Create MODEL_CARD.json
Browse files- MODEL_CARD.json +231 -0
MODEL_CARD.json
ADDED
|
@@ -0,0 +1,231 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_details": {
|
| 3 |
+
"name": "Helion-2.5-Rnd",
|
| 4 |
+
"version": "2.5.0-rnd",
|
| 5 |
+
"full_name": "DeepXR/Helion-2.5-Rnd",
|
| 6 |
+
"description": "Advanced research language model for reasoning, code generation, and multilingual understanding",
|
| 7 |
+
"organization": "DeepXR",
|
| 8 |
+
"license": "Apache-2.0",
|
| 9 |
+
"status": "research",
|
| 10 |
+
"release_date": "2025-01-30",
|
| 11 |
+
"model_type": "causal language model",
|
| 12 |
+
"architecture": "LLaMA",
|
| 13 |
+
"parameters": "70B+",
|
| 14 |
+
"base_model": "meta-llama/Meta-Llama-3.1-70B"
|
| 15 |
+
},
|
| 16 |
+
"intended_use": {
|
| 17 |
+
"primary_uses": [
|
| 18 |
+
"Research in natural language processing",
|
| 19 |
+
"Advanced reasoning and problem-solving",
|
| 20 |
+
"Code generation and programming assistance",
|
| 21 |
+
"Mathematical computation and proof generation",
|
| 22 |
+
"Multilingual text understanding and generation",
|
| 23 |
+
"Scientific analysis and research assistance",
|
| 24 |
+
"Educational applications"
|
| 25 |
+
],
|
| 26 |
+
"primary_users": [
|
| 27 |
+
"AI researchers",
|
| 28 |
+
"Software developers",
|
| 29 |
+
"Data scientists",
|
| 30 |
+
"Academic researchers",
|
| 31 |
+
"Students and educators"
|
| 32 |
+
],
|
| 33 |
+
"out_of_scope": [
|
| 34 |
+
"Production systems without extensive validation",
|
| 35 |
+
"Critical decision-making without human oversight",
|
| 36 |
+
"Medical diagnosis or treatment recommendations",
|
| 37 |
+
"Legal advice or financial guidance",
|
| 38 |
+
"Real-time safety-critical applications"
|
| 39 |
+
]
|
| 40 |
+
},
|
| 41 |
+
"factors": {
|
| 42 |
+
"relevant_factors": [
|
| 43 |
+
"Input language and complexity",
|
| 44 |
+
"Task domain and specialization",
|
| 45 |
+
"Context length requirements",
|
| 46 |
+
"Computational resources available",
|
| 47 |
+
"User expertise and validation capability"
|
| 48 |
+
],
|
| 49 |
+
"evaluation_factors": [
|
| 50 |
+
"Accuracy on benchmark datasets",
|
| 51 |
+
"Reasoning capability",
|
| 52 |
+
"Code correctness",
|
| 53 |
+
"Mathematical precision",
|
| 54 |
+
"Multilingual performance",
|
| 55 |
+
"Context utilization",
|
| 56 |
+
"Generation quality"
|
| 57 |
+
]
|
| 58 |
+
},
|
| 59 |
+
"metrics": {
|
| 60 |
+
"reasoning": {
|
| 61 |
+
"MMLU": 0.847,
|
| 62 |
+
"ARC-Challenge": 0.834,
|
| 63 |
+
"HellaSwag": 0.889,
|
| 64 |
+
"WinoGrande": 0.823
|
| 65 |
+
},
|
| 66 |
+
"mathematics": {
|
| 67 |
+
"GSM8K": 0.892,
|
| 68 |
+
"MATH": 0.567,
|
| 69 |
+
"Minerva": 0.534
|
| 70 |
+
},
|
| 71 |
+
"code": {
|
| 72 |
+
"HumanEval": 0.756,
|
| 73 |
+
"MBPP": 0.723,
|
| 74 |
+
"DS-1000": 0.645
|
| 75 |
+
},
|
| 76 |
+
"knowledge": {
|
| 77 |
+
"TruthfulQA": 0.612
|
| 78 |
+
},
|
| 79 |
+
"perplexity": 2.34
|
| 80 |
+
},
|
| 81 |
+
"training_data": {
|
| 82 |
+
"note": "Training data information is proprietary to DeepXR Research",
|
| 83 |
+
"preprocessing": [
|
| 84 |
+
"Quality filtering",
