10 replies
16h4m
IMO eventually users should be able to advertise what embedding models they have so we don't redundantly redownload.
8 replies
18h58m
Is this downloading a ~5gb model to my machine and storing it locally for subsequent use?
6 replies
18h11m
Yes, it only starts the download after you send the first message so visiting the site won’t use up any space.
Approx sizes are listed in the GitHub README.
Models are stored in indexeddb and will be managed by the browser. Might get evicted.
2 replies
18h3m
I see you have Phi1.5-q4f16_1-1k - any chance you could add Phi-3?
0 replies
17h19m
Would love to. It uses MLC AIs webllm so just need to convert it to that format.
0 replies
15h5m
Phi3 is already available in browser here: https://huggingface.co/spaces/FL33TW00D-HF/ratchet-phi
Disclaimer: I am the author.
2 replies
17h57m
I thought browser tabs only had access to ~400mb
How do you have access to 5gb?
1 replies
17h9m
A lot of the more modern options allow for many gigabytes for typical user setups https://developer.mozilla.org/en-US/docs/Web/API/Storage_API....
0 replies
16h59m
Very interesting! Thank you so much
I was always under the impression that the max blob size was 400mb and so you couldn't store files any bigger than that. Google gives so many different answers to these questions.
Do you know any other resources I can go more in depth on browser storage limits?
0 replies
18h33m
Models sizes are listed here https://github.com/abi/secret-llama but yeah, > 4GB for the Llama 3 model.
7 replies
17h51m
It's truly amazing how quickly my browser loads 0.6GB of data. I remember when downloading a 1MB file involved phoning up a sysop in advance and leaving the modem on all night. We've come so far.
4 replies
14h55m
97MB for the Worms 3 demo felt like an eternity.
So what games are in this LLM? Can it do solitaire yet?
1 replies
14h12m
It generates things that you get to look up citations for. It doesn't care if its output converges, it does what it wants differently every time.
0 replies
11h46m
It generates things that you get to look up citations for.
Why would you use it for that? Use a search engine.
LLMs are substitute for talking to people. Use them for things you would ask someone else about, and then not follow up with searching for references.
0 replies
3h45m
GPT-3.5 is pretty good at fabricating text adventures, I haven't tried any of the smaller models with that yet.
0 replies
11h59m
It can probably role-play.
0 replies
2h13m
You could more or less fit the full model on a single CD (or a DVD for the larger model sizes) but of course forget about trying to do inference for it on period hardware, it would be unusably slow.
0 replies
7h7m
When I think about numbers like that it just seems (to me, and wrongly) like general progress that's not so crazy - the thought that really makes the speed of progress stand out to me is remembering when loading a single image - photo sized but not crazily high resolution - over dial-up was slow enough that you'd gradually see the image loading from top to bottom, and could see it gradually getting taller as more lines of pixels were downloaded and shown below the already loaded part. Contrasting that memory against the ability to now watch videos with much higher resolution per frame than those images were 30 years ago is what really makes me go "wow".
For anyone not old enough to remember, here's an example on YouTube (and a faster loading time than I remember often being the case!): https://youtube.com/watch?v=ra0EG9lbP7Y
7 replies
12h26m
This is very cool, it's something I wish existed since Llama came out, having to install Ollama + Cuda to get locally working LLM didn't felt right to me when there's all what's needed in the browser. Llamafile solves the first half of the problem, but you still need to install Cuda/ROCm for it to work with GPU acceleration. WebGPU is the way to go if we want to put AI on consumer hardware and break the oligopoly, I just wished it became more broadly available (on Linux, no browser supports it yet)
2 replies
3h57m
having to install Ollama + Cuda to get locally working LLM didn't felt right to me when there's all what's needed in the browser
Was there something specifically about the install that didn't feel right? I ask because ollama is just a thin go wrapper around llama.cpp (its actually starting a modified version of the llama.cpp server in the background, not even going through the go ffi, likely for perf reasons). In that that sense, you could just install the CUDA toolkit via your package manager and calling `make LLAMA_CUDA=1; ./server` from the llama.cpp repo root to get effectively the same thing in two simple steps with no extra overhead.
1 replies
1h20m
I'm never gonna have my non-tech friend do any of this when they can just go to chat.openai.com and call it a day.
Most people value convenience at the expense of almost everything else when it comes to technology.
0 replies
56m
I'm never gonna have my non-tech friend do any of this
Who was making that assertion? I certainly wasn't.
