Helion-V2.5-Rnd / metadata.json
Trouter-Library's picture
Create metadata.json
e9b17f1 verified
raw
history blame
7.23 kB
{
"model_id": "DeepXR/Helion-2.5-Rnd",
"model_name": "Helion-2.5-Rnd",
"full_name": "Helion 2.5 Research and Development",
"organization": "DeepXR",
"release_date": "2025-01-30",
"version": "2.5.0-rnd",
"status": "research",
"description": "Advanced research language model with 70B parameters, designed for exceptional performance across reasoning, code generation, mathematics, and multilingual understanding with 131K context window.",
"architecture": {
"type": "transformer",
"variant": "llama",
"parameters": "70B",
"layers": 32,
"hidden_size": 4096,
"attention_heads": 32,
"kv_heads": 8,
"intermediate_size": 14336,
"vocabulary_size": 128256,
"context_length": 131072,
"rope_theta": 500000,
"positional_encoding": "YARN",
"activation": "SiLU",
"normalization": "RMSNorm"
},
"capabilities": {
"text_generation": {
"enabled": true,
"quality": "high",
"max_length": 131072
},
"code_generation": {
"enabled": true,
"languages": [
"Python", "JavaScript", "TypeScript", "Java", "C++", "C#", "Go",
"Rust", "Swift", "Kotlin", "Ruby", "PHP", "Scala", "R"
],
"quality": "high"
},
"mathematics": {
"enabled": true,
"capabilities": [
"arithmetic", "algebra", "calculus", "statistics", "proof_generation"
],
"quality": "high"
},
"reasoning": {
"enabled": true,
"types": [
"logical", "analytical", "common_sense", "abstract"
],
"quality": "high"
},
"multilingual": {
"enabled": true,
"languages": 50,
"primary_languages": [
"English", "Spanish", "French", "German", "Chinese", "Japanese",
"Korean", "Russian", "Arabic", "Hindi", "Portuguese", "Italian"
]
},
"long_context": {
"enabled": true,
"max_tokens": 131072,
"performance": "optimized"
}
},
"performance": {
"benchmarks": {
"mmlu": {
"score": 0.847,
"description": "Massive Multitask Language Understanding"
},
"gsm8k": {
"score": 0.892,
"description": "Grade School Math 8K"
},
"humaneval": {
"score": 0.756,
"description": "Code Generation Accuracy"
},
"mbpp": {
"score": 0.723,
"description": "Python Programming Benchmark"
},
"arc_challenge": {
"score": 0.834,
"description": "ARC Challenge Reasoning"
},
"hellaswag": {
"score": 0.889,
"description": "Common Sense Inference"
},
"winogrande": {
"score": 0.823,
"description": "Commonsense Reasoning"
},
"truthfulqa": {
"score": 0.612,
"description": "Truthfulness in QA"
}
},
"inference": {
"throughput_tokens_per_second": "30-50",
"latency_first_token_ms": "100-300",
"optimal_batch_size": "1-32",
"memory_requirement_gb": 140
}
},
"technical_details": {
"precision": "bfloat16",
"weight_format": "safetensors",
"total_shards": 96,
"shard_size_avg_gb": 1.46,
"total_size_gb": 140,
"quantization": "none",
"optimization": [
"Flash Attention 2",
"Grouped Query Attention",
"Tensor Parallelism",
"Pipeline Parallelism"
]
},
"training": {
"steps": 150000,
"warmup_steps": 2000,
"learning_rate": 2e-05,
"optimizer": "AdamW",
"scheduler": "cosine_with_restarts",
"precision": "bfloat16",
"gradient_accumulation": 8,
"batch_size": 4,
"parallelization": {
"tensor_parallel": 4,
"pipeline_parallel": 2
}
},
"hardware_requirements": {
"minimum": {
"gpus": "2x NVIDIA A100 80GB",
"vram_gb": 160,
"ram_gb": 256,
"storage_gb": 500,
"network": "10Gbps"
},
"recommended": {
"gpus": "4x NVIDIA H100 80GB",
"vram_gb": 320,
"ram_gb": 512,
"storage_gb": 1000,
"network": "100Gbps InfiniBand"
}
},
"usage": {
"intended_uses": [
"Research and development",
"Advanced reasoning tasks",
"Code generation and analysis",
"Mathematical problem solving",
"Multilingual applications",
"Long document understanding",
"Creative writing",
"Educational purposes"
],
"not_recommended": [
"Production without validation",
"Critical decision-making without oversight",
"Medical diagnosis",
"Legal advice",
"Financial advice",
"Safety-critical systems"
]
},
"limitations": [
"Research model - requires validation",
"May exhibit training data biases",
"Can generate incorrect information",
"Performance varies by domain",
"Context degradation beyond 64K tokens",
"Requires significant compute resources"
],
"ethical_considerations": {
"bias_mitigation": "Ongoing evaluation and monitoring",
"safety_features": [
"Content filtering",
"PII detection",
"Toxicity monitoring",
"Prompt injection protection"
],
"responsible_use": [
"Verify outputs for critical applications",
"Monitor for bias",
"Implement content filtering",
"Respect privacy and data protection"
]
},
"license": {
"type": "Apache-2.0",
"url": "https://www.apache.org/licenses/LICENSE-2.0",
"commercial_use": true,
"modification": true,
"distribution": true,
"patent_use": true,
"private_use": true
},
"files": {
"safetensors": {
"format": "safetensors",
"num_shards": 96,
"pattern": "model-{:05d}-of-00096.safetensors",
"index_file": "model.safetensors.index.json",
"checksums_available": true
},
"config": [
"config.json",
"generation_config.json",
"tokenizer_config.json",
"model_config.yaml"
],
"inference": [
"inference/server.py",
"inference/client.py",
"inference/utils.py",
"inference/security.py",
"inference/evaluate.py",
"inference/batch_inference.py",
"inference/optimizer.py",
"inference/benchmark.py"
]
},
"links": {
"repository": "https://huggingface.co/DeepXR/Helion-2.5-Rnd",
"organization": "https://deepxr.ai",
"documentation": "https://docs.deepxr.ai/helion",
"paper": null,
"demo": null
},
"contact": {
"email": "[email protected]",
"research_email": "[email protected]",
"security_email": "[email protected]",
"website": "https://deepxr.ai"
},
"citation": {
"format": "bibtex",
"text": "@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}"
},
"changelog": [
{
"version": "2.5.0-rnd",
"date": "2025-01-30",
"changes": [
"Initial research release",
"70B parameter model",
"131K context window with YARN",
"SafeTensors format (96 shards)",
"Comprehensive inference suite",
"Security implementation",
"Optimization tools"
]
}
]
}