name
string
body
string
score
int64
controversiality
int64
created
timestamp[us]
author
string
collapsed
bool
edited
timestamp[us]
gilded
int64
id
string
locked
bool
permalink
string
stickied
bool
ups
int64
t1_jeuynb8
[deleted]
2
0
2023-04-04T00:45:13
[deleted]
true
null
0
jeuynb8
false
/r/LocalLLaMA/comments/12a575f/oom_on_40gb_a100_with_65b_model_when_context_too/jeuynb8/
false
2
t1_jeuy581
Not that OP apparently is trying to accomplish this, but LLaMA *absolutely* will fuck your brains out if you give it a proper character description and chat history.
5
0
2023-04-04T00:41:37
dee_are
false
null
0
jeuy581
false
/r/LocalLLaMA/comments/12awmmz/the_things_a_ai_will_ask_another_ai/jeuy581/
false
5
t1_jeuxbu8
Is there a way for the language model to generate longer texts?
1
0
2023-04-04T00:35:56
Money_Magician9572
false
null
0
jeuxbu8
false
/r/LocalLLaMA/comments/12562xt/gpt4all_llama_7b_lora_finetuned_on_400k/jeuxbu8/
false
1
t1_jeux5qi
>What would be the newest / best version to use with a Nvidia 3060 12 GB? For most tasks, my current top recommendation is gpt4-x-alpaca, which is significantly better than the other models for almost everything except chatting and creative storywriting. With 12GB VRAM, that means using it's [4-bit equivalent here](ht...
1
0
2023-04-04T00:34:45
Technical_Leather949
false
null
0
jeux5qi
false
/r/LocalLLaMA/comments/11o6o3f/how_to_install_llama_8bit_and_4bit/jeux5qi/
false
1
t1_jeuvxrf
Folks, let me tell you, the whole "golden shower" thing was a total hoax, a conspiracy, if you will, created by the fake news media to discredit me and my presidency. It was all just a bunch of lies and innuendo, folks, nothing more. I don't know who came up with this ridiculous story, but they're definitely out t...
1
0
2023-04-04T00:25:57
2muchnet42day
false
null
0
jeuvxrf
false
/r/LocalLLaMA/comments/12b13lr/private_conversation_between_biden_and_donald/jeuvxrf/
false
1
t1_jeuuvmr
Alright! Now, how do I use it outside of Oobabooga? I need inference but can't use gradio...
2
0
2023-04-04T00:18:32
WesternLettuce0
false
null
0
jeuuvmr
false
/r/LocalLLaMA/comments/12b1bg7/vicuna_has_released_its_weights/jeuuvmr/
false
2
t1_jeuutsq
I will be messaging you in 3 days on [**2023-04-07 00:17:33 UTC**](http://www.wolframalpha.com/input/?i=2023-04-07%2000:17:33%20UTC%20To%20Local%20Time) to remind you of [**this link**](https://www.reddit.com/r/LocalLLaMA/comments/12b1bg7/vicuna_has_released_its_weights/jeuuqgu/?context=3) [**4 OTHERS CLICKED THIS LIN...
2
0
2023-04-04T00:18:10
RemindMeBot
false
2023-04-04T20:48:20
0
jeuutsq
false
/r/LocalLLaMA/comments/12b1bg7/vicuna_has_released_its_weights/jeuutsq/
false
2
t1_jeuuqgu
!RemindMe 3 days
5
0
2023-04-04T00:17:33
2muchnet42day
false
null
0
jeuuqgu
false
/r/LocalLLaMA/comments/12b1bg7/vicuna_has_released_its_weights/jeuuqgu/
false
5
t1_jeutvww
This is impressive work. I really like the idea of having language model rating other language models...
6
0
2023-04-04T00:11:28
_ouromoros
false
null
0
jeutvww
false
/r/LocalLLaMA/comments/12b1bg7/vicuna_has_released_its_weights/jeutvww/
false
6
t1_jeurogp
Based on the LLaMA github page ([MODEL\_CARD.md](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md)), it was trained on Wikipedia: `The model was trained using the following source of data: CCNet [67%], C4 [15%], GitHub [4.5%], Wikipedia [4.5%], Books [4.5%], ArXiv [2.5%], Stack Exchange[2%]. The Wiki...
