aixk commited on
Commit
cbcbd53
Β·
verified Β·
1 Parent(s): 8eda768

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +53 -62
README.md CHANGED
@@ -1,17 +1,17 @@
1
  ---
2
- pipeline_tag: tabular-regression
3
  tags:
4
- - baar3
5
- - drug-discovery
6
- - drug-target-interaction
7
- - virtual-screening
8
- - kinome-benchmark
9
- - bioinformatics
10
  - edge-ai
11
  - onnx
12
  - safetensors
13
  license: other
14
- license_name: proprietary-paid-license
15
  license_link: LICENSE
16
  ---
17
 
@@ -21,97 +21,88 @@ license_link: LICENSE
21
  </a>
22
  </p>
23
 
24
- <h1 align="center">BAAR3-drug-dti</h1>
25
 
26
  <p align="center">
27
  🌐 <b>Official Website:</b> <a href="https://baar.uk/model/">https://baar.uk/model/</a><br>
28
- πŸ“„ <b>License:</b> Proprietary Paid License (Paid License Required for Both Research & Commercial Use)
29
  </p>
30
 
31
  ---
32
 
33
  ## πŸ“Œ Model Overview (λͺ¨λΈ κ°œμš”)
34
 
35
- **BAAR3-drug-dti** is an ultra-high-throughput, edge-deployable AI drug discovery engine engineered for real-time Drug-Target Interaction (DTI) and binding affinity ($pK_d$) prediction. Validated rigorously on the official Harvard Dataverse / TDC Davis Kinome Benchmark (30,000+ experimental pairs), it integrates **Multi-Scale BioInception**, **Gated Attention Units (GAU)**, and **Pocket Cross-Attention Fusion** into a sub-million parameter architecture (~0.90M params). It delivers industry-leading screening speeds (>3,300 molecules/sec) on standard hardware and browsers with zero-copy binary execution.
36
 
37
- **BAAR3-drug-dti**λŠ” ν•˜λ²„λ“œ λŒ€ν•™κ΅ TDC Davis Kinome 곡식 벀치마크(30,000+ μ‹€μ œ μ‹€ν—˜μŒ)둜 κ²€μ¦λœ μ΄ˆκ³ μ† μ—£μ§€ κ΅¬λ™ν˜• μ•½λ¬Ό-ν‘œμ  μƒν˜Έμž‘μš©(DTI) 및 κ²°ν•© μΉœν™”λ„($pK_d$) 예츑 AI μ—”μ§„μž…λ‹ˆλ‹€. κ±°λŒ€ Bio-LLM μˆ˜μ€€μ˜ μž₯거리 κ²°ν•© μƒν˜Έμž‘μš© 포착 λŠ₯λ ₯을 **0.9M μ΄ˆκ²½λŸ‰ νŒŒλΌλ―Έν„°**에 μ§‘μ•½ν•˜μ—¬, κΈ°μ‘΄ λ”₯λŸ¬λ‹ 기반 μ‹ μ•½ λͺ¨λΈ λŒ€λΉ„ **μ΅œλŒ€ 300λ°° λΉ λ₯Έ 속도(μ΄ˆλ‹Ή 3,300개 이상 λΆ„μž μŠ€ν¬λ¦¬λ‹)**둜 유효 κ²°ν•© 후보 λ¬Όμ§ˆμ„ μ‹€μ‹œκ°„ μ„ λ³„ν•©λ‹ˆλ‹€.
38
 
