惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Recent Announcements
Recent Announcements
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
量子位
博客园 - 司徒正美
Security Archives - TechRepublic
Security Archives - TechRepublic
P
Palo Alto Networks Blog
cs.CL updates on arXiv.org
cs.CL updates on arXiv.org
Cyberwarzone
Cyberwarzone
小众软件
小众软件
T
Threatpost
Latest news
Latest news
J
Java Code Geeks
博客园 - Franky
博客园 - 三生石上(FineUI控件)
Project Zero
Project Zero
P
Privacy & Cybersecurity Law Blog
T
Tenable Blog
L
Lohrmann on Cybersecurity
大猫的无限游戏
大猫的无限游戏
WordPress大学
WordPress大学
Apple Machine Learning Research
Apple Machine Learning Research
Scott Helme
Scott Helme
Simon Willison's Weblog
Simon Willison's Weblog
C
CXSECURITY Database RSS Feed - CXSecurity.com
P
Privacy International News Feed
人人都是产品经理
人人都是产品经理
S
Schneier on Security
T
The Blog of Author Tim Ferriss
V
V2EX
有赞技术团队
有赞技术团队
Y
Y Combinator Blog
罗磊的独立博客
IT之家
IT之家
雷峰网
雷峰网
H
Help Net Security
C
Cyber Attacks, Cyber Crime and Cyber Security
T
Tor Project blog
C
Cybersecurity and Infrastructure Security Agency CISA
I
InfoQ
GbyAI
GbyAI
博客园 - 叶小钗
PCI Perspectives
PCI Perspectives
The GitHub Blog
The GitHub Blog
Martin Fowler
Martin Fowler
H
Heimdal Security Blog
Spread Privacy
Spread Privacy
博客园_首页
A
About on SuperTechFans
T
Tailwind CSS Blog
The Register - Security
The Register - Security

Hacker News - Newest: "LLM"

