<SPECIAL_30>",
37
+ "<SPECIAL_31>",
38
+ "[ARGS]",
39
+ "[CALL_ID]",
40
+ "[THINK]",
41
+ "[/THINK]",
42
+ "<SPECIAL_36>",
43
+ "<SPECIAL_37>",
44
+ "<SPECIAL_38>",
45
+ "<SPECIAL_39>",
46
+ "<SPECIAL_40>",
47
+ "<SPECIAL_41>",
48
+ "<SPECIAL_42>",
49
+ "<SPECIAL_43>",
50
+ "<SPECIAL_44>",
51
+ "<SPECIAL_45>",
52
+ "<SPECIAL_46>",
53
+ "<SPECIAL_47>",
54
+ "<SPECIAL_48>",
55
+ "<SPECIAL_49>",
56
+ "<SPECIAL_50>",
57
+ "<SPECIAL_51>",
58
+ "<SPECIAL_52>",
59
+ "<SPECIAL_53>",
60
+ "<SPECIAL_54>",
61
+ "<SPECIAL_55>",
62
+ "<SPECIAL_56>",
63
+ "<SPECIAL_57>",
64
+ "<SPECIAL_58>",
65
+ "<SPECIAL_59>",
66
+ "<SPECIAL_60>",
67
+ "<SPECIAL_61>",
68
+ "<SPECIAL_62>",
69
+ "<SPECIAL_63>",
70
+ "<SPECIAL_64>",
71
+ "<SPECIAL_65>",
72
+ "<SPECIAL_66>",
73
+ "<SPECIAL_67>",
74
+ "<SPECIAL_68>",
75
+ "<SPECIAL_69>",
76
+ "<SPECIAL_70>",
77
+ "<SPECIAL_71>",
78
+ "<SPECIAL_72>",
79
+ "<SPECIAL_73>",
80
+ "<SPECIAL_74>",
81
+ "<SPECIAL_75>",
82
+ "<SPECIAL_76>",
83
+ "<SPECIAL_77>",
84
+ "<SPECIAL_78>",
85
+ "<SPECIAL_79>",
86
+ "<SPECIAL_80>",
87
+ "<SPECIAL_81>",
88
+ "<SPECIAL_82>",
89
+ "<SPECIAL_83>",
90
+ "<SPECIAL_84>",
91
+ "<SPECIAL_85>",
92
+ "<SPECIAL_86>",
93
+ "<SPECIAL_87>",
94
+ "<SPECIAL_88>",
95
+ "<SPECIAL_89>",
96
+ "<SPECIAL_90>",
97
+ "<SPECIAL_91>",
98
+ "<SPECIAL_92>",
99
+ "<SPECIAL_93>",
100
+ "<SPECIAL_94>",
101
+ "<SPECIAL_95>",
102
+ "<SPECIAL_96>",
103
+ "<SPECIAL_97>",
104
+ "<SPECIAL_98>",
105
+ "<SPECIAL_99>",
106
+ "<SPECIAL_100>",
107
+ "<SPECIAL_101>",
108
+ "<SPECIAL_102>",
109
+ "<SPECIAL_103>",
110
+ "<SPECIAL_104>",
111
+ "<SPECIAL_105>",
112
+ "<SPECIAL_106>",
113
+ "<SPECIAL_107>",
114
+ "<SPECIAL_108>",
115
+ "<SPECIAL_109>",
116
+ "<SPECIAL_110>",
117
+ "<SPECIAL_111>",
118
+ "<SPECIAL_112>",
119
+ "<SPECIAL_113>",
120
+ "<SPECIAL_114>",
121
+ "<SPECIAL_115>",
122
+ "<SPECIAL_116>",
123
+ "<SPECIAL_117>",
124
+ "<SPECIAL_118>",
125
+ "<SPECIAL_119>",
126
+ "<SPECIAL_120>",
127
+ "<SPECIAL_121>",
128
+ "<SPECIAL_122>",
129
+ "<SPECIAL_123>",
130
+ "<SPECIAL_124>",
131
+ "<SPECIAL_125>",
132
+ "<SPECIAL_126>",
133
+ "<SPECIAL_127>",
134
+ "<SPECIAL_128>",
135
+ "<SPECIAL_129>",
136
+ "<SPECIAL_130>",
137
+ "<SPECIAL_131>",
138
+ "<SPECIAL_132>",
139
+ "<SPECIAL_133>",
140
+ "<SPECIAL_134>",
141
+ "<SPECIAL_135>",
142
+ "<SPECIAL_136>",
143
+ "<SPECIAL_137>",
144
+ "<SPECIAL_138>",
145
+ "<SPECIAL_139>",
146
+ "<SPECIAL_140>",
147
+ "<SPECIAL_141>",
148
+ "<SPECIAL_142>",
149
+ "<SPECIAL_143>",
150
+ "<SPECIAL_144>",
151
+ "<SPECIAL_145>",
152
+ "<SPECIAL_146>",
153
+ "<SPECIAL_147>",
154
+ "<SPECIAL_148>",
155
+ "<SPECIAL_149>",
156
+ "<SPECIAL_150>",
157
+ "<SPECIAL_151>",
158
+ "<SPECIAL_152>",
159
+ "<SPECIAL_153>",
160
+ "<SPECIAL_154>",
161
+ "<SPECIAL_155>",
162
+ "<SPECIAL_156>",
163
+ "<SPECIAL_157>",
164
+ "<SPECIAL_158>",