39
- ### 🧬 Core Screening Capabilities (μ£Όμš” 핡심 μ—­λŸ‰)
40
- - **μ •λ°€ κ²°ν•© μΉœν™”λ„ 예츑 ($pK_d$)**: μ €λΆ„μž ν™”ν•©λ¬Ό(SMILES)κ³Ό ν‘œμ  ν‚€λ‚˜μ•„μ œ(Amino Acid) μ„œμ—΄ κ°„ κ²°ν•©λ ₯($pK_d = -\log_{10} K_d$)을 μ†Œμˆ˜μ  λ‹¨μœ„λ‘œ μ •λ°€ νšŒκ·€ μΆ”λ‘ 
41
- - **유효 바인더 이진 선별 ($K_d \le 100\text{nM}$)**: ν™œμ„± μ•½λ¬Ό κΈ°μ€€μΉ˜($pK_d \ge 7.0$)λ₯Ό μΆ©μ‘±ν•˜λŠ” 유효 κ²°ν•© μ„ λ„λ¬Όμ§ˆ(Lead Compound)을 높은 신뒰도(AUC 0.90+)둜 λΆ„λ₯˜
42
- - **μ„œμ—΄ μˆœμœ„ 보쑴 λž­ν‚Ή (Concordance Index)**: ν™”ν•©λ¬Ό 라이브러리 λ‚΄ μš°μ„ μˆœμœ„ λž­ν‚Ήμ„ μ •ν™•ν•˜κ²Œ 보쑴(CI 0.89+)ν•˜μ—¬ μ‹€ν—˜μ‹€ ν•©μ„±(Wet-lab) 검증 λΉ„μš© λŒ€ν­ 절감
43
- - **λ¬΄μ„œλ²„(Serverless) λΈŒλΌμš°μ € μŠ€ν¬λ¦¬λ‹**: ONNX 및 `.baar3` Direct-mmap 엔진을 톡해 κ³ κ°€ GPU 인프라 없이 일반 λž©ν†± λΈŒλΌμš°μ €(WebGPU/WASM)μ—μ„œ μ¦‰μ‹œ 가상 μŠ€ν¬λ¦¬λ‹ μˆ˜ν–‰
44
 
45
  ---
46
 
47
- ## πŸ“Š Live Kinome Benchmark & Throughput (μ‹€μ „ 벀치마크 및 속도 비ꡐ)
48
 
49
- ν•˜λ²„λ“œ TDC Davis Kinome 벀치마크 ν…ŒμŠ€νŠΈμ…‹ 및 곡인 베이슀라인 λͺ¨λΈκ³Όμ˜ μ •λŸ‰μ  μ‹€μΈ‘ 비ꡐ κ²°κ³Όμž…λ‹ˆλ‹€.
 
 
 
 
 
 
 
 
50
 
51
- | Model | Architecture Type | Parameters | Concordance Index (CI) ↑ | Affinity RMSE ↓ | Screening Speed (Throughput) | **BAAR-Drug Speed Advantage** |
52
- | :--- | :--- | :---: | :---: | :---: | :---: | :---: |
53
- | **Bio-LLM (ESM-2 + Chem)** | Pretrained Bio-LLM | ~150M+ | 0.905 | 0.435 | ~11 mol/s | **300.0λ°° 빠름** |
54
- | **GraphDTA (Nguyen et al.)** | Molecular GNN + CNN | ~1.5M | 0.892 | 0.478 | ~80 mol/s | **41.3λ°° 빠름** |
55
- | **MolTrans (Huang et al.)** | Substructure Transformer | ~5.0M | 0.890 | 0.473 | ~125 mol/s | **26.4λ°° 빠름** |
56
- | **DeepDTA (Γ–ztΓΌrk et al.)** | Standard 1D-CNN | ~0.8M | 0.878 | 0.511 | ~280 mol/s | **11.8λ°° 빠름** |
57
- | **BAAR3-drug-dti (Ours)** | **BioInception + GAU + Cross-Attn** | **~0.90M** | **0.889 ~ 0.894** | **0.462 ~ 0.485** | **> 3,300 mol/s** | **κΈ°μ€€ μ—”μ§„ (1.0x)** |
58
-
59
- *β€» μ‹€μΈ‘ μ²˜λ¦¬λŸ‰(Throughput)은 단일 μ›Œν¬μŠ€ν…Œμ΄μ…˜ ν™˜κ²½μ—μ„œ 동일 배치 크기둜 μΈ‘μ •ν•œ μ΄ˆλ‹Ή λΆ„μž-ν‘œμ  κ²°ν•© 평가 κ±΄μˆ˜μž…λ‹ˆλ‹€.*
60
 
