Sixfinger API

Python ⭐ 3

⚑ Ultra-Fast AI Platform - 10-20x faster than popular AI services with 13 powerful models including Meta Llama 3.3 70B, Qwen3 32B, and DeepSeek-R1. Achieves 100-150 tokens/sec throughput through advanced optimizations including model quantization (8-bit/4-bit), intelligent request batching, KV cache optimization, and speculative decoding. Features OpenAI-compatible API, real-time SSE streaming, async processing with FastAPI, and comprehensive model selection from 1B to 70B parameters. Perfect for developers needing fast, cost-effective AI inference without sacrificing quality.

ai api chatgpt-api deepseek llama qwen free-ai-api
View on GitHub β†’

TurboTensors

Python ⭐ 1

πŸš€ High-performance CPU Inference Engine for Large Language Models achieving 15-20x speedup over standard PyTorch CPU implementations. Implements custom Numba-JIT compiled kernels for matrix operations, SIMD vectorization (AVX2/AVX-512), asymmetric quantization with per-channel scaling, and intelligent memory pooling. Specifically optimized for Turkish language models with specialized embedding lookup optimizations. Enables running 3B parameter models at interactive speeds (< 100ms/token) on commodity CPU hardware, making AI accessible without expensive GPU infrastructure.

cpu-inference llm numba optimization turkish-nlp
View on GitHub β†’

Genesis

Python ⭐ 1

🧬 Evolutionary Code Optimization Engine using genetic algorithms to automatically improve Python function performance. Operates on Abstract Syntax Trees (AST) to perform intelligent mutations including operator replacement, loop restructuring, algorithmic substitutions, and data structure optimizations. Includes multi-dimensional fitness evaluation (execution time, memory usage, code complexity) with mandatory correctness testing. Successfully achieved 1000x speedup on recursive functions through automatic memoization discovery and 50x improvement on data pipelines by discovering NumPy vectorization opportunities.

genetic-algorithm code-optimization ast python
View on GitHub β†’

SixFinger-8B

Python ⭐ 1

πŸ€– SixFinger AI 8B - Custom-trained 8 billion parameter language model fine-tuned for instruction following and conversational AI. Based on modern transformer architecture with optimizations for efficient inference. Trained on diverse datasets including technical documentation, programming, and general knowledge. Supports multi-turn conversations, code generation, technical explanations, and creative writing. Available in both full-precision and quantized formats for different deployment scenarios.

ai-model language-model transformer nlp
View on GitHub β†’

Turkish Sentiment Analysis

Python ⭐ 1

πŸ“Š State-of-the-art Turkish sentiment analysis using fine-tuned BERTurk (Turkish BERT). Achieves 89.2% accuracy on multi-class sentiment classification (positive/negative/neutral). Handles Turkish-specific challenges including agglutinative morphology, vowel harmony, and informal language variations. Includes comprehensive preprocessing pipeline with deasciification, normalization, and morphological analysis. Trained on 50K+ manually labeled samples from social media, e-commerce reviews, and news comments. Supports real-time inference with < 50ms latency per document.

bert nlp sentiment-analysis turkish deep-learning
View on GitHub β†’

Turkish NLP Toolkit

Python ⭐ 1

πŸ‡ΉπŸ‡· Comprehensive Natural Language Processing toolkit for Turkish language with morphological analysis, text normalization, named entity recognition, and custom tokenization. Handles Turkish-specific linguistic features: agglutinative morphology (words with 10+ suffixes), vowel harmony rules, case marking, and possessive constructions. Includes stemmer/lemmatizer achieving 94.7% root extraction accuracy, POS tagger at 92.3% accuracy, and NER for person/organization/location entities. Optimized for both formal written Turkish and informal social media text.

nlp turkish text-processing morphology tokenization
View on GitHub β†’

kayra-core

Python ⭐ 1

πŸ”¬ Experimental Turkish Language Model focused on instruction tuning research and hallucination behavior analysis in low-parameter LLMs. Investigates how model size affects instruction following capability and factual accuracy in Turkish. Tests various training approaches: supervised fine-tuning, reinforcement learning from human feedback (RLHF), and direct preference optimization (DPO). Documents hallucination patterns specific to Turkish language generation and explores mitigation strategies. Serves as research platform for understanding Turkish LLM behavior.

turkish-llm language-model experiment instruction-tuning research
View on GitHub β†’

Flask-Template

HTML ⭐ 1

🌐 Modern logistics company landing page template built with Flask featuring responsive design, user authentication, and admin dashboard. Includes complete booking system with shipment tracking, real-time status updates, and payment integration. Features dark/light theme toggle, mobile-first responsive design, form validation, and SEO optimization. Perfect starting point for logistics, shipping, or delivery service websites with professional UI components and clean code architecture.

flask web template responsive-design authentication
View on GitHub β†’

SF-Circle

HTML ⭐ 1

🎯 Interactive web game challenging users to draw perfect circles, inspired by Neal.fun. Features real-time accuracy scoring using geometric analysis, global leaderboard with user rankings, smooth canvas drawing with touch and mouse support, and detailed performance metrics. Implements advanced circle detection algorithms comparing user-drawn shapes to ideal circles using least-squares fitting. Stores high scores locally and syncs with server-side leaderboard. Includes animated feedback and achievement system for different accuracy levels.

game interactive web-app javascript canvas
View on GitHub β†’

Picart

JavaScript ⭐ 1

🎨 Feature-rich browser-based image editor with comprehensive editing capabilities including filters, transformations, layers, and effects. Supports multiple image formats (PNG, JPG, WebP), implements non-destructive editing with undo/redo, provides 20+ filters (blur, sharpen, brightness, contrast, saturation, etc.), and includes drawing tools (brush, shapes, text). Features layer management system, selection tools, crop and resize, color adjustments, and export options. Built with vanilla JavaScript for performance without framework overhead.

image-editor web-app canvas image-processing filters
View on GitHub β†’