AI & ML interests
Power To Defend Infinity ๐ค Becoming To Defend Infinity ๐
Recent Activity
Organization Card
๐ SKT AI ๐
SKT AI is a cutting-edge research and development hub dedicated to pushing the boundaries of machine learning and neural architecture. Led by Shrijan Kumar Tiwari, the lab focuses on creating high-performance models that are faster, smarter, and more efficient.
โก THE APEX OF PERFORMANCE: SKT SURYA (22T)
The Surya series represents a major leap in balancing massive scale with localized precision.
๐ 1. Performance & Inference Speed
- High-Speed Throughput: Capable of delivering 110+ tokens per second on standard optimized setups.
- Low Latency: Maintains real-time responsiveness ideal for enterprise-level deployments.
๐ฎ๐ณ 2. Contextual Nuance (The Indian Edge)
- Linguistic Mastery: Beyond simple translation; it masters local slang, idioms, and cultural references of the Indian subcontinent.
- Benchmark Superiority: Excels where global models fail due to specialized localized training data.
๐ ๏ธ 3. Architectural Optimization & VRAM Efficiency
- Smart Engineering: Optimized to run on 24GBโ32GB GPUs (like RTX 3090/4090), whereas similar models require 40GB+.
- Accessible Power: High-tier AI performance now runs on consumer-grade enterprise hardware.
๐ฏ 4. The "100% Accuracy" Mandate
- Zero-Hallucination Focus: Advanced instruction-following and fine-tuning to eliminate errors.
- Technical Precision: Perfect reliability for coding, math, and complex logic tasks.
๐ MAIN PROJECTS
| Project | Capabilities | Status |
|---|---|---|
| ST-X | Text Generation + Image Understanding | Small but Powerful |
| SKT-OMNI | Multi-Modal Vision + Text Generation (1 Trillon) | Open Sourced Now |
| SKT SURYA | Vision + Audio + Video + Text (22 Trillon Params) | IN PROGRESS TRAINING COST VERY HIGH. Private Model Soon Published |
๐ MODEL SPECIFICATIONS
- Model Type: sแดแด
- Specialization: Advanced Coding, Logic, and Hinglish Mastery
- Base Framework: SKT_ST-X-0
- Quantization: GGUF, EXL2, AWQ
- Main Dataset: 10M+ Hinglish Dataset (100x Reasoning, 10x Deeper Coding)
๐ PERFORMANCE BENCHMARKS
| Metric | Score | Status |
|---|---|---|
| Reasoning (GSM8K) | 97.2% | โ ELITE |
| Coding (HumanEval) | 90.7% | โ PRO |
| Logic (MMLU) | 87.1% | โ SUPREME |
๐ EXPLORE THE MODELS
[ ๐ SKT AI Labs ] | [ ๐งฌ าสแดแด ษชษดแด ษชแด ๐ฎ๐ณ ]
models 0
None public yet
datasets 0
None public yet