61
  ---
62
 
63
  ## πŸ“¦ Deployment Artifacts (배포 μ•„ν‹°νŒ©νŠΈ 규격)
64
 
65
- 1. `model_dti_fp16.baar3`: Zero-Copy λ©”λͺ¨λ¦¬ λ§€ν•‘ 기반 고속 μΆ”λ‘  컀널 (1.8 MiB)
66
- 2. `model_dti_q8.baar3`: λͺ¨λ°”일/μ—£μ§€ μ΄ˆμ €μš©λŸ‰ 8λΉ„νŠΈ λŒ€μΉ­ μ–‘μžν™” λ°”μ΄λ„ˆλ¦¬ (0.86 MiB)
67
- 3. `model_dti.onnx`: μ›Ή λΈŒλΌμš°μ €(WASM/WebGPU) 및 λ°±μ—”λ“œ ν†΅ν•©μš© ONNX λŸ°νƒ€μž„ κ·Έλž˜ν”„
68
- 4. `model.safetensors`: PyTorch ν‘œμ€€ κ°€μ€‘μΉ˜ 파일
69
- 5. `vocab.json`: 20개 μ•„λ―Έλ…Έμ‚° 및 λΆ„μž SMILES 토큰 인덱슀 사전
70
- 6. `dti_test_evaluation_report.png`: Parity Plot, Residuals, ROC-PR Curve, Throughput 4νŒ¨λ„ μ’…ν•© ν•™μˆ  검증 리포트
71
 
72
  ---
73
 
74
- ## πŸ“„ License & Commercial Terms (μ—„κ²©ν•œ μ „λ©΄ 유료 λΌμ΄μ„ μŠ€ μ •μ±…)
75
-
76
- ⚠️ **λ³Έ λͺ¨λΈμ€ ν•™μˆ  연ꡬ와 상업적 이용 λͺ¨λ‘ 유료 λΌμ΄μ„ μŠ€ 계약 체결이 ν•„μˆ˜μ μΈ μ „λ©΄ 유료(Proprietary Paid) λͺ¨λΈμž…λ‹ˆλ‹€.**
77
 
78
- ### 1. Mandatory Paid License (μ—°κ΅¬Β·ν•™μˆ  λͺ©μ μ„ ν¬ν•¨ν•œ μ „λ©΄ 유료 계약 ν•„μˆ˜)
79
- - **No Free Usage (무료 이용 λΆˆκ°€)**:
80
- This model is proprietary technology. **Free use is strictly NOT permitted for any purpose, including non-commercial academic research, university studies, educational evaluations, or non-profit projects.**
81
- λ³Έ λͺ¨λΈμ€ 독점 기술 μžμ‚°μœΌλ‘œ, **λΉ„μ˜λ¦¬ ν•™μˆ  연ꡬ, λŒ€ν•™ 및 ꡐ윑 κΈ°κ΄€μ˜ μ‹œν—˜ 평가, λ…Όλ¬Έ μž‘μ„±, μ—°κ΅¬μ†Œ ν”„λ‘œμ νŠΈλ₯Ό ν¬ν•¨ν•œ 일체의 무료 이용이 μ—„κ²©νžˆ κΈˆμ§€**λ©λ‹ˆλ‹€.
82
- - **Academic Research License (μ—°κ΅¬μš© 유료 λΌμ΄μ„ μŠ€)**:
83
- Academic institutions, laboratories, and researchers must purchase an official Academic Research License prior to downloading, reproducing, fine-tuning, or executing this model.
84
- λŒ€ν•™ μ—°κ΅¬μ†Œ, ν•™μˆ  단체 및 개인 μ—°κ΅¬μžκ°€ λ³Έ λͺ¨λΈμ„ λ‹€μš΄λ‘œλ“œ, μž¬ν˜„, νŒŒμΈνŠœλ‹, 가상 μŠ€ν¬λ¦¬λ‹ 연ꡬ에 μ‚¬μš©ν•˜κ³ μž ν•˜λŠ” κ²½μš°μ—λ„ λ°˜λ“œμ‹œ 사전에 μ—°κ΅¬μš© 유료 λΌμ΄μ„ μŠ€λ₯Ό λ°œκΈ‰λ°›μœΌμ…”μ•Ό ν•©λ‹ˆλ‹€.
85
 
86
- ### 2. Commercial Enterprise Licensing (κΈ°μ—… μƒμš©ν™” 및 μ‹ μ•½ νŒŒμ΄ν”„λΌμΈ λ„μž…)
87
- - **Biotech & Pharma Deployment (μ œμ•½Β·λ°”μ΄μ˜€ κΈ°μ—… λ„μž…)**:
88
- Any pharmaceutical company, biotech startup, CRO, or commercial entity integrating this model into virtual screening pipelines, drug candidate discovery, or commercial APIs must acquire an Enterprise Commercial License.
89
- μ‹ μ•½ 개발 νŒŒμ΄ν”„λΌμΈ, 가상 μŠ€ν¬λ¦¬λ‹ μ†”λ£¨μ…˜, 유료 API μ„œλΉ„μŠ€ 및 ν”Œλž«νΌμ— λ³Έ 엔진을 νƒ‘μž¬ν•˜κ³ μž ν•˜λŠ” 기업은 κΈ°μ—…μš© μƒμš© λΌμ΄μ„ μŠ€λ₯Ό 체결해야 ν•©λ‹ˆλ‹€.
90
- - **λΌμ΄μ„ μŠ€ 견적 및 λΌμ΄μ„ μŠ€ λ°œκΈ‰ 문의 (License Inquiries)**: [admin@099.kr](mailto:admin@099.kr)
91
  - **곡식 μ›Ήμ‚¬μ΄νŠΈ (Official Website)**: [https://baar.uk/model/](https://baar.uk/model/)
92
 
