dakheel commited on
Commit
f9b59d6
·
verified ·
1 Parent(s): 0526fee

Deploy HUDA-Net v43-AR frozen neural runtime; preserve v41 rollback

Browse files
DEPLOY_HUDANET_V43_AR.md ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # HUDA-Net v43-AR Stage 9
2
+
3
+ هذه هي مرحلة النشر الأخيرة للنسخة العربية المجمدة.
4
+
5
+ ## ماذا تفعل الحزمة؟
6
+
7
+ 1. تتحقق من بصمتي TOP2 وReranker المجمدتين.
8
+ 2. تستخدم Corpus Stage1 النهائي ذو 2,796 سجلًا، مع Parquet أولًا وCSV مطابق كـ fallback تقني فقط.
9
+ 3. تحذف منه حقول الاختبار والسؤال المخفي وحقول split/family/contract.
10
+ 4. تحتفظ فقط بـ `passage_ar_pure` كشاهد مرئي، وتعيد تسميته `passage_ar`.
11
+ 5. تحسب TOP2 passage embeddings محليًا مرة واحدة.
12
+ 6. تنشئ Dataset خاصًا على Hugging Face وتتحقق أنه Private قبل رفع أي بيانات، باسم:
13
+ `dakheel/hudanet-v43-ar-runtime-corpus`
14
+ 7. ترفع App v43-AR إلى Space `dakheel/hudanet-api`.
15
+ 8. تحفظ نسخة v41.0.2 الحالية كـ rollback.
16
+
17
+ المسار النشط:
18
+
19
+ `TOP2 -> Stage4-v2 -> RRF k=60 -> Stage5 Calibrator -> threshold 0.702353569...`
20
+
21
+ لا يوجد في المسار النشط:
22
+
23
+ - semantic_rules.json
24
+ - hudanet_dialects.json
25
+ - hudanet_retrieval_rules.json
26
+ - GenericEvidencePipeline
27
+ - manual synonym expansion
28
+ - English runtime
29
+
30
+ إذا أجاب النظام، نص الجواب نفسه هو `passage_ar` المرئي المختار، وRecord ID الداعم هو نفس الشاهد المرئي.
31
+
32
+ ## التشغيل
33
+
34
+ ```powershell
35
+ cd "C:\Users\iamda\OneDrive\Desktop"
36
+
37
+ Expand-Archive `
38
+ -Path ".\HUDANET_V43_AR_STAGE9_SPACE_DEPLOY.zip" `
39
+ -DestinationPath "." `
40
+ -Force
41
+
42
+ cd ".\HUDANET_V43_AR_STAGE9_SPACE_DEPLOY"
43
+
44
+ Set-ExecutionPolicy -Scope Process Bypass -Force
45
+
46
+ .\run_stage9.ps1
47
+ ```
48
+
49
+ إذا طلب تسجيل دخول Hugging Face:
50
+
51
+ ```powershell
52
+ hf auth login
53
+ .\run_stage9.ps1
54
+ ```
55
+
56
+ لا ترسل التوكن داخل المحادثة.
57
+
58
+ ## بناء محلي فقط من دون رفع
59
+
60
+ ```powershell
61
+ .\run_stage9.ps1 -PrepareOnly
62
+ ```
63
+
64
+ ## الرجوع إلى v41.0.2 إذا احتجنا
65
+
66
+ ```powershell
67
+ .\rollback_to_v41.ps1
68
+ ```
69
+
70
+ بعد رفع Stage 9، ننتظر Build الـSpace ثم نعمل Live Smoke Test فقط. لا يوجد أي tuning بعد الـBlind.
README.md CHANGED
@@ -1,5 +1,5 @@
1
  ---
2
- title: HUDA-Net
3
  emoji: 🕋
4
  colorFrom: green
5
  colorTo: green
@@ -9,4 +9,20 @@ app_file: app.py
9
  pinned: false
10
  ---
11
 
12
- # HUDA-Net
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ title: HUDA-Net v43-AR
3
  emoji: 🕋
4
  colorFrom: green
5
  colorTo: green
 
9
  pinned: false
10
  ---
11
 
12
+ # HUDA-Net v43-AR
13
+
14
+ Active frozen Arabic neural release.
15
+
16
+ `TOP2 Retriever -> Stage4-v2 Reranker -> RRF k=60 -> Stage5 Calibrator -> Fixed Abstention`
17
+
18
+ The active v43-AR application does not load the legacy semantic rule files,
19
+ dialect dictionary, manual synonym expansion, or regex intent classifier.
20
+
21
+ If an answer is emitted, its text is the selected visible `passage_ar`; the
22
+ answer-support record ID is the same visible record. Hidden answer pools are
23
+ forbidden.
24
+
25
+ English is intentionally deferred to a separately trained release.
26
+
27
+ The former v41.0.2 app is retained in this Space only as an explicit rollback
28
+ artifact and is not imported by the active v43 app.
README_v41_0_2_rollback.md ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ title: HUDA-Net
3
+ emoji: 🕋
4
+ colorFrom: green
5
+ colorTo: green
6
+ sdk: gradio
7
+ python_version: "3.11"
8
+ app_file: app.py
9
+ pinned: false
10
+ ---
11
+
12
+ # HUDA-Net
V43_ACTIVE_CONTRACT.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "version": "43.0.0-AR",
3
+ "status": "ACTIVE_FROZEN_ARABIC",
4
+ "model_repo": "dakheel/hudanet-v43-ar",
5
+ "runtime_corpus_repo": "dakheel/hudanet-v43-ar-runtime-corpus",
6
+ "top2_recursive_sha256": "b1e27b53b616f98bb406dcb94a88583a2b1d659dedb0a45464a310c4200d5173",
7
+ "reranker_recursive_sha256": "82e301c2aef211540b628f4b5517d81de79bf5c8e0cf1f69524cf0dec40d0fbc",
8
+ "rrf_k": 60,
9
+ "answer_threshold": 0.7023535690146311,
10
+ "active_language": "Arabic only",
11
+ "english_runtime_enabled": false,
12
+ "static_semantic_keyword_rules_used": false,
13
+ "dialect_dictionary_used": false,
14
+ "manual_synonym_expansion_used": false,
15
+ "regex_intent_classification_used": false,
16
+ "hidden_answer_pool_allowed": false,
17
+ "answer_support_must_equal_selected_visible_evidence": true,
18
+ "post_blind_tuning_allowed": false,
19
+ "legacy_v41_files_present_for_rollback_only": true
20
+ }
app.py CHANGED
The diff for this file is too large to render. See raw diff
 
app_v41_0_2_rollback.py ADDED
The diff for this file is too large to render. See raw diff
 
requirements.txt CHANGED
@@ -1,10 +1,10 @@
1
  numpy<3
2
  pandas<3
3
- scipy<2
4
  scikit-learn==1.6.1
5
  joblib<2
6
- pyarrow
7
- sentence-transformers<6
8
  transformers<6
9
  safetensors<1
10
- huggingface_hub>=0.30,<2
 
 
1
  numpy<3
2
  pandas<3
3
+ pyarrow
4
  scikit-learn==1.6.1
5
  joblib<2
6
+ torch<3
 
7
  transformers<6
8
  safetensors<1
9
+ huggingface_hub>=0.34,<2
10
+ gradio>=5.38,<6
requirements_v41_0_2_rollback.txt ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ numpy<3
2
+ pandas<3
3
+ scipy<2
4
+ scikit-learn==1.6.1
5
+ joblib<2
6
+ pyarrow
7
+ sentence-transformers<6
8
+ transformers<6
9
+ safetensors<1
10
+ huggingface_hub>=0.30,<2