pzo

31 ms·

361.
by pzo 2y ago
Maybe model is sensitive to quantization, by default ollama quantize it significantly.
362.
by pzo 2y ago
Depends what you mean small 4B? 7B? You can try qwen2.5 3B or 7B though 3B version is on no commercial license. Phi4-mini also should be good. Tested only on polish/english pairs should be good for spanish too. Smaller models like 1.5B
363.
by pzo 2y ago
I haven't played with more advanced feature or RAG, but at least AnythingLLM seems to support both local and remote vector databases (you can choose: lancedb, chroma, pinecone, zilliz cloud), embedding provider again both local (ollama
364.
by pzo 2y ago
open llm leaderboard [0] is probably good to compare open weights model on many different benchmarks - wish they put also some closed source one just to see what's relative ranking of best open weights to closed source one. They haven&
365.
by pzo 2y ago
would be good to see how gemma 3:4b compare to phi4-mini
366.
by pzo 2y ago
there is static hermes from Meta that do AoT compilation to native so I find it actually ambiguous. For a second I thought they did a compiler instead of transpile r.
367.
by pzo 2y ago
some other alternatives (a little more mature / feature rich): anythingllm https://github.com/Mintplex-Labs/anything-llm openwebui https://github.com/open-webui/open-webui lmstudio https:&#x
368.
by pzo 2y ago
Even that was handled badly. In my case when I went to authorized shop in central London. The guy did some test and said everything is ok and I didn't quality to change of battery (I tested using coconutBattery.app battery health was b
369.
by pzo 2y ago
I will have a look at this. Played with pipecat before and it's great, switched to sherpa-onnx though since I need something that compile to native and can run on edge devices. I'm not sure if turn detection can be really solved e
370.
by pzo 2y ago
I just wish they could provide additional options: 1) have ram soldered but provide also extra 1 slot for LPCAMM/LPDIMM ram. Soldered RAM would be used for GPU but external slower ram still could be used for non GPU related stuff (soft
371.
by pzo 2y ago
Surprisingly those open models might be savour for Apple and gift for Qualcomm too. They can finetune them to their liking and catch up to competition and also sell more of their devices in the future. Longterm even better models for Vision
372.
by pzo 2y ago
I'm wondering as well. Here in open llm leaderboard there is only preview. Better than deepseek-ai/DeepSeek-R1-Distill-Qwen-32B but surprisingly worse than deepseek-ai/DeepSeek-R1-Distill-Qwen-14B in Open LLM leaderboard over
373.
by pzo 2y ago
It's not about harder but about what error you can tolerate. Here if you have accuracy 99% for many applications it's enough. If you have 99% accuracy per trip of no crash during self driving then you gonna be dead within a year
374.
by pzo 2y ago
I love it, I always wished for something like that. Will try to later test with python. Wish there was also for JS/TS. As for rr debugger did it got by now any support for MacOS or Windows or Android? I'm also wondering how heavy
375.
by pzo 2y ago
I'm not sure if they would canibalise anything much. The sell each year around: 1) 230m iPhones 2) 50m iPads 3) 25m Macbooks Macbook is pretty much niche product to them comparing to Windows market share. Many would still wanna own Mac
376.
by pzo 2y ago
By the lawyer speak I mean for the same reason insurance put some misleading wording that other party don't understand so that you can interpret it differently. I think the intent should more important.
377.
by pzo 2y ago
I think the mess we have right now is because of many bad choices by US and allies since WW2: 1) They could reduce and rate limit lend lease to Soviet similar like they rate limited to Ukraine so buy themself more time until nuke is ready a
378.
by pzo 2y ago
Thats kind of lawyer speak, not much different when insurance company don't wanna cover you because of some tiny font paragraph or different interpretation in the contract. USA could before this war started (because they publicly on ca
379.
by pzo 2y ago
only sometimes on paper because they bought property decades ago that increased value but debt haven't been paid off yet. Even if debt paid off doesn't mean those people are rich - if they sell property they would have to still re
380.
by pzo 2y ago
The problem is that this would have be to incorporated worldwide wise. Otherwise will capital first run to different cheaper state or different (tax paradise) country. I'm also not sure if alone would solve the problem. IMHO we would
381.
by pzo 2y ago
Long term it might be hard to monetise those infrastructure considering their competition: 1) For coding (API) most probably will stick to Claude 3.5 / 3.7 - big market but still small comparing to all world wide problems 2) For non-co
382.
by pzo 2y ago
It's even hard to detect when humans stopped talking when talking to human while having high latency especially at the beginning of the call when you testing how big latency it is.
383.
by pzo 2y ago
Laptop and computers still is used as dominant for writing software including for mobile - you know software that could be infected. Still no collapse. Smartphones were not main device 10 years ago and people did bank transfers on laptop an
384.
by pzo 2y ago
As I said shady apps are/were also in App Store. Today they are not able to manually really review and detect all malicious apps. MacOS is still more open and the world didn't collapse. At minimum they could provide similar mechan
385.
by pzo 2y ago
Does nobody in you family own MacOS or Windows? Because there you can install malicious apps so your point is selective. Also there were many malicious apps approved by apple - this days they do a lot of checks automatically so nothing prev
386.
by pzo 2y ago
I wish they did also dedicated keyboard app like SwiftKey that has copilot integration
387.
by pzo 2y ago
Good point. In the old times if someone had Eclipse but installed plugins for different language than Java we wouldn't suddenly downgrade Eclipse that it is a text editor.
388.
by pzo 2y ago
Related to this I'm surprised how expensive still Tesla Powerwall is: ~$10k for 13.5kWh. Even Segway these days sell cheaper Portable Power Station Cube that is modular and stackable (upto 5kWh but IMHO much better design and useful) f
389.
by pzo 2y ago
This is still bad, When I'm developing I'm building many times during the day but uploading to appstore or TF maybe once a week. And thats as a freelancer if someone part of big team probably uses dedicated CI and special services
390.
by pzo 2y ago
I will give a little more pessimistic answer. If someone is right now studying CS then probably have expectation that can work with this profession for 30-40 years until retirement and this profession will still pay much more than average s

More ›