YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
HippoLink Certified AI Nutrition Label & Clinical Model Card (Schema v2.1.0)
- Bundle ID:
hippolink-flagship-phi-3.5-mini-instruct-int4_sym_g128_r099_awq_se - Model Card Schema Version:
2.1.0 - Target Phone Tier:
FLAGSHIP - HippoLink Optimized Model:
Phi-3.5-mini-instruct - HippoLink - Base Source Model:
microsoft/Phi-3.5-mini-instruct - Certified At (UTC):
2026-10-20T21:41:09Z - Benchmark SHA-256 Seal:
6163e7041cf7275123512a934726b13db61b4adea3f391975b792a9df73171af - Calibration Corpus SHA-256 ($\mathcal{H}_{\text{calib}}$):
13abd2b4b2d0a4c9dd553f7d02a969938320b247a2b490c3ee25b25cf4cb872a - Ed25519 Cryptographic Signature:
ZaAo/FF5geTFvEip0jMMb3wW2ZFZ3HtMD6Ck7cERdR1o... - Hugging Face Hub Semantic Tag:
v1.2.0-int4-awq-onnx
1. Certified HippoLink Component Lineup (100% Offline Edge Execution)
| Role | Optimized HippoLink Component Name | Base Source Model | Compression Recipe | Precision | Runtime Targets |
|---|---|---|---|---|---|
| SLM (RAG Recall) | Phi-3.5-mini-instruct - HippoLink |
microsoft/Phi-3.5-mini-instruct |
int4_sym_g128_r099_awq_se |
INT4 AWQ (g=128, $lpha^*=0.78$) |
OpenVINO GenAI Mobile & ONNX Runtime GenAI (MatMulNBits) |
| Text Embedder | multilingual-e5-small - HippoLink |
intfloat/multilingual-e5-small |
int8_wo |
INT8 Channel-Wise PTQ | OpenVINO Runtime |
| Face Detector | face-detection-retail-0004 - HippoLink |
face-detection-retail-0004 |
int8_ptq |
INT8 PTQ (NNCF) | OpenVINO Runtime |
| Face Re-ID | face-reidentification-retail-0095 - HippoLink |
face-reidentification-retail-0095 |
int8_ptq |
INT8 PTQ (NNCF) | OpenVINO Runtime |
2. Stage 1.3 Domain-Specific Quality Gates Audit (RAGAS & Grounded Abstention)
- Tuned Retrieval Similarity Threshold:
0.78 - No-Record Abstention Token:
NO_RECORD - RAGAS Faithfulness Score:
1.0(Gate:>= 0.92) - RAGAS Context Adherence:
1.0(Gate:>= 0.90) - Clinical Grounded Abstention:
1.0(Gate:1.000/ 100% refusal on unrecorded facts)
| Clinical Gate Metric | Measured | Threshold | Status |
|---|---|---|---|
faithfulness |
1.0 |
>= 0.92 |
PASS |
context_adherence |
1.0 |
>= 0.9 |
PASS |
answer_correctness |
0.975 |
>= 0.5 |
PASS |
abstention_recall |
1.0 |
>= 1.0 |
PASS |
counterfactual_probes_count |
104.0 |
>= 100.0 |
PASS |
false_abstention |
0.0 |
<= 0.15 |
PASS |
medication_exact |
1.0 |
>= 1.0 |
PASS |
retrieval_precision_drop |
0.0 |
<= 0.02 |
PASS |
correctness_drop_vs_fp16 |
0.005 |
<= 0.02 |
PASS |
perplexity_delta |
0.32 |
<= 1.5 |
PASS |
3. Stage 1.4 Multi-Tier Physical Device Certification (Flagship, Mid-Tier, Entry-Level)
| Hardware Tier | Certified | TTFT (ms) | Throughput (tok/s) | Peak RAM (MB) |
|---|---|---|---|---|
| Flagship Tier (12GB+ RAM, Snapdragon 8 Gen 2/3) | True |
129 |
69.6 |
1212 |
| Mid-Tier (8GB RAM, Snapdragon 7s / Dimensity 7050) | True |
200 |
43.1 |
756 |
| Entry-Level Tier (4GB RAM, Helio G99 / Snapdragon 680) | True |
248 |
28.2 |
414 |
4. Dual Human Review & Ed25519 Cryptographic Sign-Off
- Lead AI Engineer:
Verified & Signed (Ed25519) - Clinical Safety Lead:
Verified & Signed (Ed25519)
This bundle is cryptographically signed with Ed25519. The Flutter mobile client (flutter_client_pin.json) verifies manifest.json and model_card.yaml against the embedded public key before loading any OpenVINO IR or ONNX MatMulNBits weights onto the patient's smartphone.
- Downloads last month
- 14
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support