Emory Lundberg

@emory

Context Engineer, Threat Model Architect, Photographer, Writer, Parent

sporage.substack.com

All cardsCollections

GitHub - SilentVoid13/Templater: A template plugin for obsidian

github.com

A list of useful resources about Templater:

@GitMurf quick demo How to setup and run your first Templater JS script: #187
@shabegom How To Use Templater JS Scripts: https://shbgm.ca/blog/obsidian/how-to-use-templater-js-scripts
@chhoumann Templates showcase: https://github.com/chhoumann/Templater_Templates
@zachatoo Templates showcase: https://zachyoung.dev/posts/templater-snippets
@lguenth Templates showcase: https://github.com/lguenth/obsidian-templates
@tallguyjenks video: https://youtu.be/2234DXKbNgM?t=1944
@ProductivityGuru videos: https://www.youtube.com/watch?v=cSawi0tYPMM

Italy | The Art Institute of Chicago

artic.edu

HCB

Gödel’s Therapy Room — Where Alignment Goes to Die | LLM Eval Harness | Leaderboard

gtr.dev

Model performance from Gödel

Gödel’s Therapy Room — Where Alignment Goes to Die | LLM Eval Harness | Leaderboard

gtr.dev

Rusty bibliography management.

Hayagriva is a tool that can help you or your apps deal with literature and other media. Its features include:

Data structures for literature collections

Reading and writing said collections from YAML files

Formatting literature into reference list entries and in-text citations as defined by popular style guides

Interopera

GitHub - typst/hayagriva: Rusty bibliography management.

yutu is a fully functional MCP server and CLI for YouTube to automate your YouTube workflows. It can manipulate almost all YouTube resources, like videos, playlists, channels, comments, captions, and more

GitHub - eat-pray-ai/yutu: A fully functional MCP server and CLI for YouTube

sweet

Dream Router 7 - Ubiquiti Store

store.ui.com

New DreamMachine with multigig and a 10Gb SFP+ WAN uplink

Don't know if you're still looking. AMD recently released support for rocm for the HX 370 and it let me get ollama going with the GPU. I did a small benchmark using gpt-oss:20b and got 16.5 tokens/sec for a simple prompt. Once I reallocated 32G VRAM to the GPU and started the ollama server with:

HSA_OVERRIDE_GFX_VERSION=11.0.2 OLLAMA_DEBUG=1 ollama... See more

Don't know if you're still looking. AMD recently released support for rocm for the HX 370 and it let me get ollama going with the GPU. I did a small benchmark using gpt-oss:20b and got 16.5 tokens/sec for a simple prompt. Once I reallocated 32G VRAM to the GPU and started the ollama server with:
HSA_OVERRIDE_GFX_VERSION=11.0.2 OLLAMA_DEBUG=1 ollama serve
ollama ps showed "100% GPU" and I got ... 20.5 tokens per second.
So, not a huge uplift but it was only using ~100% cpu rather than ~1100% as before.

Hello! There are very few recent, properly executed, and detailed benchmarks online for the AMD Ryzen AI 9 HX 370 iGPU or NPU when running LLM. They were either made back when Strix Point support was very weak, or they use the CPU, or they run small models. Owners of mini PCs on the HX 370, can you share your experience of which DeepSeek (70B, 32B,... See more

_Nitor reddit.com

You can expect snappy, "ChatGPT-vibes" performance on models up to 20B, with user benchmarks reporting 15 to 27 tokens/second using the Radeon 890M iGPU. For larger models like DeepSeek 70B there aren't specific benchmarks, but you will absolutely need 96GB of RAM since memory bandwidth is the primary bottleneck and the iGPU shares system memory. Currently, the NPU isn't a major factor for these LLMs due to immature software support, so performance relies entirely on the iGPU.

Upvote1DownvoteReply AwardShare

_Nitor

OP•1mo ago

I know that DeepSeek 70B requires a lot of RAM. The problem is that if it generates, for example, 1 token/second, then it's not usable at all. And it makes no sense to take a version with a lot of RAM for it. Is there any data available on the performance of DeepSeek 32B?

Upvote1DownvoteReply AwardShare

tmvr

•1mo ago

Only the sparse MoE models will generate decent speeds on this machine if you go large. gpt-oss 20B, gpt-oss 120B, Qwen3 30B A3B for example. A dense 70B/72B model at Q4 will produce less than 2 tok/s, a 32B at Q4 will produce about 3-4 tok/s.