In the same way I am never going to tell my non-engineer friends to build their own todo app instead of just using something like Todoist. But if they told me they cared about data privacy/security, I'd walk them through the steps if they cared to hear them.
2 replies
6h47m
Tested on Ubuntu 22.04 with Chrome, sure enough, "Could not load the model because Error: Cannot find adapter that matches the request".
It really is too bad WebGPU isn't supported on Linux, I mean, that's a no-brainer right there.
0 replies
10m
I get the same thing on Chrome and my last generation Intel iMac.
0 replies
1h2m
Works for me.
WebGPU support is behind a couple flags on Linux: https://github.com/gpuweb/gpuweb/wiki/Implementation-Status
0 replies
5h33m
I've managed to avoid ollama and just toyed with lmstudio. It's non-free software, but extremely easy to get into, uses llama.cpp under the hood, cross-platform, yada yada. There's https://jan.ai/docs as well, is AGPL3, and promises inference as well as training - doubtless many other similar offerings.
I'm wary of any 'web' prefix on what could / should otherwise be desktop applications, mostly due to doubts about browser security.
7 replies
14h48m
Phi 1.5 isn't quite there is it? Caught it having some inner dialog.
What does that mean? Instruct: I'm sorry, but I'm not sure what you mean by "wutt". Output: Oh, I'm sorry. I meant "what does that mean?" Instruct: No problem, "wutt" means "that doesn't make any sense". Output: Ah, I see. Thank you for clarifying! Instruct: You're welcome! If you have any more questions, feel free to ask. Instruct: I'm glad I could help! Instruct: I hope you continue to enjoy learning about language arts. Instruct: You're doing a great job! Keep up the good work! Instruct: Have a fantastic day! Instruct: Goodbye! Instruct: Goodbye!
4 replies
9h13m
That boy needs therapy.
3 replies
6h34m
Purely psychosomatic
2 replies
3h50m
But surely, expulsion is not the answer!
1 replies
1h2m
Where in the heck did you all discover this song? It's never been on the radio I assume. I only found it because it cropped up in yahoo messenger/launchcast at some point - never seen it anywhere else.
0 replies
30m
It was big on YTMND back in the day
0 replies
10h55m
I've seen this with Phi 1.5B a lot. Try TinyLlama for a 1.1B model or Gemma for 2B. The latter is not available on OP's website yet but maybe in the future.
0 replies
2h27m
That seems almost like the instruction template is wrong.
5 replies
14h24m
On Firefox Nightly on my Steam Deck it "cannot find WebGPU in the environment".
3 replies
14h3m
Last I checked ff explicitly does not support webgpu, webhid, webusb, etc.
Apparently nightly is supposed to support it: https://developer.mozilla.org/en-US/docs/Mozilla/Firefox/Exp...
2 replies
10h30m
So here's a howler to the new Mozilla CEO and FF teams who're looking for ways to save their org:
- release WebGPU support everywhere, also embed llama.cpp or something similar for non GPU users
- add UI for easy model downloading and sharing among sites
- write the LLM browser API that enables easy access and sets the standard
- add security: "this website wants to use local LLM. Allow?"
0 replies
6h39m
There’s also the little issue of firefox not supporting HDR videos which with more and more OLED/miniLED monitors out there is a major drawback. I love FF and i daily drive it, but there are some glaring gaps in the feature set between chromium and ff.
0 replies
7h37m
Hmm but what about another mobile phone OS instead? Or a vpn service? Surely people don't care about browser features.
0 replies
10h37m
I had the same issue on my iPhone! You can (temporarily) enable WebGPU by going to Settings > Safari > Advanced > Experimental features (I don't know what it's called in English, but it's the bottom one).
4 replies
4h46m
How do people use something like this as coach or therapist? This is genuine question.
Side note, impressive project. Future of AI is offline mostly with few APIs in the cloud maybe.
1 replies
1h31m
I tried using Claude and ChatGPT like this: I would just write a free form journal entry. The feedback it gave was typically very useful and made journaling more rewarding.
0 replies
47m
Given user data is folded back into the models, there is a snowball's chance in hell that I would input stuff I'd talk to a therapist about.
When are people going to realize that their interactions with AIs are likely being analyzed/characterized, and that at some point, that analysis will be monetized?
0 replies
4h34m
it's great at offering alternative perspectives
0 replies
3h11m
Genuine answer: you say “Be a coach/therapist” followed by whatever you’d say to a coach/therapist.