165
+ "<SPECIAL_159>",
166
+ "<SPECIAL_160>",
167
+ "<SPECIAL_161>",
168
+ "<SPECIAL_162>",
169
+ "<SPECIAL_163>",
170
+ "<SPECIAL_164>",
171
+ "<SPECIAL_165>",
172
+ "<SPECIAL_166>",
173
+ "<SPECIAL_167>",
174
+ "<SPECIAL_168>",
175
+ "<SPECIAL_169>",
176
+ "<SPECIAL_170>",
177
+ "<SPECIAL_171>",
178
+ "<SPECIAL_172>",
179
+ "<SPECIAL_173>",
180
+ "<SPECIAL_174>",
181
+ "<SPECIAL_175>",
182
+ "<SPECIAL_176>",
183
+ "<SPECIAL_177>",
184
+ "<SPECIAL_178>",
185
+ "<SPECIAL_179>",
186
+ "<SPECIAL_180>",
187
+ "<SPECIAL_181>",
188
+ "<SPECIAL_182>",
189
+ "<SPECIAL_183>",
190
+ "<SPECIAL_184>",
191
+ "<SPECIAL_185>",
192
+ "<SPECIAL_186>",
193
+ "<SPECIAL_187>",
194
+ "<SPECIAL_188>",
195
+ "<SPECIAL_189>",
196
+ "<SPECIAL_190>",
197
+ "<SPECIAL_191>",
198
+ "<SPECIAL_192>",
199
+ "<SPECIAL_193>",
200
+ "<SPECIAL_194>",
201
+ "<SPECIAL_195>",
202
+ "<SPECIAL_196>",
203
+ "<SPECIAL_197>",
204
+ "<SPECIAL_198>",
205
+ "<SPECIAL_199>",
206
+ "<SPECIAL_200>",
207
+ "<SPECIAL_201>",
208
+ "<SPECIAL_202>",
209
+ "<SPECIAL_203>",
210
+ "<SPECIAL_204>",
211
+ "<SPECIAL_205>",
212
+ "<SPECIAL_206>",
213
+ "<SPECIAL_207>",
214
+ "<SPECIAL_208>",
215
+ "<SPECIAL_209>",
216
+ "<SPECIAL_210>",
217
+ "<SPECIAL_211>",
218
+ "<SPECIAL_212>",
219
+ "<SPECIAL_213>",
220
+ "<SPECIAL_214>",
221
+ "<SPECIAL_215>",
222
+ "<SPECIAL_216>",
223
+ "<SPECIAL_217>",
224
+ "<SPECIAL_218>",
225
+ "<SPECIAL_219>",
226
+ "<SPECIAL_220>",
227
+ "<SPECIAL_221>",
228
+ "<SPECIAL_222>",
229
+ "<SPECIAL_223>",
230
+ "<SPECIAL_224>",
231
+ "<SPECIAL_225>",
232
+ "<SPECIAL_226>",
233
+ "<SPECIAL_227>",
234
+ "<SPECIAL_228>",
235
+ "<SPECIAL_229>",
236
+ "<SPECIAL_230>",
237
+ "<SPECIAL_231>",
238
+ "<SPECIAL_232>",
239
+ "<SPECIAL_233>",
240
+ "<SPECIAL_234>",
241
+ "<SPECIAL_235>",
242
+ "<SPECIAL_236>",
243
+ "<SPECIAL_237>",
244
+ "<SPECIAL_238>",
245
+ "<SPECIAL_239>",
246
+ "<SPECIAL_240>",
247
+ "<SPECIAL_241>",
248
+ "<SPECIAL_242>",
249
+ "<SPECIAL_243>",
250
+ "<SPECIAL_244>",
251
+ "<SPECIAL_245>",
252
+ "<SPECIAL_246>",
253
+ "<SPECIAL_247>",
254
+ "<SPECIAL_248>",
255
+ "<SPECIAL_249>",
256
+ "<SPECIAL_250>",
257
+ "<SPECIAL_251>",
258
+ "<SPECIAL_252>",
259
+ "<SPECIAL_253>",
260
+ "<SPECIAL_254>",
261
+ "<SPECIAL_255>",
262
+ "<SPECIAL_256>",
263
+ "<SPECIAL_257>",
264
+ "<SPECIAL_258>",
265
+ "<SPECIAL_259>",
266
+ "<SPECIAL_260>",
267
+ "<SPECIAL_261>",
268
+ "<SPECIAL_262>",
269
+ "<SPECIAL_263>",
270
+ "<SPECIAL_264>",
271
+ "<SPECIAL_265>",
272
+ "<SPECIAL_266>",
273
+ "<SPECIAL_267>",
274
+ "<SPECIAL_268>",
275
+ "<SPECIAL_269>",
276
+ "<SPECIAL_270>",
277
+ "<SPECIAL_271>",
278
+ "<SPECIAL_272>",
279
+ "<SPECIAL_273>",
280
+ "<SPECIAL_274>",
281
+ "<SPECIAL_275>",
282
+ "<SPECIAL_276>",
283
+ "<SPECIAL_277>",
284
+ "<SPECIAL_278>",
285
+ "<SPECIAL_279>",
286