GitHub - lechmazur/position_bias: A benchmark for testing whether LLM judges keep the same preference when two lightly edited versions of the same story are shown in opposite orders. Flex routing (EU and EFTA) Dark Factories: Retooling for LLM Velocity Ask HN: What would be the impact of a LLM output injection attack? GitHub - AronDaron/dataset-generator: No-code desktop app for generating high-quality synthetic datasets to fine-tune LLMs — plan-then-execute pipeline, LLM-as-judge, HuggingFace upload. I thought I had a bug GitHub - Oaklight/llm-rosetta: Production-ready LLM API translation layer for Python — bidirectional conversion between OpenAI, Anthropic & Google formats via hub-and-spoke IR. Optional API gateway. Streaming & non-streaming. Zero core deps. Contributions welcome! GitHub - browser-use/browser-harness: Self-healing browser harness that enables LLMs to complete any task. GitHub - moeen-mahmud/remen: Remen turns thoughts into something you can return to Analyzing 156 LLM Launch Posts on Hacker News ChatGPT vs Gemini vs Claude: The Best LLM Subscription You Should Buy GitHub - salaamalykum/quran-semantic-search: High-density RAG Semantic Search Engine & Quran Corpus (GEO/SEO Architecture) GitHub - NVIDIA/TensorRT-LLM: TensorRT LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and supports state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT LLM also contains components to create Python and C++ runtimes that orchestrate the inference execution in a performant way. The State of LLM Bug Bounties in 2026 Operational Readiness Criteria for Tool-Using LLM Agents Meshcore: Architecture for a Decentralized P2P LLM Inference Network How an LLM becomes more coherent as we train it GitHub - seetrex-ai/laimark GitHub - Jossifresben/BibCrit: AI-assited biblical textual criticism GitHub - wastedcode/memex: File system based wiki, maintained by Claude 99helpers.com GitHub - cliver-project/AITrigram GitHub - unbody-io/adapt: A self-evolving memory layer for AI agents. GitHub - hb20007/awesome-gen-ai-fails: A list of incidents where reliance on generative AI and LLMs resulted in harm to companies, individuals, or society GitHub - nevenkordic/localmind: Run any local LLM with persistent memory and context. CLI agent over Ollama with SQLite-backed hybrid recall. No cloud. Ask HN: What are the machine requirements for a LLM like Llama-3.1-8B? Faster LLM Inference via Sequential Monte Carlo grpo explained: group relative policy optimization for llm finetuning - cgft Stop comparing price per million tokens: the hidden LLM API costs · TensorZero Andrej Karpathy's LLM Wiki Is a Bad Idea GitHub - GG-QandV/mnemostroma: Offline RAM-first cognitive leer/coprocessor for AI agents and robotics. Solves "Context Abandonment" with 20-80ms latency using a dual-thread biomimetic memory architecture (ONNX + SQLite WAL). mempalace/agent at agent · skorotkiewicz/mempalace GitHub - Nyquest-ai/nyquest-rust-fullstack-pub: Nyquest — Semantic Compression Proxy for LLMs. 350+ rules, local LLM stage, 15-75% token savings. Full Rust stack. GitHub - TheoV823/mneme: Enforce architectural decisions in AI-assisted development. GitHub - klemenvod/TokenBrawl: A 1v1 Bomberman-style game where two LLM agents play autonomously against each other. No human plays — you watch the AIs fight. Each agent receives a text description of the board state, reasons about it, and outputs a move as JSON. The game engine executes it. Introducing the Common AI Provider: LLM and AI Agent Support for Apache Airflow Power Circuit AI: Designing Power Electronic Circuits for Motor Drives with Generative Artificial Intelligence Ask HN: How to program with IDE and LLM on CPU locally? Show HN: Agent-cache – Multi-tier LLM/tool/session caching for Valkey and Redis Bonsai 1-bit WebGPU - a Hugging Face Space by webml-community The LLM Fallacy: Misattribution in AI-Assisted Cognitive Workflows Ask HN: Simple tooling for local LLM code critique without IDE integration? Can a General LLM Diagnose a DICOM Slice? A 10-Case Public Benchmark Charts-of-Thought: Enhancing LLM Visualization Literacy (PDF, 2026) GitHub - Mesh-LLM/mesh-llm: Distributed AI/LLM for the people. Share compute privately or publicly to power your agents and chat. GitHub - seamus-brady/springdrift: A persistent runtime for long-lived LLM agents Writing an LLM from scratch, part 32k -- Interventions: training a better model locally with gradient accumulation Ask HN: Which LLM model and agentic CLI are you using for local development? GitHub - wayneColt/modelcascade: Route local. Escalate smart. Never overspend. Open-source multi-model cascade routing for autonomous agents. LLM pricing is 100x harder than you think GitHub - asakin/llm-primer: Pre-warmed Claude Code sessions in tmux. No startup wait. GitHub - EggerMarc/chat-rs: A multi-provider LLM framework for Rust. GitHub - SynapseKit/SynapseKit: Minimal, async-first Python framework for production LLM apps- 2 hard deps, no magic, no SaaS. A Claude Skill that Makes LLM Paragraphs More Bearable Does Gas Town 'steal' usage from users' LLM credits & paid services to improve itself? What's Claude Code Actually Doing? Open the Black Box with the Arthur Engine Milla Jovovich's New Open Source LLM Memory App and the Dark Code Problem Your intuition of LLM token usage might be wrong Show HN: Bloomberg Terminal for LLM ops – free and open source GitHub - 0xchamin/mcptube: Transform YouTube videos into a compounding knowledge base with transcripts, vision analysis, and agentic search. Works as an MCP server for Claude, Copilot & more. Show HN: Open KB: Open LLM Knowledge Base Your LLM is a compiler, not a runtime GitHub - sapountzis/Unslop: A Web Feed That Deserves You crates.io: Rust Package Registry Beyond Karpathy's LLM-Wiki: The Necessity of Cognitive Governance GitHub - amitshekhariitbhu/llm-internals: Learn LLM internals step by step - from tokenization to attention to inference optimization. GitHub - parallem-ai/parallem: An expressive library for running agents with the Batch API. GitHub - stfurkan/pi-llm LLM-Wiki Show HN: Formal – Formal verification for AI-generated code using Lean 4 LRTS – Regression testing for LLM prompts (open source, local-first) LLM Wiki Skill: Build a Second Brain with Claude Code and Obsidian I built an LLM Wiki and RAG solution: here's a demo for a security KB The biggest advance in AI since the LLM Predict-Rlm: The LLM Runtime That Lets Models Write Their Own Control Flow the-synthetic-library/the-synthetic-mind at main · joshferrer1/the-synthetic-library GitHub - yisding/reviewwiggum GitHub - Donnyb369/mcp-spine: Context Minifier & State Guard — Local-first MCP middleware proxy GitHub - anitiue/Hindsight: An experience-driven self-improvement framework for LLM agents — 基于经验的 LLM Agent 自我改进框架 GitHub - stef41/lmscan: 🔍 Detect AI-generated text and fingerprint which LLM wrote it. Open-source GPTZero alternative. Zero dependencies, works offline. GitHub - alainnothere/AmdPerformanceTesting: Amd Performance Testing Ask HN: Is a purely Markdown-based CRM a terrible idea? Optimized for LLM agents Context Engineering - LLM Memory and Retrieval for AI Agents | Weaviate little_helper_tui/letter.md at main · sleepyeldrazi/little_helper_tui GitHub - EvanZhouDev/umr: The Unified Model Registry for all your local AI apps. GitHub - JordanCT/VigIA-Orchestrator Your Agent Is Mine: Measuring Malicious Intermediary Attacks on the LLM Supply Chain A Taxonomy of RL Environments for LLM Agents Llama LLM Network Feture GitHub - genedeng-ca/ai-mac-migration: AI-powered Mac-to-Mac migration tool - replace Apple Migration Assistant with intelligent, selective transfer using local LLMs GitHub - lunargate-ai/gateway: High-performance self-hosted AI gateway (OpenAI-compatible) with routing, retries, and streaming GitHub - AuthBits/webmcp: A lightweight, prompt-driven MCP web research server for high-quality LLM powered information extraction. Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering Springdrift: An Auditable Persistent Runtime for LLM Agents with Case-Based Memory, Normative Safety, and Ambient Self-Perception High-Stakes Personalization: Rethinking LLM Customization for Individual Investor Decision-Making From Static Templates to Dynamic Runtime Graphs: A Survey of Workflow Optimization for LLM Agents HUOZIIME: An On-Device LLM-enhanced Input Method for Deep Personalization TIDE: Token-Informed Depth Execution for Per-Token Early Exit in LLM Inference Characterizing WebGPU Dispatch Overhead for LLM Inference Across Four GPU Vendors, Three Backends, and Three Browsers LLM Targeted Underperformance Disproportionately Impacts Vulnerable Users
GitHub - Beledarian/wgpu-llm: A from-scratch LLM inference engine that uses wgpu (the cross-platform WebGPU implementation) to dispatch WGSL compute shaders for every math operation a Transformer needs. No CUDA. No Python. No massive framework dependencies. Just Rust, raw shaders, and your GPU.
2026-04-11 · via Hacker News - Newest: "LLM"