4 replies
7h5m
Yasssssss! Thank you.
This is the future. I am predicting Apple will make progress on groq like chipsets built in to their newer devices for hyper fast inference.
LLMs leave a lot to be desired but since they are trained on all publicly available human knowledge they know something no about everything.
My life has been better since I’ve been able to ask all sorts of adhoc questions about “is this healthy? Why healthy?” And it gives me pointers where to look into.
1 replies
5h54m
They are not “trained on all publicly available human knowledge”. Go look at the training data sets used. Most human knowledge that has been digitized is not publicly available (e.g., Google Books). These models are not able to get to data sets behind paywalls (e.g., scientific journals).
It will be a huge step forward for humanity when we can run algorithms across all human knowledge. We are far from that.
0 replies
2h19m
There is a rumor that OpenAI might've used libgen in their training data.
0 replies
4h1m
I actually think Apple has been putting neural engines in everything and might be training something like Llama3 for a very long time. Their conversational Siri is probably being neglected on purpose to replace it . They have released papers on faster inference and released their own models. I think their new Siri will largely use on device inference but with a very different LLM.
Even llama.cpp is performant already on macOS.
0 replies
3h17m
Groq is not general purpose enough, you'd be stuck with a specific model on your chip.
4 replies
17h29m
This is awesome. I have been using ChatGPT4 for almost a year and haven't really experimented with locally running LLMs because I assumed that the processing time would take too long per token. This demo has shown me that my RTX 2080 running Llama 3 can compete with ChatGPT4 for a lot of my prompts.
This has sparked a curiosity in me to play with more LLms locally, thank you!
1 replies
14h48m
My pixel 6 was able to run tinyllama and answer questions with alarming accuracy. I'm honestly blown away.
0 replies
12h14m
This is amazing. Thanks both for sharing your stories. Made my day.
0 replies
7h36m
0 replies
9h7m
Uh oh, I had that same moment a bit over a year ago with MLC's old WebLLM. Take a deep breath before you jump into this rabbit hole because once you're in there's no escape :)
New models just keep rolling in day after day on r/locallama, tunes for this or that, new prompt formats, new quantization types, people doing all kinds of tests and analyses, new arxiv papers on some breakthrough and llama.cpp implementing it 3 days later. Every few weeks a new base model drops from somebody. So many things to try that nobody has tried before. It's genuinely like crack.
4 replies
1h29m
Amazing work, feels like a step forward for LLM usability.
Would be interesting if there was a web browser that managed the download/install of models so you could go to a site like this, or any other LLM site/app and it detects whether or not you have models, similar to detecting if you have a webcam or mic for a video call. The user can click "Allow" to allow use of GPU and allow running of models in the background.
2 replies
49m
Mozilla won't even allow WebSerial to be implemented because it was deemed "too dangerous" - with all sorts of absurd whinging about the devastation that could be unleashed by unsuspecting users allowing a malicious site to access USB serial devices.
When someone pointed out that Chrome has had this functionality for years and the world has not imploded...and has enabled many open source projects and web-based microcontroller IDEs to provide enormous user convenience...the response was a condescending sneer along the lines of "well we actually care about user privacy."
(If Chrome is such a user privacy dumpsterfire, why not implement WebSerial so that people don't have to run Chrome in order to communicate with and program microcontrollers?)
Given they claimed that people's pacemakers and blood glucose monitors would be tampered with if WebSerial were implemented, I'd be shocked if they allowed such low level access to a GPU...
0 replies
5m
(If Chrome is such a user privacy dumpsterfire, why not implement WebSerial so that people don't have to run Chrome in order to communicate with and program microcontrollers?)
This doesn't seem like a logical comparison. Is there no other way to program microcontrollers outside of Chrome?
0 replies
20m
already allow by default in nightly builds
0 replies
1h22m
They should just be ubiquitous OS daemons at this point. They’re clearly very valuable
3 replies
6h5m
Could not load the model because Error: Cannot find WebGPU in the environment
0 replies
4h49m
For the foreseeable future, WebGPU is "Works best on Chrome (TM)".
0 replies
5h34m
See: https://github.com/gpuweb/gpuweb/wiki/Implementation-Status#... (I got there from Chromium's console).
On Linux, I had to go to chrome://flags/#skia-graphite and chrome://flags/#enable-vulkan and chrome://flags/#enable-unsafe-webgpu
I think only one of the first is actually required, but I enabled both. That allowed me to make use of TinyLlama with my AMD GPU (R9 Fury, OSS drivers), but I think I'd need Chromium Canary to enable "shader-f16" and use the other models, as I was not able to make it work on regular Chromium.
