DOCSEARCHbyAlgolia
Submit Docs

yyixx

website·Indexed 31 days ago

https://yyixx.com

A personal website featuring a curated collection of AI tools, coding agents, and ML deployment frameworks, alongside personal essays and technical documentation on MDX and Markdown.

Tokens
178.6K
Snippets
1K
Records
1.2K
Agent score
56%

What's inside yyixx

  1. Overview of Superpowers software development methodology

    Superpowers is a complete software development methodology designed to enable coding agents (AI agents) to effectively assist in the software development process.
    # Repository for Superpowers methodology
git clone https://github.com/obra/superpowers

  2. Overview of BentoML for AI Model Serving

    BentoML is a flexible AI application serving framework designed to simplify the process of packaging and deploying models into production-grade API services. It is specifically focused on ease of use and application building, making it suitable for creating complete AI cloud services, including Model Inference APIs, Job queues, LLM applications, and multi-model pipelines.
    # Source code and documentation available at:
https://github.com/bentoml/BentoML

  3. Overview of AI Large Model Career Paths

    The path from training to inference in AI Large Models consists of four primary technical domains. While AI Algorithms and Chips are the 'dreamers' and 'builders,' AI Infra and Compilers are the 'deliverers' that make models stable, efficient, and affordable for production.
    AI Algorithm -> AI Infra (Infrastructure) -> AI Compiler -> AI Chip

  4. Overview of the Claude Code Agentic Workflow

    A comprehensive software engineering workflow for Claude Code is structured into four layers to move from 'Vibe Coding' to 'Agentic Engineering':

    1. Behavioral Layer: Defines AI behavioral guidelines to standardize Agent actions (e.g., Andrej Karpathy Skills).
    2. Process Management Layer: Provides 'driving' modes for development, ranging from fully autonomous (Superpowers) to manual orchestration of virtual engineering teams (gstack).
    3. Specialized Skill Library: Task-specific skill sets for professional engineers (e.g., Matt Pocock and Addy Osmani skills).
    4. Code Intelligence Layer: Provides deep context and code knowledge graphs to support all upper layers (e.g., GitNexus).

  5. Overview of Hermes Agent by Nous Research

    Hermes Agent is a self-improving AI agent developed by Nous Research. The project is hosted on GitHub.
    # Project repository
git clone https://github.com/nousresearch/hermes-agent

  6. Overview of OpenAI Triton

    Triton is a language and compiler for parallel programming designed to provide a Python-based DSL for writing custom DNN compute kernels. It allows developers to achieve maximal throughput on modern GPU hardware while maintaining productivity. It serves two primary roles: as a tool for algorithm researchers to implement high-performance operators within the PyTorch ecosystem, and as a compiler framework for hardware developers to build compilation infrastructure for custom AI chips using MLIR.

  7. Overview of nndeploy AI Deployment Framework

    nndeploy is a high-performance AI deployment framework designed to streamline the implementation of AI algorithms across various platforms and hardware using visual workflows and multi-terminal inference. It specifically focuses on edge deployment, supporting the following environments:

    • Desktop: Windows, macOS
    • Mobile: Android, iOS
    • Edge Computing Devices: NVIDIA Jetson, Ascend310B, Rockchip (RK), etc.
    • Single-machine Servers: NVIDIA RTX series, T4, Ascend310P, etc.

  8. Overview of LLM architectural evolution components

    Modern Large Language Models (LLMs) have evolved across five core dimensions to improve efficiency and performance:

    • Architecture: Shifted from Encoder-Decoder (original Seq2Seq) to Decoder-only (Causal), which is the current standard validated by Scaling Laws.
    • Attention: Evolved from MHA (Multi-Head Attention) $\rightarrow$ MQA (Multi-Query Attention) $\rightarrow$ GQA (Grouped-Query Attention) $\rightarrow$ MLA (Multi-head Latent Attention) to optimize KV Cache size and memory compression. Computational optimizations include FlashAttention, Sliding Window Attention, and Sparse Attention.
    • Positional Encoding: Evolved from Absolute and Relative encoding to RoPE (Rotary Positional Embedding) and ALiBi, with RoPE being the current mainstream for better length extrapolation.
    • Activation Functions: Evolved from ReLU $\rightarrow$ GeLU $\rightarrow$ SwiGLU to enhance non-linearity.
    • Normalization: Use of RMSNorm to ensure training stability.

  9. Overview of Istio Service Mesh

    Istio is an open-source service mesh that layers transparently onto existing distributed applications. It provides a unified way to secure, connect, and monitor services without requiring significant changes to service code. Key capabilities include:

    • Security: Service-to-service communication within clusters using TLS encryption, strong identity authentication, and authorization.
    • Traffic Management: Automatic load balancing for HTTP, gRPC, WebSocket, and TCP traffic, with fine-grained control via routing rules, retries, failovers, and fault injection.
    • Policy Enforcement: A pluggable policy layer and configuration API for access control, rate limiting, and quotas.
    • Observability: Automatic metrics, logging, and tracing for all cluster traffic, including ingress and egress.

    Istio's control plane runs on Kubernetes and can manage applications within the same cluster, extend to other clusters, or connect to VMs and endpoints outside of Kubernetes.

  10. Overview of LM Studio for local AI model execution

    LM Studio is a tool that allows developers and users to run large language models (AI models) locally on their own hardware, ensuring that data remains private and does not leave the local machine.
    Visit https://lmstudio.ai/ for installation and more details.

  11. Overview of Hugging Face Skills specification

    Hugging Face Skills is a set of definition specifications used to standardize and describe AI/ML tasks, including dataset creation, model training, and evaluation.
    # Repository for Hugging Face Skills
git clone https://github.com/huggingface/skills

  12. Overview of VideoPipe video analysis framework

    VideoPipe is a cross-platform video structuring (video analysis) framework designed for rapid integration and deployment. It is suitable for scenarios such as video structuring, image search, face recognition, and behavior analysis in traffic or security domains (e.g., traffic event detection). The framework handles core video pipeline tasks including stream pulling (ingestion) and stream pushing (egress).