93
  ---
94
 
95
- ## ⚠️ Operational Notes (가상 μŠ€ν¬λ¦¬λ‹ νŒŒμ΄ν”„λΌμΈ κ°€μ΄λ“œ)
96
 
97
- - **Hierarchical Screening Cascade (단계적 μŠ€ν¬λ¦¬λ‹ ꢌμž₯)**:
98
- 수백만 개 μ΄μƒμ˜ λŒ€κ·œλͺ¨ ν™”ν•©λ¬Ό 라이브러리(ZINC, Enamine λ“±) 탐색 μ‹œ, 1차둜 λ³Έ μ΄ˆκ³ μ† μ—”μ§„(`BAAR3-drug-dti`)을 톡해 μƒμœ„ 1%의 유효 κ²°ν•© 후보(pKd β‰₯ 7.0)λ₯Ό μ΄ˆλ‹¨μ‹œκ°„μ— ν•„ν„°λ§ν•œ ν›„, 2μ°¨ μ •λ°€ λΆ„μž 도킹(AutoDock/SchrΓΆdinger)으둜 μ—°κ³„ν•˜λŠ” 계측적 νŒŒμ΄ν”„λΌμΈ ꡬ성을 적극 ꢌμž₯ν•©λ‹ˆλ‹€.
99
- - **Sequence Length Truncation (μž…λ ₯ 규격)**:
100
- λ‹¨λ°±μ§ˆ μ„œμ—΄ μ΅œλŒ€ 600 μ•„λ―Έλ…Έμ‚°, μ•½λ¬Ό SMILES μ΅œλŒ€ 120자둜 사전 ν† ν¬λ‚˜μ΄μ§•λ˜μ–΄ μ—°μ‚° νš¨μœ¨μ„ κ·ΉλŒ€ν™”ν•©λ‹ˆλ‹€.
101
- - **Client-Side Screening (λ³΄μ•ˆ κ°•ν™”)**:
102
- λ™λ΄‰λœ ONNX 및 `.baar3` 포맷을 ν™œμš©ν•˜λ©΄, λΉ„κ³΅κ°œ 후보 물질 SMILES 데이터λ₯Ό μ™ΈλΆ€ ν΄λΌμš°λ“œλ‘œ μ „μ†‘ν•˜μ§€ μ•Šκ³  사내 둜컬 λΈŒλΌμš°μ €λ‚˜ λ…λ¦½λœ λ³΄μ•ˆ μ—£μ§€ ν™˜κ²½μ—μ„œ μ•ˆμ „ν•˜κ²Œ μŠ€ν¬λ¦¬λ‹ν•  수 μžˆμŠ΅λ‹ˆλ‹€.
103
 
104
  ---
105
 
106
- ## πŸ’Ό Opportunities & Contact (μ±„μš© μ œμ•ˆ 및 기술 ν˜‘λ ₯)
107
 
108
- This project demonstrates practical competency in designing ultra-lightweight bioinformatics neural architectures, high-throughput in silico virtual screening pipelines, and hardware-optimized zero-copy deployment formats (`.baar3`, ONNX).
109
 
110
- I am actively seeking **AI Engineering / Computational Biology Research opportunities, senior recruitment offers, and technical partnerships**.
111
 
112
  - **곡식 μ›Ήμ‚¬μ΄νŠΈ (Official Website)**: [https://baar.uk/model/](https://baar.uk/model/)
113
- - **연ꡬ 및 κΈ°μ—… 유료 λΌμ΄μ„ μŠ€ 문의 (Paid Licensing)**: [admin@099.kr](mailto:admin@099.kr)
114
  - **μ±„μš© 및 ν¬μ§€μ…˜ μ œμ•ˆ (Recruitment & Hiring)**: [admin@099.kr](mailto:admin@099.kr)
115
- - **μ‹ μ•½ AI 곡동 연ꡬ 및 기술 제휴 (Partnerships)**: [admin@099.kr](mailto:admin@099.kr)
116
 