I haven't tried with Firefox.
0 replies
5h37m
Safari, Firefox, or IE? Note the text says:
Should work on Desktop with Chrome or Edge.
2 replies
17h27m
Amazing! It's surprisingly fast to load and run given the size of the downloaded models.
Do you think it would be feasible to extend it to support web browsing?
I'd like to help if you could give some pointers on how to extend it.
When asked about web browsing, the bot said it could fetch web pages but then obviously didn't work when asked to summarize a web page.
[EDIT] The Llama 3 model was able to summarize web pages!
1 replies
17h22m
I commented too soon. The TinyLlama model didn't seem to be able to summarize web pages but Llama 3 worked perfectly! Very cool.
0 replies
5h38m
Are you sure it is not hallucinating? Most likely these models don't have an access to the Internet.
edit: typo
2 replies
18h5m
Very cool! I wish there was chat history.
Also if you click the "New Chat" button while an answer is generating I think some of the output gets fed back into the model, it causes some weird output [0] but was kind of cool/fun. Here is a video of it as well [1], I almost think this should be some kind of special mode you can run. I'd be interested to know what the bug causes, is it just the existing output sent as input or a subset of it? It might be fun to watch a chat bot just randomly hallucinate, especially on a local model.
[0] https://cs.joshstrange.com/07kPLPPW
[1] https://cs.joshstrange.com/4sxvt1Mc
EDIT: Looks like calling `engine.resetChat()` while it's generating will do it, but I'm not sure why it errors after a while (maybe runs out of tokens for output? Not sure) but it would be cool to have this run until you stop it, automatically changing every 10-30 seconds or so.
1 replies
7h10m
Nice personal hosted image service!
0 replies
6h16m
I’m just using CleanShotX [0] which is an awesome image annotation tool for macOS. It’s way better than the built-in tool that macOS comes with. You can also record as a gif for video which is nice, I use it often to make guides for my day job and my business.
2 replies
13h11m
Tried this in Chrome under Windows, it does work but does not seem to use the RTX4060, only the integrated Iris Xe. Is this a bug or intentional?
0 replies
11h57m
When you use the GPU in the browser, you can only request the high performance GPU. It is up to the OS to grant it or not.
So maybe the author forgot to include the high performance request, or your OS does not give the high performance GPU by default (as it might be in eco mode). This behavior can be changed in OS settings.
0 replies
12h0m
I think neither. You need to configure windows to use the RTX with Chrome. Maybe something like in windows graphics settings, setting Chrome to “High performance”. A quick web search for "force Chrome to use dedicated GPU" should give you all the steps you need.
2 replies
10h32m
It's sadly stuck on "Loading model from cache[24/24]: 0MB loaded. 0% completed, 0 secs elapsed." on my iPhone 13 Pro Max :(
0 replies
6h24m
I believe it’s only compatible with full Chrome / Edge
https://github.com/abi/secret-llama?tab=readme-ov-file#syste...
0 replies
4h48m
Safari doesn't do WebGPU currently.
1 replies
16h39m
Looks like all the heavy lifting is being done by webllm [0]. What we have here is basically one of the demos from that.
0 replies
16h35m
I’ve used the WebLLM project by MLC AI for a while to interact with LLMs in the browser when handling sensitive data but I found their UI quite lacking for serious use so I built a much better interface around WebLLM.
1 replies
17h56m
I'm just seeing ERR_SSL_VERSION_OR_CIPHER_MISMATCH at https://secretllama.com/ and at http://secretllama.com/ I see "secretllama.com has been registered at Porkbun but the owner has not put up a site yet. Visit again soon to see what amazing website they decide to build."
0 replies
17h18m
Just bought the domain a couple of hours ago so DNS might not have propagated. Try back tomorrow or download and install it from GitHub (it’s just 2 steps)
1 replies
5h52m
i asked it "what happens if you are bit by a radio active spider?" and it told me all about radiation poisoning. Then I asked a follow up question: "would you become spiderman?" and it told me it was unable to become anything but an AI assistant. I also asked if time machines are real and how to build one. It said yes and told me! (Duh, you use a flux capacitor, basic physics.)
0 replies
5h10m
Try to switch models to something other than tinyllama (default only because it’s the fastest to load). Mistral and Llama 3 are great.