+ "<SPECIAL_280>",
287
+ "<SPECIAL_281>",
288
+ "<SPECIAL_282>",
289
+ "<SPECIAL_283>",
290
+ "<SPECIAL_284>",
291
+ "<SPECIAL_285>",
292
+ "<SPECIAL_286>",
293
+ "<SPECIAL_287>",
294
+ "<SPECIAL_288>",
295
+ "<SPECIAL_289>",
296
+ "<SPECIAL_290>",
297
+ "<SPECIAL_291>",
298
+ "<SPECIAL_292>",
299
+ "<SPECIAL_293>",
300
+ "<SPECIAL_294>",
301
+ "<SPECIAL_295>",
302
+ "<SPECIAL_296>",
303
+ "<SPECIAL_297>",
304
+ "<SPECIAL_298>",
305
+ "<SPECIAL_299>",
306
+ "<SPECIAL_300>",
307
+ "<SPECIAL_301>",
308
+ "<SPECIAL_302>",
309
+ "<SPECIAL_303>",
310
+ "<SPECIAL_304>",
311
+ "<SPECIAL_305>",
312
+ "<SPECIAL_306>",
313
+ "<SPECIAL_307>",
314
+ "<SPECIAL_308>",
315
+ "<SPECIAL_309>",
316
+ "<SPECIAL_310>",
317
+ "<SPECIAL_311>",
318
+ "<SPECIAL_312>",
319
+ "<SPECIAL_313>",
320
+ "<SPECIAL_314>",
321
+ "<SPECIAL_315>",
322
+ "<SPECIAL_316>",
323
+ "<SPECIAL_317>",
324
+ "<SPECIAL_318>",
325
+ "<SPECIAL_319>",
326
+ "<SPECIAL_320>",
327
+ "<SPECIAL_321>",
328
+ "<SPECIAL_322>",
329
+ "<SPECIAL_323>",
330
+ "<SPECIAL_324>",
331
+ "<SPECIAL_325>",
332
+ "<SPECIAL_326>",
333
+ "<SPECIAL_327>",
334
+ "<SPECIAL_328>",
335
+ "<SPECIAL_329>",
336
+ "<SPECIAL_330>",
337
+ "<SPECIAL_331>",
338
+ "<SPECIAL_332>",
339
+ "<SPECIAL_333>",
340
+ "<SPECIAL_334>",
341
+ "<SPECIAL_335>",
342
+ "<SPECIAL_336>",
343
+ "<SPECIAL_337>",
344
+ "<SPECIAL_338>",
345
+ "<SPECIAL_339>",
346
+ "<SPECIAL_340>",
347
+ "<SPECIAL_341>",
348
+ "<SPECIAL_342>",
349
+ "<SPECIAL_343>",
350
+ "<SPECIAL_344>",
351
+ "<SPECIAL_345>",
352
+ "<SPECIAL_346>",
353
+ "<SPECIAL_347>",
354
+ "<SPECIAL_348>",
355
+ "<SPECIAL_349>",
356
+ "<SPECIAL_350>",
357
+ "<SPECIAL_351>",
358
+ "<SPECIAL_352>",
359
+ "<SPECIAL_353>",
360
+ "<SPECIAL_354>",
361
+ "<SPECIAL_355>",
362
+ "<SPECIAL_356>",
363
+ "<SPECIAL_357>",
364
+ "<SPECIAL_358>",
365
+ "<SPECIAL_359>",
366
+ "<SPECIAL_360>",
367
+ "<SPECIAL_361>",
368
+ "<SPECIAL_362>",
369
+ "<SPECIAL_363>",
370
+ "<SPECIAL_364>",
371
+ "<SPECIAL_365>",
372
+ "<SPECIAL_366>",
373
+ "<SPECIAL_367>",
374
+ "<SPECIAL_368>",
375
+ "<SPECIAL_369>",
376
+ "<SPECIAL_370>",
377
+ "<SPECIAL_371>",
378
+ "<SPECIAL_372>",
379
+ "<SPECIAL_373>",
380
+ "<SPECIAL_374>",
381
+ "<SPECIAL_375>",
382
+ "<SPECIAL_376>",
383
+ "<SPECIAL_377>",
384
+ "<SPECIAL_378>",
385
+ "<SPECIAL_379>",
386
+ "<SPECIAL_380>",
387
+ "<SPECIAL_381>",
388
+ "<SPECIAL_382>",
389
+ "<SPECIAL_383>",
390
+ "<SPECIAL_384>",
391
+ "<SPECIAL_385>",
392
+ "<SPECIAL_386>",
393
+ "<SPECIAL_387>",
394
+ "<SPECIAL_388>",
395
+ "<SPECIAL_389>",
396
+ "<SPECIAL_390>",
397
+ "<SPECIAL_391>",
398
+ "<SPECIAL_392>",
399
+ "<SPECIAL_393>",
400
+ "<SPECIAL_394>",
401
+ "<SPECIAL_395>",
402
+ "<SPECIAL_396>",
403
+ "<SPECIAL_397>",
404
+ "<SPECIAL_398>",
405
+ "<SPECIAL_399>",
406
+ "<SPECIAL_400>",
407
+ "<SPECIAL_401>",
408
