frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: NeuralForge – Fine-Tune LLMs on Your Mac Using Apple Neural Engine

https://github.com/Khaeldur/NeuralForge
1•khaeldur•1h ago

Comments

khaeldur•1h ago
Hi HN, I built NeuralForge because I wanted to fine-tune small LLMs on my MacBook without renting cloud GPUs or setting up CUDA.

It uses Apple's Neural Engine directly (not Metal, not CPU) to hit ~1.2 TFLOPS on a consumer Mac. The app wraps a C/Obj-C training engine in a SwiftUI GUI with live loss curves, LoRA support, and one-click export to GGUF/CoreML.

What actually works today: - Real training on ANE (110M parameter llama2.c models) - LoRA fine-tuning (rank 4-64 on attention weights) - Cosine LR schedule with warmup - Checkpoint save/resume (survives crashes) - Text generation at 66ms/token with top-p sampling - Export to GGUF, llama2c, CoreML formats - 648 automated tests (unit + integration on real hardware)

Current limitations: - Only llama2.c format models (110M tested, larger planned) - macOS 14+ on Apple Silicon only - Some UI features are still stubs (being honest)

Setup: git clone + bash setup.sh (one command)

The hardest part was ANE itself. There's basically zero documentation on using it for training — it's designed for inference only. I had to reverse-engineer the MIL compiler, figure out the 119-kernel compilation limit per process, and build an exec() restart mechanism that transparently re-launches the training process to get fresh kernel budget.

MIT licensed: https://github.com/Khaeldur/NeuralForge

Happy to answer questions about ANE internals.ye

Apple Releases New Versions of iOS for Older iPhones

https://www.macrumors.com/2026/03/11/apple-ios-16-7-15-release/
1•mgh2•44s ago•0 comments

Add per-repo output language preference

1•nishiohiroshi•59s ago•0 comments

Feedback on a local-first MCP memory system for AI assistants?

https://github.com/ptobey/local-memory-mcp
1•ptobey•1m ago•1 comments

Does MAGA Want American Science to Win?

https://www.thenewatlantis.com/publications/does-maga-actually-want-american-science-to-win
1•SCEtoAux•1m ago•0 comments

Compare Claude 4.6 Opus and GPT-5.2 to Boost E-E-A-T Content Quality

https://searchfit.ai/blog/claude-4-6-opus-vs-gpt-5-2-for-eeat-content-accuracy-citations-hallucin...
1•kasaei•3m ago•0 comments

OpenClaw Meets Microsoft Agentic Identity [video]

https://www.youtube.com/watch?v=7uD2vyfBUUs
1•umangsehgal93•5m ago•0 comments

Live Stream Oscar Wilde Importance of Being Earnest

https://www.youtube.com/watch?v=obX-HGs-PS8
1•frm88•6m ago•1 comments

Lobsters Interview with Ngoldbaum

https://alexalejandre.com/programming/interview-with-ngoldbaum/
1•birdculture•9m ago•1 comments

Show HN: Homecastr - AI home price forecasts on a map

https://www.homecastr.com/app
1•dhardestylewis•9m ago•0 comments

Bokuchi: A lightweight, cross-platform Markdown editor built with Tauri

https://bokuchi.com/
1•nogajun•10m ago•0 comments

Trump class 'battleship' revives Navy's railgun project

https://www.twz.com/sea/navy-is-firing-its-railgun-again-after-abandoning-it-for-years
1•samizdis•10m ago•1 comments

Proof SDK: Editor, collab server, provenance model, and agent HTTP bridge

https://github.com/everyinc/proof-sdk
1•azhenley•13m ago•0 comments

What it will mean for the economy if the Strait of Hormuz stays closed

https://www.axios.com/2026/03/12/oil-prices-iran-strait-of-hormuz
1•toomuchtodo•13m ago•1 comments

How are people debugging multi-agent AI workflows in production?

https://www.agentsentinelai.com/
1•skhatter•14m ago•2 comments

Gotta have a permit to kayak and paddleboard In Oregon

https://twitter.com/JayinWashington/status/2032161546215047623
1•bilsbie•18m ago•0 comments

Show HN: Sergamon – I defined 3,700 font glyphs as plain-text pixel grids

https://github.com/sgmonda/sergamon
1•sgmonda•19m ago•1 comments

Show HN: AEF – Agents State Machine

https://github.com/mikemasam/aef-spec
1•mikemasam•20m ago•0 comments

Sidekiq in the Terminal

https://www.mikeperham.com/2026/03/10/sidekiq-in-the-terminal/
1•butterlesstoast•24m ago•0 comments

Crypto whale loses nearly $50M swapping USDT for AAVE

https://www.theblock.co/post/393466/crypto-whale-loses-nearly-50-million-swapping-usdt-for-aave
1•gdeglin•24m ago•0 comments

Data Science Weekly – Issue 642

https://datascienceweekly.substack.com/p/data-science-weekly-issue-642
1•sebg•26m ago•0 comments

Catching silent data loss using BigQuery's native ML and Write API

https://robertsahlin.substack.com/p/your-pipeline-succeeded-your-data
1•boxer_shorts•26m ago•1 comments

Portify: Generate a developer portfolio from your GitHub

https://www.portify.ca/
1•lucasadilla•26m ago•2 comments

A semantic history: How the term 'vibe coding' went from a tweet to prod

https://www.coderabbit.ai/blog/a-semantic-history-how-the-term-vibe-coding-went-from-a-tweet-to-prod
1•dmkravets•27m ago•0 comments

Crypto investor turns $50M into $36,000 in one botched move

https://www.coindesk.com/markets/2026/03/12/crypto-investor-turns-usd50-million-into-usd36-000-in...
1•scrlk•30m ago•2 comments

US carrying out rescue effort after military aircraft crash in Iraq

https://www.reuters.com/world/middle-east/us-carrying-out-rescue-effort-after-losing-aircraft-ira...
2•tartoran•30m ago•0 comments

Steve Yegge Wants You to Stop Looking at Your Code

https://www.oreilly.com/radar/steve-yegge-wants-you-to-stop-looking-at-your-code/
1•metadat•31m ago•0 comments

Anthropic and OpenAI just exposed SAST's structural blind spot with free tools

https://venturebeat.com/security/anthropic-openai-sast-reasoning-scanners-security-directors-guide
1•mooreds•31m ago•0 comments

US fuel tanker aircraft crashes in Iraq – what we know and don't know

https://www.bbc.com/news/live/c4gqjyk0vx3t
3•tartoran•31m ago•0 comments

Before you let AI agents loose, you'd better know what they're capable of

https://thenewstack.io/risk-mitigation-agentic-ai/
1•chhum•35m ago•0 comments

To use your brain, first accept the Terms and Conditions

https://ctlj.colorado.edu/?p=1460
2•hhs•36m ago•0 comments