Upload Maaza-MLM-135M-JSON-v1 - v1.0.0 production release

README.md

@@ -21,6 +21,7 @@ Micro Language Model (135M parameters) specialized for JSON extraction on edge d
 
 ## Model Details
 
+- **Family**: CycleCore Maaza Series (Micro/Small Language Models for structured JSON extraction)
 - **Developer**: CycleCore Technologies
 - **Model Name**: CycleCore Maaza MLM-135M-JSON
 - **Version**: v1.0.0
@@ -158,6 +159,29 @@ Output:
 - **Format**: JSON mode enforced
 - **Platform**: CUDA (GPU) or CPU
 
+### JSON Validation
+
+After generation, validate the output:
+
+```python
+import json
+from jsonschema import validate  # pip install jsonschema
+
+# Decode model output
+output = tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+# Parse JSON
+try:
+    obj = json.loads(output)
+    # Validate against schema
+    validate(instance=obj, schema=your_json_schema)
+    print("✅ Valid JSON matching schema")
+except json.JSONDecodeError:
+    print("❌ Invalid JSON")
+except jsonschema.exceptions.ValidationError as e:
+    print(f"❌ Schema validation failed: {e.message}")
+```
+
 ## Limitations and Bias
 
 ### Known Limitations
@@ -180,6 +204,12 @@ Output:
 - **Energy**: Ultra-fast training (48.7s) and efficient inference reduce carbon footprint
 - **Transparency**: 100% open training methodology, reproducible results
 
+## Links
+
+- **EdgeJSON Benchmark**: [GitHub Repository](https://github.com/CycleCore/SLMBench) (evaluation harness & dataset)
+- **Technical Documentation**: [SLMBench Docs](https://github.com/CycleCore/SLMBench/tree/main/benchmarks/edge_json)
+- **Model Card (360M)**: [Maaza-SLM-360M-JSON](https://huggingface.co/CycleCoreTechnologies/Maaza-SLM-360M-JSON-v1)
+
 ## How to Use
 
 ### Installation