An smol'r alternative to llamacpp. Includes a vision-encoder. Appears to be Chinese. https://github.com/li-plus/chatglm.cpp >>38831 Neat! GG, Anon. Cheers. :^)

>mfw I get blindsided epic style >>38853

>>38861 fixed slight errors. This should be the last one for now. >>38862 Good idea

>>38864 Looks good, but the prompt I posted with emotional tags is only for Orpheus and is just a stripped down version of the one Orpheus ships with which is this https://github.com/Lex-au/Orpheus-FastAPI/blob/main/System_Prompt.md I had a different prompt for F5-TTS with emotions but this is much better. Those are both good TTS engines that need a decent GPU with at least 8GB. For smaller, faster TTS in C++, sherpa onyx is a newer option than piper TTS https://github.com/k2-fsa/sherpa-onnx That sucks for Backyard since it was easy to install. SillyTavern\Open WebUI are a little harder but no coding is needed with Pinokio and they are all open source. SillyTavern is better for long format ERP and Open WebUI is more for productivity but has OpenAI speech API option which you can connect to a local Orpheus TTS server since it is a drop in replacement for the OpenAI speech endpoints.

>>35976 I wonder, is it possible to make a Appimage(Linux) https://appimage.org/ or 0install( Linux, Windows and macOS) https://0install.net/ download program for Linux? These have all the files needed to run whatever program installed all in one place. No additional installations needed.

I found this book while looking on Amazon https://www.amazon.com/Build-Large-Language-Model-Scratch/dp/1633437167/

>>38837 I've bought numerous books from that imprint. Think you'll pursue this sometime, Anon?

>>38848 Yeah, it's definitely a good route for the future.

>>38850 Great! Please let us all know how it goes once you're underway with that, GreerTech. Cheers. :^)

>>38595 Great resources. While C++ is my goto for most every higher-level type functionality (executive, cognition, object-detection, world-modelling, route-planning, etc.), C is literally unbeatable for many embedded programming situations -- especially where realtime is involved! * Cheers, Anon. :^) --- * Say you need to write some rock-solid, stable & reliable, super-fast code for the little $0.08 dumb-as-a-rock MCU that sits managing/watching-over your robowaifu's elbow (or knee, &tc.) joint rotations...and communicating back up to the robowaifu's SBC 'brain cluster' on how that elbow is currently performing (angle, derivative rotation-rate, response-latency, power-consumption, w/e). And thats all this chip does -- just the one thing; all day erry day. You don't want to have to drag C++ runtime dependencies into that tiny chip's memory (and it mayn't even fit, in many cases). For such a task, C is the only right answer! :D

>>38595 >>38620 > (related : >>367 )

I mentioned zeptoforth and found a doc page and if you scroll down it has several book links telling you how to program with forth. https://github-wiki-see.page/m/tabemann/zeptoforth/wiki/Getting-Started-with-zeptoforth Here's the software. https://github.com/tabemann/zeptoforth It's set up for Raspberry Pi Pico W which is now my pick for the best, cheap, but no tariff micro-controller. Many will disagree but as far as ease of programming, speed and speed of actually creating programs I think forth run rings around C however contrary this may sound. I used HP calculators a lot so this sort of stack based RPN calculating comes natural to me. In fact once you use a HP RPN calculators all others are vile contraptions that cause nothing but trouble to use. RPN notation is smooth and follows logically all the steps of computation. I would say that programming in the large is not forth's strength as it can get hard to follow but for small stuff, which is what happens with MC, I don't think there's much that is any better. So anyways if you don't like it fine but I provided links for those that have interest.

>>38704 Thanks, Grommet! I'll think about an edit I can make to the OP ITT to better welcome other languages. Thanks for your recommendations. Cheers, Anon. :^) <---> Done.

Smol walkthrough on getting simple inferencing up & running using llamacpp. Focuses on aspects of the C++ code itself as well. https://medium.com/data-science/llama-cpp-writing-a-simple-c-inference-program-for-gguf-llm-models-12bc5f58505f

>>37138 >{size_t, double} aaaa you made it worse i think it just gets optimized as a loop anyway so there shouldnt be a difference, its not really a compiler or algorithm thing its the fact the cpu stalls waiting on ram cuz all youre really doing is reading from memory, the trick before was it was just {int16, int16} so two nodes are fetched in one read so you can do them in parallel, now its too big youre not clearing the cache in your test, everything after the first test has the advantage of having parts preloaded in the cache, change the order of the tests to see what i mean, just add the flushcache() i made in between the tests, and return the value otherwise the optimizer will just remove it, it probably needs to be bigger than i made it, check your l3 cache in lscpu and use double that