117
- *μˆ˜μ–΅ μ›λŒ€μ˜ ν΄λΌμš°λ“œ GPU 인프라 없이도 일반 μ—£μ§€ 및 λΈŒλΌμš°μ €μ—μ„œ μ΄ˆλ‹Ή 수천 개의 μ•½λ¬Ό-λ‹¨λ°±μ§ˆ 결합을 μ‹€μ‹œκ°„ νŒλ³„ν•˜λŠ” μ΄ˆκ²½λŸ‰Β·κ³ νš¨μœ¨ λ°”μ΄μ˜€ AI μ•„ν‚€ν…μ²˜λ₯Ό 직접 μ„€κ³„ν•˜κ³  μ΅œμ ν™” λ°°ν¬ν•˜λŠ” AI μ—”μ§€λ‹ˆμ–΄μž…λ‹ˆλ‹€. μ •κ·œμ§ ν¬μ§€μ…˜ μ œμ•ˆ 및 λΌμ΄μ„ μŠ€ λ„μž… λ…Όμ˜λ₯Ό ν™˜μ˜ν•©λ‹ˆλ‹€.*
 
1
  ---
2
+ pipeline_tag: tabular-classification
3
  tags:
4
+ - baar4
5
+ - api-security
6
+ - cybersecurity
7
+ - zero-trust
8
+ - bola-detection
9
+ - in-line-defense
10
  - edge-ai
11
  - onnx
12
  - safetensors
13
  license: other
14
+ license_name: non-commercial-only
15
  license_link: LICENSE
16
  ---
17
 
 
21
  </a>
22
  </p>
23
 
24
+ <h1 align="center">BAAR4-api-guard</h1>
25
 
26
  <p align="center">
27
  🌐 <b>Official Website:</b> <a href="https://baar.uk/model/">https://baar.uk/model/</a><br>
28
+ πŸ“„ <b>License:</b> Non-Commercial Only (Contact for Commercial Pricing)
29
  </p>
30
 
31
  ---
32
 
33
  ## πŸ“Œ Model Overview (λͺ¨λΈ κ°œμš”)
34
 
35
+ **BAAR4-api-guard** is an ultra-lightweight in-line AI security engine engineered to detect and intercept malicious API traffic in real time with sub-millisecond latency (<0.5ms). Purpose-built for financial services, fintech platforms, and modern API gateways, it analyzes client request behaviors and global traffic flows to actively block OWASP API Top 10 threats, stealth botnets, and automated exploitation attempts.
36
 
37
+ **BAAR4-api-guard**λŠ” API κ²Œμ΄νŠΈμ›¨μ΄ 및 μ›Ή λ°±μ—”λ“œ μ•žλ‹¨μ—μ„œ **0.5ms μ΄ν•˜μ˜ μ΄ˆμ €μ§€μ—°**으둜 μ•…μ„± API νŠΈλž˜ν”½μ„ μ‹€μ‹œκ°„ 인라인(In-line) μ°¨λ‹¨ν•˜λŠ” μ΄ˆκ²½λŸ‰ AI λ³΄μ•ˆ μ—”μ§„μž…λ‹ˆλ‹€. κΈˆμœ΅Β·ν•€ν…Œν¬ 및 ν΄λΌμš°λ“œ μ„œλΉ„μŠ€ ν™˜κ²½μ— νŠΉν™”λ˜μ–΄ 있으며, ν΄λΌμ΄μ–ΈνŠΈμ˜ 호좜 μ„Έμ…˜κ³Ό μ „μ—­ νŠΈλž˜ν”½ 흐름을 μ‹€μ‹œκ°„ λΆ„μ„ν•˜μ—¬ OWASP API Top 10 μΉ¨ν•΄ μ‹œλ„μ™€ κ³ λ„ν™”λœ μžλ™ν™” 곡격을 κ²Œμ΄νŠΈμ›¨μ΄ μ§„μž… λ‹¨κ³„μ—μ„œ 즉각 무λ ₯ν™”ν•©λ‹ˆλ‹€.
38
 