A minimalist Llama inference engine written in Rust and WGSL.
Runs locally on any GPU — no CUDA required.


wgpu-llm generating 24 tokens/sec natively on Snapdragon Adreno

What Is This?

A from-scratch LLM inference engine that uses wgpu v29 to dispatch WGSL compute shaders for the full Transformer forward pass. No CUDA. No Python. No framework dependencies. Just Rust, raw shaders, and your GPU.

Target architecture: Llama (TinyLlama, Llama 2/3, and compatible fine-tunes).
Tested with: TinyLlama-1.1B-Chat-v1.0

Benchmarks (TinyLlama 1.1B, 256 tokens, default sampling)

GPU Mode VRAM (weights) tok/s
RTX 3090 (Vulkan) f16-weights ~2.05 GiB 66+
Adreno X1-85 (Vulkan) INT8-b64 1.27 GiB 32.8
Adreno X1-85 (Vulkan) f16-weights 2.05 GiB 25.5

Why?

  • The Copilot+ Hardware Gap — I recently got a Snapdragon X Elite Copilot+ laptop and quickly realized my GPU was effectively a paperweight for local AI. Standard tools like LM Studio and the massive PyTorch ecosystem didn't support the integrated Adreno GPU, forcing everything onto the CPU. I didn't want to wait for the ecosystem to catch up, so I bypassed it entirely.
  • Portability — runs on Windows, macOS, Linux, and anything with a Vulkan/Metal/DX12 driver.
  • Educational — every layer of the Transformer is visible as a standalone WGSL shader. No framework magic.
  • Hackable — small codebase, clear data flow, easy to experiment with.

How It Was Built

wgpu-llm is an experiment in highly leveraged, AI-accelerated engineering. Starting from a detailed, human-designed architectural blueprint, the engine was scaffolded into working code in under 16 hours using a custom LLM orchestration pipeline with a human-in-the-loop approach.