+ "<SPECIAL_402>",
409
+ "<SPECIAL_403>",
410
+ "<SPECIAL_404>",
411
+ "<SPECIAL_405>",
412
+ "<SPECIAL_406>",
413
+ "<SPECIAL_407>",
414
+ "<SPECIAL_408>",
415
+ "<SPECIAL_409>",
416
+ "<SPECIAL_410>",
417
+ "<SPECIAL_411>",
418
+ "<SPECIAL_412>",
419
+ "<SPECIAL_413>",
420
+ "<SPECIAL_414>",
421
+ "<SPECIAL_415>",
422
+ "<SPECIAL_416>",
423
+ "<SPECIAL_417>",
424
+ "<SPECIAL_418>",
425
+ "<SPECIAL_419>",
426
+ "<SPECIAL_420>",
427
+ "<SPECIAL_421>",
428
+ "<SPECIAL_422>",
429
+ "<SPECIAL_423>",
430
+ "<SPECIAL_424>",
431
+ "<SPECIAL_425>",
432
+ "<SPECIAL_426>",
433
+ "<SPECIAL_427>",
434
+ "<SPECIAL_428>",
435
+ "<SPECIAL_429>",
436
+ "<SPECIAL_430>",
437
+ "<SPECIAL_431>",
438
+ "<SPECIAL_432>",
439
+ "<SPECIAL_433>",
440
+ "<SPECIAL_434>",
441
+ "<SPECIAL_435>",
442
+ "<SPECIAL_436>",
443
+ "<SPECIAL_437>",
444
+ "<SPECIAL_438>",
445
+ "<SPECIAL_439>",
446
+ "<SPECIAL_440>",
447
+ "<SPECIAL_441>",
448
+ "<SPECIAL_442>",
449
+ "<SPECIAL_443>",
450
+ "<SPECIAL_444>",
451
+ "<SPECIAL_445>",
452
+ "<SPECIAL_446>",
453
+ "<SPECIAL_447>",
454
+ "<SPECIAL_448>",
455
+ "<SPECIAL_449>",
456
+ "<SPECIAL_450>",
457
+ "<SPECIAL_451>",
458
+ "<SPECIAL_452>",
459
+ "<SPECIAL_453>",
460
+ "<SPECIAL_454>",
461
+ "<SPECIAL_455>",
462
+ "<SPECIAL_456>",
463
+ "<SPECIAL_457>",
464
+ "<SPECIAL_458>",
465
+ "<SPECIAL_459>",
466
+ "<SPECIAL_460>",
467
+ "<SPECIAL_461>",
468
+ "<SPECIAL_462>",
469
+ "<SPECIAL_463>",
470
+ "<SPECIAL_464>",
471
+ "<SPECIAL_465>",
472
+ "<SPECIAL_466>",
473
+ "<SPECIAL_467>",
474
+ "<SPECIAL_468>",
475
+ "<SPECIAL_469>",
476
+ "<SPECIAL_470>",
477
+ "<SPECIAL_471>",
478
+ "<SPECIAL_472>",
479
+ "<SPECIAL_473>",
480
+ "<SPECIAL_474>",
481
+ "<SPECIAL_475>",
482
+ "<SPECIAL_476>",
483
+ "<SPECIAL_477>",
484
+ "<SPECIAL_478>",
485
+ "<SPECIAL_479>",
486
+ "<SPECIAL_480>",
487
+ "<SPECIAL_481>",
488
+ "<SPECIAL_482>",
489
+ "<SPECIAL_483>",
490
+ "<SPECIAL_484>",
491
+ "<SPECIAL_485>",
492
+ "<SPECIAL_486>",
493
+ "<SPECIAL_487>",
494
+ "<SPECIAL_488>",
495
+ "<SPECIAL_489>",
496
+ "<SPECIAL_490>",
497
+ "<SPECIAL_491>",
498
+ "<SPECIAL_492>",
499
+ "<SPECIAL_493>",
500
+ "<SPECIAL_494>",
501
+ "<SPECIAL_495>",
502
+ "<SPECIAL_496>",
503
+ "<SPECIAL_497>",
504
+ "<SPECIAL_498>",
505
+ "<SPECIAL_499>",
506
+ "<SPECIAL_500>",
507
+ "<SPECIAL_501>",
508
+ "<SPECIAL_502>",
509
+ "<SPECIAL_503>",
510
+ "<SPECIAL_504>",
511
+ "<SPECIAL_505>",
512
+ "<SPECIAL_506>",
513
+ "<SPECIAL_507>",
514
+ "<SPECIAL_508>",
515
+ "<SPECIAL_509>",
516
+ "<SPECIAL_510>",
517
+ "<SPECIAL_511>",
518
+ "<SPECIAL_512>",
519
+ "<SPECIAL_513>",
520
+ "<SPECIAL_514>",
521
+ "<SPECIAL_515>",
522
+ "<SPECIAL_516>",
523
+ "<SPECIAL_517>",
524
+ "<SPECIAL_518>",
525
+ "<SPECIAL_519>",
526
+ "<SPECIAL_520>",
527
+ "<SPECIAL_521>",
528
+ "<SPECIAL_522>",
529
+ "<SPECIAL_523>",