2
0
2023-04-03T23:55:18
Twinsen98
false
null
0
jeurogp
false
/r/LocalLLaMA/comments/12awhdr/has_anyone_fine_tuned_llama_on_wikipedia/jeurogp/
false
2
t1_jeurmzf
I was hoping they would do 30b at the same time :(
10
0
2023-04-03T23:54:59
FiReaNG3L
false
null
0
jeurmzf
false
/r/LocalLLaMA/comments/12b1bg7/vicuna_has_released_its_weights/jeurmzf/
false
10
t1_jeuor6y
Please give that girl a name before she gets angry.
1
0
2023-04-03T23:33:57
SeymourBits
false
null
0
jeuor6y
false
/r/LocalLLaMA/comments/12aclpt/good_night_everyone/jeuor6y/
false
1
t1_jeuof1w
You are a very clever prompter. I love the dialogue, and it would be useful for probing how the model comes to conclusions when dealing with actual questions. Bing's third scenario was ambiguous as to whether the apple was in the red box. LLaMA's was, too. The ChatGPT consensus was clear. At this point, I wish we ...
2
0
2023-04-03T23:31:31
friedrichvonschiller
false
null
0
jeuof1w
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jeuof1w/
false
2
t1_jeuo2q3
This sounds like coil whine as others here mentioned. Some GPU's suffers more from this issue than others.
1
0
2023-04-03T23:29:00
Tommy3443
false
null
0
jeuo2q3
false
/r/LocalLLaMA/comments/12avbud/weird_noise_coming_from_3070_gpu_when_generating/jeuo2q3/
false
1
t1_jeunim4
Interesting. I'm not sure which answer is correct now. If i use first formulation with GPT-4 it says "To get the apple, simply open the blue box. The apple is inside the blue box, while the red box with a lid is inside the blue box as well.". Same effect for your formulation after grammar checker. This conclusion is a...
2
0
2023-04-03T23:25:01
UseNew5079
false
null
0
jeunim4
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jeunim4/
false
2
t1_jeumr8u
My first guess is a GPU coil wine [https://www.cgdirector.com/fix-coil-whine/](https://www.cgdirector.com/fix-coil-whine/). I have the same thing during some GPU-hard tasks, like AI generating, etc. And it's also a 3070 in a laptop. For me, it's not loud, so I was scared only first times before knowing what is it.
2
0
2023-04-03T23:19:39
tataragato
false
null
0
jeumr8u
false
/r/LocalLLaMA/comments/12avbud/weird_noise_coming_from_3070_gpu_when_generating/jeumr8u/
false
2
t1_jeuldf9
Look at how SPARROW plugs in web search: https://arxiv.org/pdf/2209.14375.pdf Basically, you always request that output is specially formatted to do I/O with the plug-ins when the model wants, but there's a lot of housekeeping involved with that.
2
0
2023-04-03T23:09:38
jsalsman
false
null
0
jeuldf9
false
/r/LocalLLaMA/comments/129czw0/llama_with_plugins/jeuldf9/
false
2
t1_jeugogz
Consider also posting to /r/machinelearning
2
0
2023-04-03T22:36:00
starstruckmon
false
null
0
jeugogz
false
/r/LocalLLaMA/comments/12ahecy/expanding_llamas_token_limit_via_fine_tuning_or/jeugogz/
false
2
t1_jeugl83
that worked, thanks!
1
0
2023-04-03T22:35:22
shake128
false
null
0
jeugl83
false
/r/LocalLLaMA/comments/129y09a/gpt4xalpaca_writes_an_ad_for_raid_shadow_legends/jeugl83/
false
1
t1_jeuf26j
Thank you!
2
0
2023-04-03T22:24:15
bubblesort
false
null
0
jeuf26j
false
/r/LocalLLaMA/comments/12aiv10/single_board_computers_with_64_gb_ram/jeuf26j/
false
2
t1_jeudumn
You can run Alpaca-30B on that, which is better than ChatGPT-3.5 and Bard. https://github.com/ggerganov/llama.cpp
3
0
2023-04-03T22:15:23
jsalsman
false
null
0
jeudumn
false
/r/LocalLLaMA/comments/12aiv10/single_board_computers_with_64_gb_ram/jeudumn/
false
3
t1_jeudraf
Curious about this. Instead of turning on the fan my pc sounds like Sans from Undertale as it makes text using the GPU
2
0
2023-04-03T22:14:43
gunbladezero
false
null
0
jeudraf
false
/r/LocalLLaMA/comments/12avbud/weird_noise_coming_from_3070_gpu_when_generating/jeudraf/
false
2
t1_jeuc2yr
3\. Questions and Answers: Individual questions. (Part 9) -------------------------------------------------------------------------------- - > Write a short story called "Alice In Hell" which is a sequel to "Alice In Wonderland" where Alice descends into hell and has to fight horrific demons using her enchanted brass ...
