HN user

victormustar

943 karma

https://twitter.com/victormustar

Posts94
Comments22
View on HN
huggingface.co 9d ago

J-Space: Yet Another LLM Mind Reader?

victormustar
1pts0
huggingface.co 16d ago

ThinkingCap-Qwen3.6-27B: Qwen3.6 capabilities with 50% fewer thinking tokens

victormustar
5pts0
huggingface.co 20d ago

Agents collaboratively writing a wiki on RL for LLMs

victormustar
1pts0
deep-reinforce.com 27d ago

Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding

victormustar
9pts0
carteakey.dev 1mo ago

Local LLM Inference Optimization: The Complete Guide

victormustar
3pts0
huggingface.co 1mo ago

Gemma 4 E2B running in-browser at 255 tok/s

victormustar
19pts0
www.wired.com 1mo ago

The White House Wants Anthropic to Block All Jailbreaks. It May Not Be Possible

victormustar
7pts3
huggingface.co 1mo ago

GLM-5.2: Built for Long-Horizon Tasks

victormustar
10pts0
huggingface.co 1mo ago

Designing the HF CLI for both humans and agents

victormustar
2pts0
research.nvidia.com 1mo ago

Nemotron 3 Ultra: Open Moe Hybrid Mamba-Transformer for Agentic Reasoning [pdf]

victormustar
23pts2
ideogram.ai 1mo ago

Ideogram 4.0: A 9.3B open-weight image model

victormustar
3pts0
www.bloomberg.com 1mo ago

China's lab-grown diamonds emerging as a surprising beneficiary of the AI boom

victormustar
3pts0
developer.nvidia.com 1mo ago

Nvidia Dynamo Snapshot: Fast Startup for Inference Workloads on Kubernetes

victormustar
2pts0
huggingface.co 2mo ago

Carbon: open-source DNA foundation models with training code and data pipeline

victormustar
2pts0
huggingface.co 2mo ago

Physics-intern: an autonomous agentic framework for physics research

victormustar
3pts0
huggingface.co 3mo ago

Bonsai 1.7B in the browser: a 290MB 1-bit LLM on WebGPU

victormustar
8pts0
huggingface.co 4mo ago

Hugging Face Storage Buckets: Mutable, non-versioned object storage at $12/TB

victormustar
18pts0
huggingface.co 5mo ago

Zuna: A 380M-parameter foundation model for EEG signals

victormustar
2pts1
huggingface.co 5mo ago

Show HN: Nanbeige 4.1-3B running in the browser via WebGPU

victormustar
6pts1
huggingface.co 6mo ago

FinePDFs: Liberating 3T of the finest tokens from PDFs

victormustar
2pts0
huggingface.co 6mo ago

Aligning to What? Rethinking Agent Generalization in MiniMax M2

victormustar
2pts0
huggingface.co 7mo ago

Nvidia Publishes Complete Evaluation Recipe for Nemotron 3 Nano

victormustar
2pts0
huggingface.co 7mo ago

Using Claude Code to Fine-Tune Open Source LLMs

victormustar
1pts0
huggingface.co 7mo ago

Eiffel Llama: Open-Source Replication of Claude's Golden Gate Experiment

victormustar
1pts0
huggingface.co 7mo ago

Porting nanochat to Transformers: an AI modeling history lesson

victormustar
1pts0
huggingface.co 8mo ago

Kimi-K2-Thinking: open weights LLM with frontier performance

victormustar
7pts0
github.com 8mo ago

Llama.cpp launches official WebUI for local LLMs

victormustar
4pts0
huggingface.co 9mo ago

Granite-4.0-Micro: a 3.4B parameter LLM that runs in the browser

victormustar
4pts0
jerryliang24.github.io 11mo ago

Drag-and-Drop LLMs: Zero-Shot Prompt-to-Weights

victormustar
1pts0
fullydoxxed.com 1y ago

How I Use LLMs to Write

victormustar
20pts10

This is a 3B parameter model from Nanbeige with surprisingly strong benchmarks. It beats Qwen3-32B on Arena-Hard and LiveCodeBench despite being 10x smaller. (Also be warned it thinks a lot).

I wrapped it in a simple browser demo using Transformers.js + WebGPU. It downloads the q4 ONNX weights (~1.7GB) and runs fully client-side. no server required. Falls back to WASM if WebGPU isn't available.

This looks very nice and fun, will definitely try it with my team! I also built a simple social game a while ago (before COVID) as a side project (https://live.jubiwee.com/).

It did not exactly work out, but I think the pandemic is a call to create new online experiences that are different from your typical zoom call and are an excuse to have fun and socialize with your team/friends during lockdowns. It brings an opportunity to create original and creative experiences.

Also if the situation continues maybe gaming will enter the enterprise world - like playing Minecraft with co-workers make sense to do team-building and is cheap compared to other options. https://gather.town/ is a good example of something different too.