530
+ "<SPECIAL_524>",
531
+ "<SPECIAL_525>",
532
+ "<SPECIAL_526>",
533
+ "<SPECIAL_527>",
534
+ "<SPECIAL_528>",
535
+ "<SPECIAL_529>",
536
+ "<SPECIAL_530>",
537
+ "<SPECIAL_531>",
538
+ "<SPECIAL_532>",
539
+ "<SPECIAL_533>",
540
+ "<SPECIAL_534>",
541
+ "<SPECIAL_535>",
542
+ "<SPECIAL_536>",
543
+ "<SPECIAL_537>",
544
+ "<SPECIAL_538>",
545
+ "<SPECIAL_539>",
546
+ "<SPECIAL_540>",
547
+ "<SPECIAL_541>",
548
+ "<SPECIAL_542>",
549
+ "<SPECIAL_543>",
550
+ "<SPECIAL_544>",
551
+ "<SPECIAL_545>",
552
+ "<SPECIAL_546>",
553
+ "<SPECIAL_547>",
554
+ "<SPECIAL_548>",
555
+ "<SPECIAL_549>",
556
+ "<SPECIAL_550>",
557
+ "<SPECIAL_551>",
558
+ "<SPECIAL_552>",
559
+ "<SPECIAL_553>",
560
+ "<SPECIAL_554>",
561
+ "<SPECIAL_555>",
562
+ "<SPECIAL_556>",
563
+ "<SPECIAL_557>",
564
+ "<SPECIAL_558>",
565
+ "<SPECIAL_559>",
566
+ "<SPECIAL_560>",
567
+ "<SPECIAL_561>",
568
+ "<SPECIAL_562>",
569
+ "<SPECIAL_563>",
570
+ "<SPECIAL_564>",
571
+ "<SPECIAL_565>",
572
+ "<SPECIAL_566>",
573
+ "<SPECIAL_567>",
574
+ "<SPECIAL_568>",
575
+ "<SPECIAL_569>",
576
+ "<SPECIAL_570>",
577
+ "<SPECIAL_571>",
578
+ "<SPECIAL_572>",
579
+ "<SPECIAL_573>",
580
+ "<SPECIAL_574>",
581
+ "<SPECIAL_575>",
582
+ "<SPECIAL_576>",
583
+ "<SPECIAL_577>",
584
+ "<SPECIAL_578>",
585
+ "<SPECIAL_579>",
586
+ "<SPECIAL_580>",
587
+ "<SPECIAL_581>",
588
+ "<SPECIAL_582>",
589
+ "<SPECIAL_583>",
590
+ "<SPECIAL_584>",
591
+ "<SPECIAL_585>",
592
+ "<SPECIAL_586>",
593
+ "<SPECIAL_587>",
594
+ "<SPECIAL_588>",
595
+ "<SPECIAL_589>",
596
+ "<SPECIAL_590>",
597
+ "<SPECIAL_591>",
598
+ "<SPECIAL_592>",
599
+ "<SPECIAL_593>",
600
+ "<SPECIAL_594>",
601
+ "<SPECIAL_595>",
602
+ "<SPECIAL_596>",
603
+ "<SPECIAL_597>",
604
+ "<SPECIAL_598>",
605
+ "<SPECIAL_599>",
606
+ "<SPECIAL_600>",
607
+ "<SPECIAL_601>",
608
+ "<SPECIAL_602>",
609
+ "<SPECIAL_603>",
610
+ "<SPECIAL_604>",
611
+ "<SPECIAL_605>",
612
+ "<SPECIAL_606>",
613
+ "<SPECIAL_607>",
614
+ "<SPECIAL_608>",
615
+ "<SPECIAL_609>",
616
+ "<SPECIAL_610>",
617
+ "<SPECIAL_611>",
618
+ "<SPECIAL_612>",
619
+ "<SPECIAL_613>",
620
+ "<SPECIAL_614>",
621
+ "<SPECIAL_615>",
622
+ "<SPECIAL_616>",
623
+ "<SPECIAL_617>",
624
+ "<SPECIAL_618>",
625
+ "<SPECIAL_619>",
626
+ "<SPECIAL_620>",
627
+ "<SPECIAL_621>",
628
+ "<SPECIAL_622>",
629
+ "<SPECIAL_623>",
630
+ "<SPECIAL_624>",
631
+ "<SPECIAL_625>",
632
+ "<SPECIAL_626>",
633
+ "<SPECIAL_627>",
634
+ "<SPECIAL_628>",
635
+ "<SPECIAL_629>",
636
+ "<SPECIAL_630>",
637
+ "<SPECIAL_631>",
638
+ "<SPECIAL_632>",
639
+ "<SPECIAL_633>",
640
+ "<SPECIAL_634>",
641
+ "<SPECIAL_635>",
642
+ "<SPECIAL_636>",
643
+ "<SPECIAL_637>",
644
+ "<SPECIAL_638>",
645
+ "<SPECIAL_639>",
646
+ "<SPECIAL_640>",
647
+ "<SPECIAL_641>",
648
+ "<SPECIAL_642>",
649
+ "<SPECIAL_643>",
650
+ "<SPECIAL_644>",
651
+ "<SPECIAL_645>",