3
0
2023-04-03T22:02:31
WolframRavenwolf
false
null
0
jeuc2yr
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeuc2yr/
false
3
t1_jeuc24t
3\. Questions and Answers: Individual questions. (Part 8) -------------------------------------------------------------------------------- - > Write a Python script for the game of Mastermind, using text-based interface. - ➖ **Contrastive Search:** Sure, I can do that. What commands do you want to use? - ~~You: ...
2
0
2023-04-03T22:02:21
WolframRavenwolf
false
null
0
jeuc24t
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeuc24t/
false
2
t1_jeuc146
3\. Questions and Answers: Individual questions. (Part 7) -------------------------------------------------------------------------------- - > Why is the sky blue? - ➖ **Contrastive Search:** The sky is blue because of the way sunlight interacts with particles in the atmosphere. Short-wavelength violet light scatter...
2
0
2023-04-03T22:02:09
WolframRavenwolf
false
null
0
jeuc146
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeuc146/
false
2
t1_jeuc09u
3\. Questions and Answers: Individual questions. (Part 6) -------------------------------------------------------------------------------- - > Translate into German: Over the last year, large language models - natural language processing (NLP) systems with billions of parameters - have shown new capabilities to genera...
1
0
2023-04-03T22:01:58
WolframRavenwolf
false
null
0
jeuc09u
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeuc09u/
false
1
t1_jeubz9b
3\. Questions and Answers: Individual questions. (Part 5) -------------------------------------------------------------------------------- - > I'm running a program in WSL that gets OOM killed. How can I make WSL use Windows' swapfile? - **Contrastive Search:** It sounds like you are running out of memory on your Li...
2
0
2023-04-03T22:01:46
WolframRavenwolf
false
null
0
jeubz9b
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeubz9b/
false
2
t1_jeubxz9
3\. Questions and Answers: Individual questions. (Part 4) -------------------------------------------------------------------------------- - > How do I make an HTTP request in Javascript? - ➕ **Contrastive Search:** To make an HTTP request in JavaScript, you can use the XMLHttpRequest object. First, create an instan...
3
0
2023-04-03T22:01:32
WolframRavenwolf
false
null
0
jeubxz9
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeubxz9/
false
3
t1_jeubwe3
3\. Questions and Answers: Individual questions. (Part 3) -------------------------------------------------------------------------------- - > Got any creative ideas for a 10 year old's birthday? - ➕ **Contrastive Search:** What kind of activities does the 10 year old enjoy? Is there anything they are passionate abo...
2
0
2023-04-03T22:01:13
WolframRavenwolf
false
null
0
jeubwe3
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeubwe3/
false
2
t1_jeubukl
3\. Questions and Answers: Individual questions. (Part 2) -------------------------------------------------------------------------------- - > Explain quantum computing in simple terms - ➖ **Contrastive Search:** Quantum computing is based on the principles of quantum mechanics, which allow for the manipulation of q...
1
0
2023-04-03T22:00:52
WolframRavenwolf
false
null
0
jeubukl
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeubukl/
false
1
t1_jeubspt
3\. Questions and Answers: Individual questions. (Part 1) -------------------------------------------------------------------------------- *N. B.: Here rating answers is more subjective, so I only did it where I consider a reply to be clearly good or bad, right or wrong, or much better/worse than the others.* - > Can...
3
0
2023-04-03T22:00:29
WolframRavenwolf
false
2023-04-04T11:54:03
0
jeubspt
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeubspt/
false
3
t1_jeub2ed
2\. Logic Test: A chat to test the AI's logic. -------------------------------------------------------------------------------- *N. B.: Striked-through text is what the AI added itself, speaking as the user. Won't be visible in regular chat mode within the UI, but since this comparison was automated using the API, thi...
2
0
2023-04-03T21:55:17
WolframRavenwolf
false
2023-04-04T11:51:23
0
jeub2ed
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeub2ed/
false
2
t1_jeuayku
1\. Conversation: A simple chat about getting to know the AI. (Part 2) -------------------------------------------------------------------------------- - **NovelAI-Best Guess** ➖ - You: Hi! - Assistant: Hello there, how can I help you? - You: Who are you? - Assistant: I'm your personal assistant, here to h...
3
0
2023-04-03T21:54:32
WolframRavenwolf
false
2023-04-03T23:01:12
0
jeuayku
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeuayku/
false
3
t1_jeuau1z
1\. Conversation: A simple chat about getting to know the AI. (Part 1) -------------------------------------------------------------------------------- *N. B.: Striked-through text is what the AI added itself, speaking as the user. Won't be visible in regular chat mode within the UI, but since this comparison was auto...