|
| 85 |
+
"Deduplication",
|
| 86 |
+
"PII removal",
|
| 87 |
+
"Format standardization",
|
| 88 |
+
"Language identification",
|
| 89 |
+
"Toxicity filtering"
|
| 90 |
+
]
|
| 91 |
+
},
|
| 92 |
+
"ethical_considerations": {
|
| 93 |
+
"risks": [
|
| 94 |
+
"Potential for generating biased content",
|
| 95 |
+
"May produce factually incorrect information",
|
| 96 |
+
"Could be misused for harmful content generation",
|
| 97 |
+
"Privacy concerns with training data",
|
| 98 |
+
"Environmental impact of training and inference"
|
| 99 |
+
],
|
| 100 |
+
"mitigations": [
|
| 101 |
+
"Content filtering mechanisms",
|
| 102 |
+
"Regular bias auditing",
|
| 103 |
+
"Clear documentation of limitations",
|
| 104 |
+
"User education on responsible use",
|
| 105 |
+
"Toxicity detection and prevention",
|
| 106 |
+
"PII detection in outputs"
|
| 107 |
+
],
|
| 108 |
+
"recommendations": [
|
| 109 |
+
"Implement additional safety layers for production use",
|
| 110 |
+
"Regular monitoring and evaluation of outputs",
|
| 111 |
+
"Human oversight for critical applications",
|
| 112 |
+
"Transparency about model capabilities and limitations",
|
| 113 |
+
"Respect for user privacy and data protection"
|
| 114 |
+
]
|
| 115 |
+
},
|
| 116 |
+
"caveats_and_recommendations": {
|
| 117 |
+
"limitations": [
|
| 118 |
+
"Research model - requires validation before production use",
|
| 119 |
+
"May exhibit biases present in training data",
|
| 120 |
+
"Can generate plausible but incorrect information",
|
| 121 |
+
"Performance varies across specialized domains",
|
| 122 |
+
"Long context performance degrades beyond 64K tokens",
|
| 123 |
+
"Computational requirements are substantial",
|
| 124 |
+
"Not optimized for real-time applications"
|
| 125 |
+
],
|
| 126 |
+
"recommendations": [
|
| 127 |
+
"Always verify outputs for critical applications",
|
| 128 |
+
"Implement appropriate content filtering",
|
| 129 |
+
"Monitor for bias in specific use cases",
|
| 130 |
+
"Test thoroughly before deployment",
|
| 131 |
+
"Use temperature=0 for deterministic tasks",
|
| 132 |
+
"Implement retry logic for API failures",
|
| 133 |
+
"Consider quantization for resource constraints"
|
| 134 |
+
]
|
| 135 |
+
},
|
| 136 |
+
"technical_specifications": {
|
| 137 |
+
"context_window": 131072,
|
| 138 |
+
"vocabulary_size": 128256,
|
| 139 |
+
"hidden_size": 4096,
|
| 140 |
+
"num_layers": 32,
|
| 141 |
+
"num_attention_heads": 32,
|
| 142 |
+
"num_key_value_heads": 8,
|
| 143 |
+
"intermediate_size": 14336,
|
| 144 |
+
"rope_theta": 500000.0,
|
| 145 |
+
"rope_scaling": {
|
| 146 |
+
"type": "yarn",
|
| 147 |
+
"factor": 8.0,
|
| 148 |
+
"original_max_position_embeddings": 16384
|
| 149 |
+
},
|
| 150 |
+
"weight_format": "safetensors",
|
| 151 |
+
"supported_precisions": [
|
| 152 |
+
"fp16"
|
| 153 |
+
],
|
| 154 |
+
"quantization": "none",
|
| 155 |
+
"safetensors_shards": 82,
|
| 156 |
+
"shard_naming": "shard_01 to shard_82",
|
| 157 |
+
"shard_size_gb": 1.57,
|
| 158 |
+
"supported_frameworks": [
|
| 159 |
+
"transformers",
|
| 160 |
+