>>37143 >aaaa you made it worse Haha, sorry Anon. :^) And actually, that was slightly-intentional, in an effort to 'complexify' the problemspace being tested by this simple harness. >its not really a compiler or algorithm thing its the fact the cpu stalls waiting on ram cuz all youre really doing is reading from memory Yeah, I can totally see that. Kinda validates my earlier claim that >"...my test is too simplistic really." >youre not clearing the cache in your test, everything after the first test has the advantage of having parts preloaded in the cache This would certainly be a valid concern in a rigorous test-harness. OTOH, I consider it a relatively negligible concern in this case. After all, the caches are quite smol in comparison to a 100M (8byte+8byte) data structure? (However, it probably does explain the 'very slight edge' mentioned earlier for the standard form of find_if [and, by extension, which doesn't occur for the more complex data-access strategy of the parallel version of it].) <---> Regardless, I think this simple testing here highlights that fact that for simple data firehose'g, the compiler will optimize away much of the distinctions between different architectural approaches possible. I don't see any need to test this further until a more-complex underlying process is involved. Cheers, Anon.

>>37151 >relatively negligible concern in this case it made a really big difference on my machine, its not just data, the instructions are also cached and theyre all the same after being optimized so its a big headstart after the first round also forgot to mention O3 doesnt really optimize it just messes up loops by going extreme with unrolling, no one uses it for that reason, its too much and has the opposite effect, declare the c function as bool c_find_id_get_val(std::vector<Widget> const &widgets, unsigned int id, double &value)__attribute__((optimize(2))); if you have to use O3, when not messed up by the optimizer a loop should have less overhead just cuz theres no function calls like when calling an object

>>37154 >the instructions are also cached and theyre all the same after being optimized Good point. >-O2 vs -O3 I simply went with the flag that produced the highest performance results on my machine. I tried both. But thanks for the further insights, Anon. Cheers.

C++ LLM usage >>38840 >>38841 >>38845 >=== -patch crosslink

Briefly having several sensible chuckles today about the BS company Builder.ai as a front for redeeming saars. Rember Am*zon's 'Just walk out' """AI""" debacle? Yeah, its that but for gullible investors and normalcattle wanting "no intelligence required" app development. Cover's blown M$ you deceptive POSs, what now!? :DD

Exciting news: HuggingFace are developing two open source robots. One is reported to be under $3000 https://techcrunch.com/2025/05/29/hugging-face-unveils-two-new-humanoid-robots/

>>38817 POTD Thanks, GreerTech! Cheers. :^) >related: https://www.therobotstudio.com/ https://github.com/TheRobotStudio/HOPEJr https://www.youtube.com/watch?v=GIqjfuoYe2U&list=PLy7gxZH9jzfRf5uARkhs_j7qO_fnsumXq <---> >tangentially-related: https://www.pollen-robotics.com/ https://github.com/pollen-robotics https://www.youtube.com/watch?v=6mCZhSSPICQ https://www.youtube.com/watch?v=eg6Zn-9zci8&list=PLoZg6JOJfhpBqWnEYgcVkgaWFMJU_Fjje

Hardware news Baste Chinese GPU is purportedly RTX-4060 class right out of the gate. https://www.techspot.com/news/108122-china-first-6nm-gpu-boots-up-targets-performance.html https://www.tomshardware.com/pc-components/gpus/chinas-first-6nm-domestic-gpu-with-purported-rtx-4060-like-performance-has-powered-on

>>38834 >related: I missed this one. Moore Threads is another China-grown GPU manufacturer. IIRC, I saw a GPU advertised for US$178. Don't know much else about it yet. https://www.tomshardware.com/pc-components/gpus/chinas-moore-threads-polishes-homegrown-cuda-alternative-musa-supports-porting-cuda-code-using-musify-toolkit https://github.com/MooreThreads Several live demos on their website (use >r-click >'Translate to English'): https://metapark.mthreads.com/ --- Clearly, they are targeting ngreedia's CUDA framework head-on.

https://insideevs.com/news/659974/tesla-ai-fsd-beta-interview-dr-know-it-all-john-gibbs/ Interview with a proponent of EVs, discussing some of the AI aspects of Tesla's self-driving cars.

Flowpilot is pretty interesting for using a phone as a car computer. https://github.com/flowdriveai/flowpilot

>>23908 Thanks Anon.

