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

博客园_首页
J
Java Code Geeks
博客园 - 聂微东
量子位
C
Check Point Blog
T
The Blog of Author Tim Ferriss
T
Tailwind CSS Blog
G
Google Developers Blog
Google DeepMind News
Google DeepMind News
B
Blog
罗磊的独立博客
腾讯CDC
GbyAI
GbyAI
博客园 - 【当耐特】
A
About on SuperTechFans
M
MIT News - Artificial intelligence
U
Unit 42
D
Docker
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
Y
Y Combinator Blog
大猫的无限游戏
大猫的无限游戏
小众软件
小众软件
S
SegmentFault 最新的问题
有赞技术团队
有赞技术团队

Hacker News: Show HN

PurrrrrFocus: Pomodoro Timer App - App Store Workflow Engine — Multi-Step Orchestration for Bun RapidPhoto: Pro Photo Editor App - App Store GitHub - DheerG/swarms: Achieve extraordinary results with claude code across a variety of tasks SPICE simulation → oscilloscope → verification with Claude Code — Lucas Gerads Show HN: VCoding – A 5 MB native Windows IDE with no dynamic dependencies Show HN: LLMs don't hallucinate because they're bad at math, it's the format GitHub - Agent-FM/agentfm-core: AgentFM is a peer-to-peer network that turns everyday computers into a decentralized AI supercomputer. AgentFM lets you run massive AI workloads directly across a global mesh of idle CPUs and GPUs. Show HN: Tracking Top US Science Olympiad Alumni over Last 25 Years GitHub - Potarix/agent-hub: One place to talk to all your agents Show HN: Runtime security for AI agents(injection,tool abuse, data exfiltration) GitHub - dubeyKartikay/lazyspotify: Terminal Spotify client for macOS and Linux GitHub - the-banana-tool/king-louie: Easy to use GUI Personal AI Assistant. Win/Linux/Mac. Show HN I made my vacation rental bookable by AI agents–no Airbnb, 0% commission GitHub - basteez/jsf-autoreload: maven plugin to enable hot reload on jsf projects uvm32/hosts/host-gdbstub at main · ringtailsoftware/uvm32 GitHub - labsai/EDDI: Config-driven engine that turns JSON into production-grade AI agents. Multi-agent orchestration, 12+ LLM providers, MCP/A2A protocols, RAG, persistent memory, and enterprise compliance (EU AI Act, GDPR, HIPAA). Built on Quarkus. GitHub - glitchnsec/fortyone-oss: AI Executive Assistant Platform Quickstart | Alien GitHub - muxshed/shed: One stream in, or many. Every destination, simultaneously. No cloud middleman, no per-channel fees, no limits. GitHub - ocrbase-hq/ocrbase: 📄 PDF/IMG ->.MD/JSON Document OCR API for PaddleOCR and GLMOCR. Self-hostable. GitHub - impactjo/home-memory: MCP server that lets your AI assistant remember everything about your home. GitHub - Sets88/dbcls: DbCls is a powerful terminal database client that supports various databases GitHub - neptun2000/heor-agent-mcp GitHub - SeanFDZ/macmind: Single-layer transformer in HyperTalk for the classic Macintosh RollQuation: Math Puzzles - Apps on Google Play GitHub - dropbox/witchcraft Show HN: Agent-cache – Multi-tier LLM/tool/session caching for Valkey and Redis GitHub - opentalon/opentalon: OpenTalon is an open-source platform built from the ground up in Go as a robust alternative to OpenClaw LinkedIn™ 职位抓取工具 - Chrome 应用商店
Release v2.0.0 — The Own Everything Release · Zyora-Dev/zse
zyoraclub · 2026-06-02 · via Hacker News: Show HN

ZSE v2.0.0 — The "Own Everything" Release

A complete rewrite. Zero third-party dependencies. No PyTorch, no Triton, no transformers, no bitsandbytes. Pure-Python kernel compiler emits CUDA C, HIP C, and Metal Shading Language directly.

Install size: ~3 GB → ~5 MB.

Headline numbers (Qwen2.5-14B INT4 vs vLLM AWQ INT4 on A100-80GB)

Metric ZSE vLLM Δ
Cold start 6.29s 127.02s 20.2×
VRAM used 12.28 GB 71.45 GB 5.82× less
Single-seq tok/s 37.0 26.5 1.40×

Validated on 6 platforms

GPU Cold start vs vLLM AWQ INT4 cold
NVIDIA T4 (sm_75) 7.25s 30.2× faster
NVIDIA L4 (sm_89) 5.58s 26.0× faster
NVIDIA A10G (sm_86) 6.01s 32.1× faster
NVIDIA A100-80GB 6.29s 20.2× faster
AMD MI300X 3.14s 13.6× faster (vs vLLM-ROCm FP16)
Apple M1 E2E vector_add validated, full inference pending

Install

pip install zse-engine
zse serve model.zse --port 8000

Or run the kernel compiler standalone:

What's in this release

  • ZSE Kernel Compiler@zse.kernel Python DSL → CUDA / HIP / Metal. Warp primitives, vectorized memory, block reductions, tiling, fusion, WMMA, CDNA3 MFMA matrix cores, auto-tuning.
  • .zse model format v2 — pre-quantized INT4/INT8/FP16, mmap-friendly, C-accelerated quantization (~600× faster). Adapters for Llama / Mistral / Qwen2 / Gemma2 / Phi3.
  • Own PagedAttention — adaptive block sizing, token-level eviction, FNV-1a dedup, COW forking.
  • ZStreamer — continuous batching, disaggregated prefill/decode, chunked prefill, speculative decoding (n-gram + self-draft).
  • Orchestrator — unified VRAM allocator, 29 GPU kernels on MI300X, CUDA Graphs + HIP Graphs, LoRA hot-swap.
  • Server — OpenAI-compatible API, API key auth, rate limiting, SQLite store, built-in RAG (/v1/rag/*), web dashboard.
  • RAG — BM25 + TF-IDF + dense embeddings (via the loaded LLM, zero extra deps) + Reciprocal Rank Fusion + LLM cross-encoder rerank.
  • Tensor Parallelism — pure-ctypes NCCL/RCCL wrapper, multi-process workers.

Breaking changes

  • Package rename: zllm-zsezse-engine on PyPI
  • Module rename: zsezse_engine
  • .zse format v2 is incompatible with 1.x — re-convert with zse convert
  • bnb / bitsandbytes backend removed
  • PyTorch / Triton / transformers dependencies removed

Full migration guide and detailed change log: CHANGELOG.md

Acknowledgments

AMD MI300X validation, 32B-parameter benchmarks, and our ROCm wave-64 kernel development were made possible by DigitalOcean's Open Source Sponsorship Program.


447 tests passing. Zero dependencies. Three GPU backends. One package.