"vllm",
|
| 161 |
+
"text-generation-inference"
|
| 162 |
+
]
|
| 163 |
+
},
|
| 164 |
+
"hardware_requirements": {
|
| 165 |
+
"minimum": {
|
| 166 |
+
"gpu": "2x NVIDIA A100 80GB",
|
| 167 |
+
"vram": "160GB",
|
| 168 |
+
"ram": "256GB",
|
| 169 |
+
"storage": "500GB NVMe"
|
| 170 |
+
},
|
| 171 |
+
"recommended": {
|
| 172 |
+
"gpu": "4x NVIDIA H100 80GB",
|
| 173 |
+
"vram": "320GB",
|
| 174 |
+
"ram": "512GB",
|
| 175 |
+
"storage": "1TB+ NVMe"
|
| 176 |
+
},
|
| 177 |
+
"inference_speed": {
|
| 178 |
+
"tokens_per_second": "30-50 (depending on hardware)",
|
| 179 |
+
"latency": "100-300ms first token",
|
| 180 |
+
"throughput": "High with batch processing"
|
| 181 |
+
}
|
| 182 |
+
},
|
| 183 |
+
"model_sources": {
|
| 184 |
+
"repository": "https://huggingface.co/DeepXR/Helion-2.5-Rnd",
|
| 185 |
+
"paper": null,
|
| 186 |
+
"demo": null,
|
| 187 |
+
"organization": "https://deepxr.ai"
|
| 188 |
+
},
|
| 189 |
+
"citation": {
|
| 190 |
+
"bibtex": "@misc{helion-2.5-rnd-2025,\n title={Helion-2.5-Rnd: Advanced Research Language Model},\n author={DeepXR Research Team},\n year={2025},\n publisher={DeepXR},\n url={https://huggingface.co/DeepXR/Helion-2.5-Rnd}\n}",
|
| 191 |
+
"apa": "DeepXR Research Team. (2025). Helion-2.5-Rnd: Advanced Research Language Model. DeepXR. https://huggingface.co/DeepXR/Helion-2.5-Rnd"
|
| 192 |
+
},
|
| 193 |
+
"contact": {
|
| 194 |
+
"email": "research@deepxr.ai",
|
| 195 |
+
"website": "https://deepxr.ai",
|
| 196 |
+
"github": "https://github.com/DeepXR",
|
| 197 |
+
"support": "support@deepxr.ai"
|
| 198 |
+
},
|
| 199 |
+
"additional_information": {
|
| 200 |
+
"languages_supported": [
|
| 201 |
+
"English", "Spanish", "French", "German", "Italian", "Portuguese",
|
| 202 |
+
"Chinese (Simplified)", "Chinese (Traditional)", "Japanese", "Korean",
|
| 203 |
+
"Russian", "Arabic", "Hindi", "Bengali", "Turkish", "Vietnamese",
|
| 204 |
+
"Polish", "Ukrainian", "Romanian", "Dutch", "Greek", "Czech",
|
| 205 |
+
"Swedish", "Hungarian", "Finnish", "Norwegian", "Danish", "Hebrew",
|
| 206 |
+
"Thai", "Indonesian", "Malay", "Filipino", "Persian", "Urdu",
|
| 207 |
+
"Tamil", "Telugu", "Kannada", "Malayalam", "Gujarati", "Marathi",
|
| 208 |
+
"Punjabi", "Swahili", "Amharic", "Yoruba", "Igbo", "Hausa"
|
| 209 |
+
],
|
| 210 |
+
"programming_languages": [
|
| 211 |
+
"Python", "JavaScript", "TypeScript", "Java", "C++", "C#", "Go",
|
| 212 |
+
"Rust", "Swift", "Kotlin", "Ruby", "PHP", "Scala", "R", "MATLAB",
|
| 213 |
+
"SQL", "Shell", "PowerShell", "HTML", "CSS", "LaTeX"
|
| 214 |
+
],
|
| 215 |
+
"deployment_options": [
|
| 216 |
+
"Docker containers",
|
| 217 |
+
"Kubernetes clusters",
|
| 218 |
+
"Cloud platforms (AWS, GCP, Azure)",
|
| 219 |
+
"On-premise servers",
|
| 220 |
+
"API endpoints",
|
| 221 |
+
"Batch processing pipelines"
|
| 222 |
+
],
|
| 223 |
+
"monitoring_tools": [
|
| 224 |
+
"Prometheus metrics",
|
| 225 |
+
"Grafana dashboards",
|
| 226 |
+
"Custom logging",
|
| 227 |
+
"Performance profiling",
|
| 228 |
+
"Token usage tracking"
|
| 229 |
+
]
|
| 230 |
+
}
|
| 231 |
+
}
|