652
+ "<SPECIAL_646>",
653
+ "<SPECIAL_647>",
654
+ "<SPECIAL_648>",
655
+ "<SPECIAL_649>",
656
+ "<SPECIAL_650>",
657
+ "<SPECIAL_651>",
658
+ "<SPECIAL_652>",
659
+ "<SPECIAL_653>",
660
+ "<SPECIAL_654>",
661
+ "<SPECIAL_655>",
662
+ "<SPECIAL_656>",
663
+ "<SPECIAL_657>",
664
+ "<SPECIAL_658>",
665
+ "<SPECIAL_659>",
666
+ "<SPECIAL_660>",
667
+ "<SPECIAL_661>",
668
+ "<SPECIAL_662>",
669
+ "<SPECIAL_663>",
670
+ "<SPECIAL_664>",
671
+ "<SPECIAL_665>",
672
+ "<SPECIAL_666>",
673
+ "<SPECIAL_667>",
674
+ "<SPECIAL_668>",
675
+ "<SPECIAL_669>",
676
+ "<SPECIAL_670>",
677
+ "<SPECIAL_671>",
678
+ "<SPECIAL_672>",
679
+ "<SPECIAL_673>",
680
+ "<SPECIAL_674>",
681
+ "<SPECIAL_675>",
682
+ "<SPECIAL_676>",
683
+ "<SPECIAL_677>",
684
+ "<SPECIAL_678>",
685
+ "<SPECIAL_679>",
686
+ "<SPECIAL_680>",
687
+ "<SPECIAL_681>",
688
+ "<SPECIAL_682>",
689
+ "<SPECIAL_683>",
690
+ "<SPECIAL_684>",
691
+ "<SPECIAL_685>",
692
+ "<SPECIAL_686>",
693
+ "<SPECIAL_687>",
694
+ "<SPECIAL_688>",
695
+ "<SPECIAL_689>",
696
+ "<SPECIAL_690>",
697
+ "<SPECIAL_691>",
698
+ "<SPECIAL_692>",
699
+ "<SPECIAL_693>",
700
+ "<SPECIAL_694>",
701
+ "<SPECIAL_695>",
702
+ "<SPECIAL_696>",
703
+ "<SPECIAL_697>",
704
+ "<SPECIAL_698>",
705
+ "<SPECIAL_699>",
706
+ "<SPECIAL_700>",
707
+ "<SPECIAL_701>",
708
+ "<SPECIAL_702>",
709
+ "<SPECIAL_703>",
710
+ "<SPECIAL_704>",
711
+ "<SPECIAL_705>",
712
+ "<SPECIAL_706>",
713
+ "<SPECIAL_707>",
714
+ "<SPECIAL_708>",
715
+ "<SPECIAL_709>",
716
+ "<SPECIAL_710>",
717
+ "<SPECIAL_711>",
718
+ "<SPECIAL_712>",
719
+ "<SPECIAL_713>",
720
+ "<SPECIAL_714>",
721
+ "<SPECIAL_715>",
722
+ "<SPECIAL_716>",
723
+ "<SPECIAL_717>",
724
+ "<SPECIAL_718>",
725
+ "<SPECIAL_719>",
726
+ "<SPECIAL_720>",
727
+ "<SPECIAL_721>",
728
+ "<SPECIAL_722>",
729
+ "<SPECIAL_723>",
730
+ "<SPECIAL_724>",
731
+ "<SPECIAL_725>",
732
+ "<SPECIAL_726>",
733
+ "<SPECIAL_727>",
734
+ "<SPECIAL_728>",
735
+ "<SPECIAL_729>",
736
+ "<SPECIAL_730>",
737
+ "<SPECIAL_731>",
738
+ "<SPECIAL_732>",
739
+ "<SPECIAL_733>",
740
+ "<SPECIAL_734>",
741
+ "<SPECIAL_735>",
742
+ "<SPECIAL_736>",
743
+ "<SPECIAL_737>",
744
+ "<SPECIAL_738>",
745
+ "<SPECIAL_739>",
746
+ "<SPECIAL_740>",
747
+ "<SPECIAL_741>",
748
+ "<SPECIAL_742>",
749
+ "<SPECIAL_743>",
750
+ "<SPECIAL_744>",
751
+ "<SPECIAL_745>",
752
+ "<SPECIAL_746>",
753
+ "<SPECIAL_747>",
754
+ "<SPECIAL_748>",
755
+ "<SPECIAL_749>",
756
+ "<SPECIAL_750>",
757
+ "<SPECIAL_751>",
758
+ "<SPECIAL_752>",
759
+ "<SPECIAL_753>",
760
+ "<SPECIAL_754>",
761
+ "<SPECIAL_755>",
762
+ "<SPECIAL_756>",
763
+ "<SPECIAL_757>",
764
+ "<SPECIAL_758>",
765
+ "<SPECIAL_759>",
766
+ "<SPECIAL_760>",
767
+ "<SPECIAL_761>",
768
+ "<SPECIAL_762>",
769
+ "<SPECIAL_763>",
770
+ "<SPECIAL_764>",
771
+ "<SPECIAL_765>",
772
+ "<SPECIAL_766>",
773