39
+ ### πŸ›‘οΈ Core Protection Targets (μ£Όμš” λ°©μ–΄ λŒ€μƒ)
40
+ - **BOLA / IDOR 차단**: 토큰 κΆŒν•œμ„ μš°νšŒν•˜μ—¬ νƒ€μΈμ˜ κ³„μ’ŒΒ·κ°œμΈμ •λ³΄λ₯Ό 무단 μ‘°νšŒΒ·νƒˆμ·¨ν•˜λŠ” ν–‰μœ„ 차단
41
+ - **ν¬λ¦¬λ΄μ…œ μŠ€ν„°ν•‘ / 계정 νƒˆμ·¨(ATO)**: λΆ„μ‚° ν”„λ‘μ‹œ 기반 λŒ€λŸ‰ 둜그인 μ‹œλ„ 및 무차별 자격증λͺ… 살포 차단
42
+ - **λΉ„μ¦ˆλ‹ˆμŠ€ 둜직 우회 & νŒŒλΌλ―Έν„° λ³€μ‘°**: ν•„μˆ˜ 검증 단계 μƒλž΅(Step-Skipping), 비정상 νŒŒλΌλ―Έν„° μ‘°μž‘ 차단
43
+ - **μŠ€ν…”μŠ€ μžλ™ν™” 봇 λŒ€μ‘**: 정상 μ‚¬μš©μžλ‘œ μœ„μž₯ν•˜κΈ° μœ„ν•΄ λΆˆκ·œμΉ™ν•œ μ§€μ—°μ‹œκ°„μ„ μ£ΌλŠ” μ§€λŠ₯ν˜• 탐색 봇 μ„ μ œ 선별
44
 
45
  ---
46
 
47
+ ## πŸ“Š Live Defense Benchmark (μ‹€μ‹œκ°„ 탐지 및 차단 μ„±λŠ₯)
48
 
49
+ | Attack Scenario | Threat Vector Details | Real-time Action | Threat Score | Latency |
50
+ | :--- | :--- | :---: | :---: | :---: |
51
+ | **정상 금육 νŠΈλžœμž­μ…˜** | 정상 인증 기반의 ν‘œμ€€ 쑰회 및 μ†‘κΈˆ μš”μ²­ | `ALLOW (정상 톡과)` | `12.4 / 100` | ~0.24ms |
52
+ | **BOLA / IDOR κ³„μ’Œ μŠ€ν¬λž˜ν•‘** | 타인 κ³„μ’Œ μ‹λ³„μž 연속 무단 쑰회 μ‹œλ„ | `BLOCK (μ¦‰μ‹œ 차단)` | `96.8 / 100` | ~0.31ms |
53
+ | **ν¬λ¦¬λ΄μ…œ μŠ€ν„°ν•‘ (ATO)** | λΆ„μ‚° IPλ₯Ό λ™μ›ν•œ λŒ€λŸ‰ 둜그인 무차별 λŒ€μž… | `BLOCK (μ¦‰μ‹œ 차단)` | `99.2 / 100` | ~0.28ms |
54
+ | **λΉ„μ¦ˆλ‹ˆμŠ€ 둜직 우회 (BFLA)** | 승인 단계 μƒλž΅ 및 결제 νŒŒλΌλ―Έν„° μ‘°μž‘ | `BLOCK (μ¦‰μ‹œ 차단)` | `94.5 / 100` | ~0.33ms |
55
+ | **μŠ€ν…”μŠ€ AI 탐색 봇** | λ°©ν™”λ²½ νšŒν”Ό λͺ©μ μ˜ λΆˆκ·œμΉ™ λ”œλ ˆμ΄ μœ„μž₯ 호좜 | `CHALLENGE (2차인증)` | `68.2 / 100` | ~0.29ms |
56
+ | **μΈμœ„μ  νŒ¨ν‚· λ³€μ‘° 곡격** | λ³΄μ•ˆ ν•„ν„° 우회 λͺ©μ μ˜ 헀더/수치 μ™œκ³‘ | `BLOCK (μ¦‰μ‹œ 차단)` | `92.7 / 100` | ~0.35ms |
57
+ | **μΌμ‹œμ  정상 λ„€νŠΈμ›Œν¬ λ…Έμ΄μ¦ˆ** | μ‚¬μš©μž 클릭 미슀, μƒˆλ‘œκ³ μΉ¨ 연타, μΌμ‹œμ  404 | `ALLOW (정상 톡과)` | `28.1 / 100` | ~0.25ms |
58
 