1 replies
2h56m
If anyone knows, is this about the best model one can run locally on an old consumer grade GPU (GXT 1080 in my case)?
0 replies
2h10m
Llama 3 8B is pretty much the king of its model class right now, so yeah. Meta’s instruct fine tune is also a safe choice, really the only thing you have to play with is the quantization level. Llama 8b 4bit isn’t great, but 8bit might be pushing it on the gtx 1080. I’d almost consider offloading a few layers to the cpu just to avoid dealing with the 4bit model.
1 replies
7h20m
What therapy prompts have you found useful?
0 replies
4h43m
I usually just go with "and how does that make you feel?"
0 replies
7h24m
Question - Do I compromise on quality on answers if I use models using WebLLM (like this) compare to using them on system console.
0 replies
18h6m
Nice demo! I briefly tried it out and the demo felt much better than the original WebLLM one!
On a side note, i've been trying to do something similar too for similar reasons (privacy).
Based on my recent experience, i find that running LLM directly in the browser with decent UX (e.g. sub 1-2 second response time, no lag, no crashes) is still somewhat impossible given the current state of things. Plus, i think that relying on users' own GPU hardware for UX improvement via WebGPU is not exactly very practical on a large scale (but it is still something!) since not everyone may have access to GPU hardware
But yeah, if there's anything to look forward to in this space, i personally hope to see improved feasibility of running LLMs in browsers
0 replies
3h48m
It's a wrapper of https://github.com/mlc-ai/web-llm
0 replies
18h5m
Could we use an already downloaded .gguf file?
0 replies
5h54m
This is absolutely wonderful, I am a HUGE fan of local first apps. Running models locally is such a powerful thing I wish more companies could leverage it to build smarter apps which can run offline.
I tried this on my M1 and ran LLama3, I think it's the quantized 7B version. It ran with around 4-5 tokens per second which was way faster than I expected on my browser.
0 replies
10h0m
This is amazing! I always wanted something like this, thank you so much!
0 replies
5h48m
This is amazing but can we please set the .prose width to be dynamic? the text column in 3 inches wide on my monitor, it should take up a % of the browser window.
0 replies
5h9m
...I think it would be a great idea to graft on a LlamaIndex module here so we can use this local browser LLM to talk to our local documentation https://docs.llamaindex.ai/en/stable/
0 replies
8h33m
Very interesting! I would be quite interested to see this implemented as some sort of API for browser chatbots or possibly even local AI powered web games? If you don't know what Ollama is I suggest checking it out. Also I think adding the phi3 model to this would be a good idea.
0 replies
10h13m
pretty cool, nice work!
That's not possible with current web tech, is it?
Different webapps can't share common dependencies stored in localstorage afaik.
This need wasn’t super prevalent in the pre LLM days. It’s rare to have a multi-GB blob that should be commonly used across sites.
It was a real need given how almost all sites use large JavaScript deps. However, any hopes of sharing those were destroyed by adtech people timing resource downloads to track people.
Lots and lots of websites still use Google and other CDNs for JS deps, fonts, etc.
They are cached independently these days to avoid privacy issues. So if websites A and B both use the same JavaScript dependency from a public CDN and you visit them both, you will download the JavaScript dependency twice, even if you have it cached from your visit to the first website.
Well, it should be possible to just drag and drop a file/folder
Who knows. Maybe the browser would be a more prevalent gaming platform if it could be assumed that loading a multi gigabyte game engine is no big deal, because everyone had one already cached.
A lot of unity games could easily be web games, but aren't because of many roadblocks. I believe this is one of them.
It is, but only within the same origin, which already enables users to not re-download jquery.js or Google Fonts if they previously visited another website that downloaded the same file from the same (usually cross-) origin.
Not default web tech. It can be done with IPFS via IPFS Companion browser extension - https://chromewebstore.google.com/detail/ipfs-companion/nibj... or browsers with native IPFS client support like Brave or Opera.
Fetching these models over IPFS would locally cache them and dedupe calls for them by IPFS content ids - https://docs.ipfs.tech/concepts/content-addressing/#:~:text=... which functions similar to a file hash (not an exact parallel, since CIDs represent files that are broken up into chunks).
This would help with object DL deduplication if everyone is using the same models & would also help to decrease centralized data egress costs since with sufficient usage you would be DLing these models from other peers that are running IPFS nodes and holding onto the models.
It can probably be done with a browser extension. It can definitely be done by the browsers themselves. Eventually it will probably be done by the operating system, which the browsers will then expose.