+ "<SPECIAL_767>",
774
+ "<SPECIAL_768>",
775
+ "<SPECIAL_769>",
776
+ "<SPECIAL_770>",
777
+ "<SPECIAL_771>",
778
+ "<SPECIAL_772>",
779
+ "<SPECIAL_773>",
780
+ "<SPECIAL_774>",
781
+ "<SPECIAL_775>",
782
+ "<SPECIAL_776>",
783
+ "<SPECIAL_777>",
784
+ "<SPECIAL_778>",
785
+ "<SPECIAL_779>",
786
+ "<SPECIAL_780>",
787
+ "<SPECIAL_781>",
788
+ "<SPECIAL_782>",
789
+ "<SPECIAL_783>",
790
+ "<SPECIAL_784>",
791
+ "<SPECIAL_785>",
792
+ "<SPECIAL_786>",
793
+ "<SPECIAL_787>",
794
+ "<SPECIAL_788>",
795
+ "<SPECIAL_789>",
796
+ "<SPECIAL_790>",
797
+ "<SPECIAL_791>",
798
+ "<SPECIAL_792>",
799
+ "<SPECIAL_793>",
800
+ "<SPECIAL_794>",
801
+ "<SPECIAL_795>",
802
+ "<SPECIAL_796>",
803
+ "<SPECIAL_797>",
804
+ "<SPECIAL_798>",
805
+ "<SPECIAL_799>",
806
+ "<SPECIAL_800>",
807
+ "<SPECIAL_801>",
808
+ "<SPECIAL_802>",
809
+ "<SPECIAL_803>",
810
+ "<SPECIAL_804>",
811
+ "<SPECIAL_805>",
812
+ "<SPECIAL_806>",
813
+ "<SPECIAL_807>",
814
+ "<SPECIAL_808>",
815
+ "<SPECIAL_809>",
816
+ "<SPECIAL_810>",
817
+ "<SPECIAL_811>",
818
+ "<SPECIAL_812>",
819
+ "<SPECIAL_813>",
820
+ "<SPECIAL_814>",
821
+ "<SPECIAL_815>",
822
+ "<SPECIAL_816>",
823
+ "<SPECIAL_817>",
824
+ "<SPECIAL_818>",
825
+ "<SPECIAL_819>",
826
+ "<SPECIAL_820>",
827
+ "<SPECIAL_821>",
828
+ "<SPECIAL_822>",
829
+ "<SPECIAL_823>",
830
+ "<SPECIAL_824>",
831
+ "<SPECIAL_825>",
832
+ "<SPECIAL_826>",
833
+ "<SPECIAL_827>",
834
+ "<SPECIAL_828>",
835
+ "<SPECIAL_829>",
836
+ "<SPECIAL_830>",
837
+ "<SPECIAL_831>",
838
+ "<SPECIAL_832>",
839
+ "<SPECIAL_833>",
840
+ "<SPECIAL_834>",
841
+ "<SPECIAL_835>",
842
+ "<SPECIAL_836>",
843
+ "<SPECIAL_837>",
844
+ "<SPECIAL_838>",
845
+ "<SPECIAL_839>",
846
+ "<SPECIAL_840>",
847
+ "<SPECIAL_841>",
848
+ "<SPECIAL_842>",
849
+ "<SPECIAL_843>",
850
+ "<SPECIAL_844>",
851
+ "<SPECIAL_845>",
852
+ "<SPECIAL_846>",
853
+ "<SPECIAL_847>",
854
+ "<SPECIAL_848>",
855
+ "<SPECIAL_849>",
856
+ "<SPECIAL_850>",
857
+ "<SPECIAL_851>",
858
+ "<SPECIAL_852>",
859
+ "<SPECIAL_853>",
860
+ "<SPECIAL_854>",
861
+ "<SPECIAL_855>",
862
+ "<SPECIAL_856>",
863
+ "<SPECIAL_857>",
864
+ "<SPECIAL_858>",
865
+ "<SPECIAL_859>",
866
+ "<SPECIAL_860>",
867
+ "<SPECIAL_861>",
868
+ "<SPECIAL_862>",
869
+ "<SPECIAL_863>",
870
+ "<SPECIAL_864>",
871
+ "<SPECIAL_865>",
872
+ "<SPECIAL_866>",
873
+ "<SPECIAL_867>",
874
+ "<SPECIAL_868>",
875
+ "<SPECIAL_869>",
876
+ "<SPECIAL_870>",
877
+ "<SPECIAL_871>",
878
+ "<SPECIAL_872>",
879
+ "<SPECIAL_873>",
880
+ "<SPECIAL_874>",
881
+ "<SPECIAL_875>",
882
+ "<SPECIAL_876>",
883
+ "<SPECIAL_877>",
884
+ "<SPECIAL_878>",
885
+ "<SPECIAL_879>",
886
+ "<SPECIAL_880>",
887
+ "<SPECIAL_881>",
888
+ "<SPECIAL_882>",
889
+ "<SPECIAL_883>",
890
+ "<SPECIAL_884>",
891
+ "<SPECIAL_885>",
892
+ "<SPECIAL_886>",
893
+ "<SPECIAL_887>",
894
+ "<SPECIAL_888>",
895