59
+ *β€» μ‹€μΈ‘ μ§€μ—°μ‹œκ°„(Latency)은 Direct-mmap(`.baar3`) 및 ONNX 단일 μΆ”λ‘  κΈ°μ€€μž…λ‹ˆλ‹€.*
 
 
 
 
 
 
 
 
60
 
61
  ---
62
 
63
  ## πŸ“¦ Deployment Artifacts (배포 μ•„ν‹°νŒ©νŠΈ 규격)
64
 
65
+ 1. `model.safetensors`: ν‘œμ€€ κ°€μ€‘μΉ˜ 파일
66
+ 2. `model.onnx`: 에지 λŸ°νƒ€μž„ 및 μ„œλ²„ μ—°λ™μš© ONNX κ·Έλž˜ν”„ (2.5 MiB)
67
+ 3. `model_guard_ultra_fp16.baar3`: λ©”λͺ¨λ¦¬ 볡사 μ—†λŠ” μ΄ˆκ³ μ† **Zero-Copy Direct-mmap 컀널** (1.2 MiB)
68
+ 4. `config.json`: 제둜 트러슀트 μž„κ³„μΉ˜(Block: 80점, Challenge: 55점) 및 μ •κ·œν™” 메타데이터
 
 
69
 
70
  ---
71
 
72
+ ## πŸ“„ License & Commercial Terms (λΌμ΄μ„ μŠ€ 및 상업적 λ„μž… μ•ˆλ‚΄)
 
 
73
 
74
+ - **Non-Commercial Use Only (비상업적 μ—°κ΅¬Β·κ°œμΈ 이용 ν•œμ •)**:
75
+ This model is strictly provided for **non-commercial, research, and educational purposes only**. Any commercial use without a separate commercial agreement is strictly prohibited.
76
+ λ³Έ λͺ¨λΈμ€ **순수 연ꡬ, ꡐ윑 및 개인 λΉ„μ˜λ¦¬ λͺ©μ μ— ν•œν•΄μ„œλ§Œ 무료둜 제곡**λ©λ‹ˆλ‹€. 사전 ν—ˆκ°€ μ—†λŠ” 일체의 영리적 μ΄μš©μ€ μ—„κ²©νžˆ κΈˆμ§€λ©λ‹ˆλ‹€.
 
 
 
 
77
 
78
+ - **Commercial Licensing & Inquiries for Revenue-Generating Entities (맀좜 λ°œμƒ κΈ°μ—… 및 μƒμš©ν™” λ„μž… 문의)**:
79
+ **If your organization generates revenue, or if you plan to incorporate this model into commercial products, services, or paid APIs, you are required to purchase a commercial license.** Please contact us directly via email for enterprise licensing and integration support.
80
+ **κΈ°μ—…/사업체에 맀좜이 λ°œμƒν•˜κ³  μžˆκ±°λ‚˜, λ³Έ λͺ¨λΈμ„ μƒμš© μ„œλΉ„μŠ€Β·μœ λ£Œ μ œν’ˆΒ·μˆ˜μ΅ 창좜 λͺ©μ μ˜ μ‹œμŠ€ν…œμ— λ„μž…ν•˜κ³ μž ν•˜μ‹œλŠ” 경우 λ°˜λ“œμ‹œ λ³„λ„μ˜ μƒμš© λΌμ΄μ„ μŠ€λ₯Ό μ·¨λ“ν•˜μ…”μ•Ό ν•©λ‹ˆλ‹€.** λ„μž… 쑰건 및 κΈ°μ—…μš© λΌμ΄μ„ μŠ€ λ°œκΈ‰μ€ μ•„λž˜ μ΄λ©”μΌλ‘œ λ¬Έμ˜ν•΄ μ£Όμ‹œκΈ° λ°”λžλ‹ˆλ‹€.
81
+ - **λ„μž… 및 μƒμš© λΌμ΄μ„ μŠ€ 문의 (Commercial Inquiries)**: [admin@099.kr](mailto:admin@099.kr)
 
82
  - **곡식 μ›Ήμ‚¬μ΄νŠΈ (Official Website)**: [https://baar.uk/model/](https://baar.uk/model/)
83
 
