tgtweak

15 ms·

151.
by tgtweak 10mo ago
The manageability of having everything on one host is kind of nice at that scale, but yeah you can stack free tiers on various providers for less.
152.
by tgtweak 10mo ago
Just a note - you can very much limit cpu usage on the docker containers by setting --cpus="0.5" (or cpus:0.5 in docker compose) if you expect it to be a very lightweight container, this isolation can help prevent one roudy contai
153.
by tgtweak 10mo ago
I trust companies that immediately and regularly update their status/issues page and follow up any outages with proper and comprehensive post-mortems. Sadly this is becoming the exception these days and not the norm.
154.
by tgtweak 10mo ago
There really should be an http header dedicated to "outage status" with a link to the service outage details page... clients (for example, in this case, your code IDE) could intercept this and notify users. 503 is cool and yes, th
155.
by tgtweak 10mo ago
If the models are designed around it, and not resorting to compression to get to higher input token lengths, they don't 'fall off' as they get near the context window limit. When working with large codebases, exhausting or c
156.
by tgtweak 10mo ago
So what are these tiny portable ones? I always assumed they were digitally augmented or virtual even - is there a minimum size for it to be a "real" telescope?
157.
by tgtweak 10mo ago
have been on 1M context window with claude since 4.0 - it gets pretty expensive when you run 1M context on a long running project (mostly using it in cline for coding). I think they've realized more context length = more $ when dealin
158.
by tgtweak 10mo ago
Cool - now make it analyze all of those and come up with the 10 commandments of commenting factually and insightfully on HN posts...
159.
by tgtweak 10mo ago
Have you looked at getting an openai api key for gpt5? you have to do selfie-ID verification...
160.
by tgtweak 10mo ago
You can probably follow the vLLM instructions for omni here, then use the included voice demo html to interface with it: https://github.com/QwenLM/Qwen3-Omni#vllm-usage https://github.com/QwenLM/Qw
161.
by tgtweak 10mo ago
Depends where you are plugging them in - but yes they are older gen - despite this, 8xV100 will outperform most of what you can buy for that price simply by way of memory and nvlink bandwidth. If you want to practically run a local model t
162.
by tgtweak 10mo ago
PSA: 10X savings when you have to prompt it 10 times to get the correct solution is not actually faster.
163.
by tgtweak 10mo ago
dual 3090's (24GB each) on 8x+8x pcie has been a really reliable setup for me (with nvlink bridge... even though it's relatively low bandwidth compared to tesla nvlink, it's better than going over pcie!) 48GB of vram and lots
164.
by tgtweak 11mo ago
Came with receipts lol - hopefully they can repro and fix this but the fact it as omitted for 8 months kind of hints at how little people are using it. Yeah you can really see the resize comparing the before and after. https://jp
165.
by tgtweak 11mo ago
It's actually kind of concerning how these just look normal at first glance
166.
by tgtweak 11mo ago
There is one parity bit per 8 data bits, which is decently resilient - plus the recovery is pretty simple on the occasional bit flip. Combine that with the fact you can reference other sources to make up for missing/corrupted files -
167.
by tgtweak 11mo ago
Very interesting storage format too - Those tapes actually held quite a bit of data (comparatively) - around 45MB (Although this one is shorter ~1000ft and probably carries about 10-15MB which is close to V4's source code, binary and d
168.
by tgtweak 11mo ago
The larger systems (solar + AC Inverter + home-scale battery): PowerHub Core (2 kW AC inverter 230VAC, 2.5kWh lfp battery storage, 1800-2700W solar panels - ₦48,000 per week, 80 weekly payments, ₦240,000 down payment ₦4,080,000 ($2,836.94 U
169.
by tgtweak 11mo ago
Overall, democratization of utility and finance is great, but this article is so inconsistent with the pricing and statements it really reads like a sycophantic llm wrote the entire thing. Here is the actual numbers from Sun King on their e
170.
by tgtweak 1y ago
300CPM above background is considered very low - likely why they classified this as non-emergency - the only reason it was reported was per NRC cfr that states any time there is transportation of a radioactively contained person offsite, it
171.
by tgtweak 1y ago
The amount of public WiFi's (including in-flight ones) I've bypassed by running a vpn server on udp port 53 is honestly insane. Sadly, this is becoming less commonplace many captive portals don't allow any egress at all asid
172.
by tgtweak 1y ago
Skills run remotely in the llm environment, not locally on your system running claude - worth noting.
173.
by tgtweak 1y ago
At term (and not even far term) - LLMs will be able to churn up their own "skills" using their sandbox code environments - and possibly recycle them through context on a per-user basis. While I like the flexibility of deploying yo
174.
by tgtweak 1y ago
Even the USB key .mp3 homebrew?
175.
by tgtweak 1y ago
Feels like it could be an extension to the compression lib (and would retain newlines as such) vs requiring external document tailoring. Also feels like a very specific use case but this optimization might have larger applications outside
176.
by tgtweak 1y ago
Missed opportunity to put a 503:Vape Unavailable http error.
177.
by tgtweak 1y ago
At 1 rack scale you're saving ~20-30k/mo in cloud fees - you can hire an excellent sysadmin in the 12-15k/mo range and they can do a lot more than just go to the datacenter as needed.
178.
by tgtweak 1y ago
We had single-datacenter resiliency (meaning n+1 on power, cooling, network + isp, servers) and it was fine. You still need offsite DRS strategy here - this is one of the things having that hybrid cloud is great for: you can replicate your
179.
by tgtweak 1y ago
Not always - you can lease your servers from the vendor as well, in which case you're renting the rack space, power and cooling from the datacenter and you're renting the servers from the vendor - most of the leases are designed s
180.
by tgtweak 1y ago
I've been doing hybrid colo+public cloud for over a decade and it's always been the most cost effective route at a certain scale. That specific break even point is lowering over time with the density and cost effectiveness of hard

More ›