gentile

OP could be doing a lot better with a MoE model (uses some RAM as well).

I'm doing Qwen 3.6 35BA3B-NVFP4 (a larger ~20GB model) in 8GB VRAM and 20 GB RAM (I also get more than 16k context, and more t/s).

Just create a custom filter in ublock origin:

www.linkedin.com##div:matches-attr(data-testid="mainFeed")

You could also add in

www.linkedin.com##aside:matches-attr(aria-label="Aside")

for good measure.

It's still typically for code, like python data stuff, bash scripts, general web search, small javascript stuff for my website. By saying not SWE, I mean I don't really see much benefit from "agentic" stuff, although I've tried. 50 t/s means 40~60 seconds for a typical thinking response. I used to run gemma 4 e4b-it-qat fully in GPU (~150 t/s), but the quality improvement moving to a much larger MoE model was 100% worth the switch. Especially because I had a lot of idle ram (from the before times :( ). There is some tradeoff in terms of context length, but I just use it as a chat. Also it's fairly trivial to setup as long as the GPU is supported.

Not a SWE, but I do Freetoken (backend) + unsloth (front) to run qwen 3.6 35B A3B NVFP4 (~20GB) in 8GB VRAM + ~20GB RAM. Getting around 50 t/s on a 3070. I don't think I can even share through this site.

You can download Osmand~ from F-droid which doesn't limit downloads. Or you could even download the map from another provider like bbbike or geofabrik, use osmandmapcreator (needs java) to convert the pbf to obf and then import it into the phone.

I've been making some custom maps for both Osmand and Comaps. The main big positive for Comaps I've seen is Android Auto being available freely (but you have osmand+ so moot point). But having used both, the tooling for osmand is far more mature. Creating an obf file (osmand file type) is a lot easier than the process for an mwm file (comaps file type). Far more routing or ui customization options in osmand and routing.xml files for even more advanced custom time penalties/weighting profiles.

The 100k figure is an overestimate by a few percent. The OpenStreetMap data for ALPRs is pretty good, but there is some duplication. I (recently) programmatically identified ~2.5k such instances. https://pickpj.github.io/Mapping/FIock/similar.html It has openstreetmap links attached for those who want to help fix the data.

Consider balling up some electrical tape underneath the Ram stick. This solved this very specific issue with my laptop that was flexing too much and crashing.

Download osm data, extract roads and surveillance, gpd overlay how=difference, remove/edit the different osmid's, write to pbf file, convert to obf file w/ osmandmapcreator, import into OsmAnd.

Now you have turn by turn navigation around ALPRs on your phone.

Edit: link https://github.com/pickpj/Big-B-Router - I tend to find ALPRs that are missing in the OSM data, so keep on updating OSM data.

To add, a bit complicated, but osm data can be edited with osmium (cpp) or pyosmium (python). Then the edited data can be put into osmandmapcreator to generate the file to use in Osmand. (I used this to route around ALPRs)

Yes, it just uses the shapefile from the tokyo open data site.

https://catalog.data.metro.tokyo.lg.jp/dataset/t000008d00000...

It's a quarto site so if you navigate to the repository of the github page you can see the python code used to generate the map/choropleth.

Some more context for Japan's land use (it does not differ prefecturally; some local regulations like kyoto and building colors)

The basics of Japan's Land use rules in english (only 8 pages; mostly tables/pictures and very straightforward): https://www.mlit.go.jp/common/001050453.pdf

- In the "Control of Building Use by Land Use Zones" you can see how even the most exclusive of zoning enables "Houses with other small scale function", Clinics, Schools or stores with very small footprints.

- Structures are restricted by the shape, shadow, and floor area.

- No mention of "single-family" housing.

A map (of Tokyo) overlaying the gradient of zoning from least to most permissive. You can see how the up-zoning follows the major roads: https://tokyochizu.github.io/zoning.html

I like:

Kaeru no tsura ni shonnbenn, カエルの面に小便, (Like) piss on a frogs face;

Basically the idiom "like water off a ducks back"

It's mentioned in the article, the US embassy advises against unauthorized employment. https://jp.usembassy.gov/services/visas-japan-u-s-citizens/

A license isn't needed for an electric bicycle. The law states that it is classified as an electric bicycle if it is only pedal assist and only up to 24kph (15mph). The issue arises when either there are no pedals, or it can be operated without using pedals, like these suitcases. Electric only operation moves the classification to a scooter, thus license needed.

https://archive.ph/Fw5lX

https://www.keishicho.metro.tokyo.lg.jp/kotsu/jikoboshi/elec...

I only have some light use cases, so I use a cheap laptop (<$250) with a ryzen APU 8gb soldered/shared ram. Then added a 16gb ram stick, booted off a usb bios from github, increased uma buffer to 8gb. I had stable diffusion working, it was slow, but I'm pretty sure its faster than cpu/ram (2 min for 512x512 20-25 step).

On youtube you can block the section with shorts in them with ublock (rclick > block element), this was good enough for my use. I don't know about other sites.

There is a spelling mistake in the japanese translation under language tokenization. In こんにちわ, わ should be は.