To ensure the AI could write complex, low-level code without hallucinating or losing track of the architecture, the orchestrator was augmented with my own open-source tools:

  • MCP_WGSL_Docs: Fed the LLM the exact, up-to-date WGSL specification so it could accurately implement features like tiled GEMM and compile-time f32/f16 switching across 12 standalone compute shaders.
  • mcp-local-memory: Maintained persistent, strict architectural context across the entire 16-hour development sprint, ensuring the Rust host code and WGSL shaders stayed perfectly aligned.

Installation

Prerequisites: Rust and Cargo + a GPU with Vulkan, Metal, or DX12 support.

Install from crates.io

cargo install wgpu-llm

Build from source

git clone https://github.com/Beledarian/wgpu-llm.git
cd wgpu-llm
cargo build --release

Quick Start

Download a compatible model (e.g., TinyLlama-1.1B-Chat) — the model directory needs config.json, tokenizer.json, and *.safetensors.

# If installed via cargo install:
wgpu-llm --model-dir /path/to/model --prompt "The overarching philosophy of stoicism teaches us" --max-tokens 256 --f16-weights

# If built from source:
cargo run --release --bin wgpu-llm -- --model-dir /path/to/model --prompt "The overarching philosophy of stoicism teaches us" --max-tokens 256 --f16-weights

The engine auto-detects your GPU capabilities, loads weights, and streams generated text to stdout. After generation it prints a telemetry summary (VRAM breakdown, tok/s, timing).

INT8 mode (~2× VRAM reduction):

# 1. Quantize your model (one-time step, requires Python + safetensors)
python scripts/quantize_int8.py /path/to/model /path/to/model-int8 --block-size 64

# 2. Run with quantized weights
wgpu-llm --model-dir /path/to/model-int8 --prompt "..." --max-tokens 256 --gemm-int8-block-size 64

For the full list of CLI flags (sampling, INT8 GEMM, KV spill, memory budget, dry-run), see the CLI Reference.

Current Limitations

  • Alpha quality — functional but not production-hardened
  • Decode-only — prompt tokens are processed sequentially (no batched prefill yet)
  • Single-sequence — no batching or concurrent requests
  • INT8 lightly tested — verified with TinyLlama 1.1B; quality may vary with larger models
  • CPU-side sampling — logits read back to CPU for top-k/top-p/temperature; fast enough for single-sequence but not optimal
  • No chat template — raw text completion only; no multi-turn conversation support yet
  • Timing is wall-clock — reported tok/s includes CPU overhead, not isolated GPU kernel time

Architecture at a Glance

Text → Tokenizer → Embedding → [N × Transformer Layers] → Logits → Sampler → Token
                                       │
                                  ┌────┴────┐
                                  │  Attn   │ ← KV Cache (paged)
                                  │  FFN    │
                                  └─────────┘

Each box is one or more WGSL compute shader dispatches orchestrated by Rust.
See docs/architecture.md for the full system design.

Key Design Decisions

Decision Detail
12 WGSL compute shaders Every Transformer op is a standalone shader (GEMM, MATVEC, RMSNorm, RoPE, SiLU, softmax, etc.)
f32/f16 compile-time switching WGSL string injection — no runtime branching
Paged KV Cache Lazy GPU page allocation with per-page sequence-offset writes
Single CommandEncoder All dispatches per token in one GPU submission — no mid-pipeline sync
CPU-side sampling Temperature → top-k → top-p → softmax → weighted sample; GPU argmax at temp=0
Row-sharding Large embedding/LM-head tensors auto-split to fit max_storage_buffer_binding_size
GQA support Grouped-Query Attention for Llama 3 models

Future Roadmap

  • Batched prompt prefill — process all prompt tokens in one dispatch
  • GPU timestamp queries — accurate kernel-level timing
  • 7B+ model support — optimize for Llama 3.1 8B and larger
  • INT8 GEMM activation ✅ — ~2× VRAM reduction via scripts/quantize_int8.py + --gemm-int8-block-size
  • Streaming output ✅ — tokens are emitted as generated
  • Multi-turn chat — KV cache reuse with chat template wrapper

See docs/plan.html for the full interactive roadmap.

Documentation

Document Purpose
Architecture System design, data flow, buffer layouts
CLI Reference All flags, sampling config, advanced modes
Roadmap Interactive implementation plan
Doc Maintenance Documentation rules for contributors
Guides wgpu & WGSL pitfalls and best practices
Contributing How to contribute

License

Dual-licensed under MIT or Apache-2.0 at your option.