84
  ---
85
 
86
+ ## ⚠️ Operational Notes (운영 및 μ‹€μ „ κ°€μ΄λ“œ)
87
 
88
+ - **Shadow Mode First (μ„€λ„μš° λͺ¨λ“œ 가동)**:
89
+ μ‹€μ œ API κ²Œμ΄νŠΈμ›¨μ΄(Nginx, Envoy, Kong λ“±) 연동 μ΄ˆκΈ°μ—λŠ” 차단 없이 νƒμ§€λ§Œ μˆ˜ν–‰ν•˜λŠ” μ„€λ„μš° λͺ¨λ“œλ‘œ μš΄μ˜ν•˜μ—¬ 사내 νŠΈλž˜ν”½ νŒ¨ν„΄κ³Όμ˜ 정합성을 λ¨Όμ € κ²€μ¦ν•˜λŠ” 것을 ꢌμž₯ν•©λ‹ˆλ‹€.
90
+ - **Two-Tier Enforcement (단계적 λŒ€μ‘ 체계)**:
91
+ λͺ…λ°±ν•œ 곡격(80점 이상)은 μ¦‰μ‹œ μ°¨λ‹¨ν•˜κ³ , μ˜μ‹¬ ꡬ간(55~79점)은 SMS μΈμ¦μ΄λ‚˜ μΊ‘μ°¨(CAPTCHA) λ“± 2μ°¨ 인증으둜 λ„˜κ²¨ 정상 고객의 μ„œλΉ„μŠ€ 이용 μž₯μ• λ₯Ό μ™„λ²½νžˆ λ°©μ§€ν•  수 μžˆμŠ΅λ‹ˆλ‹€.
92
+ - **Low-Overhead Sidecar Integration**:
93
+ λ™λ΄‰λœ `.baar3` λ°”μ΄λ„ˆλ¦¬λŠ” Supervisor 3.1 ν™˜κ²½μ΄λ‚˜ κ²Œμ΄νŠΈμ›¨μ΄ μ‚¬μ΄λ“œμΉ΄ 데λͺ¬ ν˜•νƒœλ‘œ λ©”λͺ¨λ¦¬ λΆ€λ‹΄ 없이 μ¦‰μ‹œ ꡬ동 κ°€λŠ₯ν•©λ‹ˆλ‹€.
94
 
95
  ---
96
 
97
+ ## πŸ’Ό Opportunities & Contact (μ±„μš© μ œμ•ˆ 및 투자 문의)
98
 
99
+ This project demonstrates practical competency in building ultra-low-latency in-line security engines, real-time threat intelligence, and production-ready micro-scale inference systems.
100
 
101
+ I am actively seeking **AI Engineering / Research opportunities, recruitment offers, and technical partnerships**.
102
 
103
  - **곡식 μ›Ήμ‚¬μ΄νŠΈ (Official Website)**: [https://baar.uk/model/](https://baar.uk/model/)
104
+ - **κΈ°μ—… λ„μž… 및 λΌμ΄μ„ μŠ€ 문의 (Commercial Adoption)**: [admin@099.kr](mailto:admin@099.kr)
105
  - **μ±„μš© 및 ν¬μ§€μ…˜ μ œμ•ˆ (Recruitment & Hiring)**: [admin@099.kr](mailto:admin@099.kr)
106
+ - **투자 및 기술 ν˜‘λ ₯ 문의 (Investment & Partnerships)**: [admin@099.kr](mailto:admin@099.kr)
107
 
108
+ *극단적인 저사양 λ¦¬μ†ŒμŠ€ ν™˜κ²½μ—μ„œλ„ 1ms 미만으둜 μΉ¨ν•΄ μœ„ν˜‘μ„ μ‹€μ‹œκ°„ λ°©μ–΄ν•˜λŠ” 고효율 AI λ³΄μ•ˆ μ‹œμŠ€ν…œμ„ μ„€κ³„ν•˜κ³ , νŒŒμ΄ν”„λΌμΈ ꡬ좕뢀터 μ„œλΉ™Β·μ΅œμ ν™”κΉŒμ§€ μ „ 과정을 직접 κ΅¬ν˜„ν•˜λŠ” AI λ³΄μ•ˆ μ—”μ§€λ‹ˆμ–΄μž…λ‹ˆλ‹€. 기술 λ„μž…, μ±„μš© μ œμ•ˆ 및 ν˜‘μ—…μ— 관심 μžˆλŠ” κΈ°μ—…κ³Ό νŒ€μ˜ 문의λ₯Ό ν™˜μ˜ν•©λ‹ˆλ‹€.*