+ "<SPECIAL_889>",
896
+ "<SPECIAL_890>",
897
+ "<SPECIAL_891>",
898
+ "<SPECIAL_892>",
899
+ "<SPECIAL_893>",
900
+ "<SPECIAL_894>",
901
+ "<SPECIAL_895>",
902
+ "<SPECIAL_896>",
903
+ "<SPECIAL_897>",
904
+ "<SPECIAL_898>",
905
+ "<SPECIAL_899>",
906
+ "<SPECIAL_900>",
907
+ "<SPECIAL_901>",
908
+ "<SPECIAL_902>",
909
+ "<SPECIAL_903>",
910
+ "<SPECIAL_904>",
911
+ "<SPECIAL_905>",
912
+ "<SPECIAL_906>",
913
+ "<SPECIAL_907>",
914
+ "<SPECIAL_908>",
915
+ "<SPECIAL_909>",
916
+ "<SPECIAL_910>",
917
+ "<SPECIAL_911>",
918
+ "<SPECIAL_912>",
919
+ "<SPECIAL_913>",
920
+ "<SPECIAL_914>",
921
+ "<SPECIAL_915>",
922
+ "<SPECIAL_916>",
923
+ "<SPECIAL_917>",
924
+ "<SPECIAL_918>",
925
+ "<SPECIAL_919>",
926
+ "<SPECIAL_920>",
927
+ "<SPECIAL_921>",
928
+ "<SPECIAL_922>",
929
+ "<SPECIAL_923>",
930
+ "<SPECIAL_924>",
931
+ "<SPECIAL_925>",
932
+ "<SPECIAL_926>",
933
+ "<SPECIAL_927>",
934
+ "<SPECIAL_928>",
935
+ "<SPECIAL_929>",
936
+ "<SPECIAL_930>",
937
+ "<SPECIAL_931>",
938
+ "<SPECIAL_932>",
939
+ "<SPECIAL_933>",
940
+ "<SPECIAL_934>",
941
+ "<SPECIAL_935>",
942
+ "<SPECIAL_936>",
943
+ "<SPECIAL_937>",
944
+ "<SPECIAL_938>",
945
+ "<SPECIAL_939>",
946
+ "<SPECIAL_940>",
947
+ "<SPECIAL_941>",
948
+ "<SPECIAL_942>",
949
+ "<SPECIAL_943>",
950
+ "<SPECIAL_944>",
951
+ "<SPECIAL_945>",
952
+ "<SPECIAL_946>",
953
+ "<SPECIAL_947>",
954
+ "<SPECIAL_948>",
955
+ "<SPECIAL_949>",
956
+ "<SPECIAL_950>",
957
+ "<SPECIAL_951>",
958
+ "<SPECIAL_952>",
959
+ "<SPECIAL_953>",
960
+ "<SPECIAL_954>",
961
+ "<SPECIAL_955>",
962
+ "<SPECIAL_956>",
963
+ "<SPECIAL_957>",
964
+ "<SPECIAL_958>",
965
+ "<SPECIAL_959>",
966
+ "<SPECIAL_960>",
967
+ "<SPECIAL_961>",
968
+ "<SPECIAL_962>",
969
+ "<SPECIAL_963>",
970
+ "<SPECIAL_964>",
971
+ "<SPECIAL_965>",
972
+ "<SPECIAL_966>",
973
+ "<SPECIAL_967>",
974
+ "<SPECIAL_968>",
975
+ "<SPECIAL_969>",
976
+ "<SPECIAL_970>",
977
+ "<SPECIAL_971>",
978
+ "<SPECIAL_972>",
979
+ "<SPECIAL_973>",
980
+ "<SPECIAL_974>",
981
+ "<SPECIAL_975>",
982
+ "<SPECIAL_976>",
983
+ "<SPECIAL_977>",
984
+ "<SPECIAL_978>",
985
+ "<SPECIAL_979>",
986
+ "<SPECIAL_980>",
987
+ "<SPECIAL_981>",
988
+ "<SPECIAL_982>",
989
+ "<SPECIAL_983>",
990
+ "<SPECIAL_984>",
991
+ "<SPECIAL_985>",
992
+ "<SPECIAL_986>",
993
+ "<SPECIAL_987>",
994
+ "<SPECIAL_988>",
995
+ "<SPECIAL_989>",
996
+ "<SPECIAL_990>",
997
+ "<SPECIAL_991>",
998
+ "<SPECIAL_992>",
999
+ "<SPECIAL_993>",
1000
+ "<SPECIAL_994>",
1001
+ "<SPECIAL_995>",
1002
+ "<SPECIAL_996>",
1003
+ "<SPECIAL_997>",
1004
+ "<SPECIAL_998>",
1005
+ "<SPECIAL_999>"
1006
+ ],
1007
+ "is_local": true,
1008
+ "local_files_only": false,
1009
+ "model_max_length": 4096,
1010
+ "pad_token": "<pad>",
1011
+ "padding_side": "left",
1012
+ "tokenizer_class": "TokenizersBackend",
1013
+ "unk_token": "<unk>"
1014
+ }