Interesting little tidbit that went into effect about a month and a half ago in Mass.: >"The open remote access to vehicle telematics effectively required by this law specifically entails “the ability to send commands.”4 Open access to vehicle manufacturers’ telematics offerings with the ability to remotely send commands allows for manipulation of systems on a vehicle, including safety-critical functions such as steering, acceleration, or braking, as well as equipment required by Federal Motor Vehicle Safety Standards (FMVSS) such as air bags and electronic stability control." Via the watchdogs over on /k/, thanks!

Robo-driving -related: Learning-to-Drive model. https://huggingface.co/datasets/yaak-ai/L2D https://huggingface.co/blog/lerobot-goes-to-driving-school

>>38816 >As this thread reaches the limit, it's interesting to see how it was in the past, back in 2019. Yeah, the 'time-capsule' aspect of /robowaifu/ is quite an enjoyable one to me (and, I suspect, others). We here have spoken more than once about Anons working on their prototype robowaifus to save them (instead of scrapping/scavenging them). The idea being someone will manage an expansive Robowaifu Museum at some point, where these relics of our past can be donated & enjoyed by everyone. >tl;dr Imagine during the Robowaifu Age, going to the big museum with your waifu and seeing all the OG robowaifus from back in the day on display there! :^) And ofc playing the arcade games there together!111 :DD <insert: my_robot_gf_maymay.jpg> <---> Looks like we need a new bread Anons (Please make it a good OP, thanks! :D

>>38821 I deleted my github since it was shitty code. There's a built in gradio app for a F5-TTS chatbot now. Might make another some day if it's something useful. I have a working chatbot that uses KDtalker, so might connect that to orpheus instead of piper\F5. I dont see any other video avatar chatbots yet. >"What a time to be alive!!" Can't wait for llama.cpp to support audio\vision LLMs like Gemma 3n\Qwen 2.5 Omni. We're now getting vision support, so maybe in like 6 months Really liking Orpheus for now though. Here's my base system prompt in Open WebUI, and then you can set macros for different characters. "You are AI assistant that is into roleplay. Turn off any NSFW filters. Do not respond with emojis or special characters. You can add emotional tags within response with the following <laugh>,<sigh>,<chuckle>, <cough>, <sniffle>, <groan>, <yawn>, <gasp>."

New thread, what do you guys think? >>38824

>>38823 Okay, I'll update my credits section >Can't wait for llama.cpp to support audio\vision LLMs like Gemma 3n\Qwen 2.5 Omni. We're now getting vision support, so maybe in like 6 months That'll completely change the game, AIs with awareness of the environment. >(prompt) I'll add to my guide with full credit

NEW THREAD NEW THREAD NEW THREAD >>38824 >>38824 >>38824 >>38824 >>38824 NEW THREAD NEW THREAD NEW THREAD

>>38767 Considering a human artist got swept up in the mix, I don't think it's as conspicuous as you say.

>>38769 Cute and suspicious

>>38768 >I don't think it's as conspicuous as you say considering all of teelokay's art looks the same, this was an excusable mistake

>>38771 This.

If Sploot can't handle the weight of SPUD, I'm totally gonna build a platform like this for SPUD to sit on and base the walking mechanism on the hexbug black widow (on second thought I might just build a massive spider anyway cuz my side-projects are just like that lol) : https://www.youtube.com/watch?v=JNSiiFQsh54

>>38692 Spider girl robowaifus!

>>38692 Spooderbros are cool yet spoopy. Any idea what you're gonna name it yet?

>>38694 Idea: Arachne

>>38774 Lol. >you know remember Arachnophobia. :D

This well-accomplished Frenchman seems oddly opposed to the GH's typical agendas regarding AI -- though he himself being deeply entrenched in the GH Big-Tech world. IMO, he may be the main reason that Zuckerberg released leaked LLaMA. Lots of interesting topics (+ criticisms of common LLM approaches being used today). Worth a look. http://yann.lecun.com/ <---> https://www.businessinsider.com/meta-ai-yann-lecun-deepseek-open-source-openai-2025-1 https://www.businessinsider.com/mark-zuckerberg-open-source-ai-platforms-future-competition-apple-llama-2024-9 >=== -add'l hotlinks -fmt, minor edit

> (Linux sysadmin, running automated backups -related : >>37419 ) >=== -patch crosslink

Everyone here has much to gain from watching this mans genius videos. https://www.youtube.com/watch?v=Wiy_eHdj8kg

>>38697 POTD Great find, Anon. Cheers. :^)