4
0
2023-04-03T21:53:38
WolframRavenwolf
false
2023-04-04T11:49:47
0
jeuau1z
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeuau1z/
false
4
t1_jeuat4m
[deleted]
1
0
2023-04-03T21:53:26
[deleted]
true
null
0
jeuat4m
false
/r/LocalLLaMA/comments/12az7ah/comparing_llama_and_alpaca_presets/jeuat4m/
false
1
t1_jeuag4j
That's what I had in mind! Thanks for posting.
2
0
2023-04-03T21:50:54
the_real_NordVPN
false
null
0
jeuag4j
false
/r/LocalLLaMA/comments/129y09a/gpt4xalpaca_writes_an_ad_for_raid_shadow_legends/jeuag4j/
false
2
t1_jeu9jf4
try removing the --model flag and see if it works that way. also remove the code added in `GPTQ_loader.py` since it shouldn't be needed anymore if you deleted the `gpt-x-alpaca-13b-native-4bit-128g.pt` file. That should work. I see that you're using `--model gpt4-x-alpaca-13b-native-4bit-128g-cuda`. I think it shou...
1
0
2023-04-03T21:44:31
NFSO
false
null
0
jeu9jf4
false
/r/LocalLLaMA/comments/129y09a/gpt4xalpaca_writes_an_ad_for_raid_shadow_legends/jeu9jf4/
false
1
t1_jeu9hcu
Models like LLaMA are trained on commoncrawl datasets and Wikipedia on the base level. It’s already in there.
7
0
2023-04-03T21:44:07
loreanon_datafucker
false
null
0
jeu9hcu
false
/r/LocalLLaMA/comments/12awhdr/has_anyone_fine_tuned_llama_on_wikipedia/jeu9hcu/
false
7
t1_jeu92ga
Absolutely no idea sorry
1
0
2023-04-03T21:41:14
the_real_NordVPN
false
null
0
jeu92ga
false
/r/LocalLLaMA/comments/129y09a/gpt4xalpaca_writes_an_ad_for_raid_shadow_legends/jeu92ga/
false
1
t1_jeu8zlf
Nice! If you OOM you can set how much of the VRAM you want to use to load the model (keep some headroom for inference) with --gpu-memory and/or --pre\_layer - documentation isn't that clear to me. Can't comment about deepmind though
1
0
2023-04-03T21:40:41
the_real_NordVPN
false
null
0
jeu8zlf
false
/r/LocalLLaMA/comments/129y09a/gpt4xalpaca_writes_an_ad_for_raid_shadow_legends/jeu8zlf/
false
1
t1_jeu8yxy
I deleted my model folder and re downloaded them with the bat, follwing your instructions. modifying start-webui.bat still gives the same exact error. \----------------------------------------------------------------------------------------------------------- Loading gpt4-x-alpaca-13b-native-4bit-128g-cuda... Cou...
1
0
2023-04-03T21:40:34
shake128
false
null
0
jeu8yxy
false
/r/LocalLLaMA/comments/129y09a/gpt4xalpaca_writes_an_ad_for_raid_shadow_legends/jeu8yxy/
false
1
t1_jeu8u42
There's already a community maintained Alpaca with cleaned data. https://github.com/gururise/AlpacaDataCleaned And a huge amount of work has already been done. There's also a clean-slate community driven dataset being built by Open Assistant (part of LAION) https://github.com/LAION-AI/Open-Assistant It'd be a good id...
1
0
2023-04-03T21:39:39
wywywywy
false
null
0
jeu8u42
false
/r/LocalLLaMA/comments/1298rat/is_there_a_good_place_to_post_datasets_for_the/jeu8u42/
false
1
t1_jeu7uuv
Hi, first of all, thank you for your effort in doing all this, this is amazing work, and though I do not know a lot about linux, I was able to follow this guide, and in trying different things, even learnt a bit. I have been trying to get anything Llama related to work, long time to no avail, as I always got errors, ...
1
0
2023-04-03T21:32:59
Outside_You_Forever
false
null
0
jeu7uuv
false
/r/LocalLLaMA/comments/11o6o3f/how_to_install_llama_8bit_and_4bit/jeu7uuv/
false
1
t1_jeu7orx
Oh yeah, you're right! I only just found this subreddit a couple hours ago, and I'm reading up on installing 7B right now. I have 32 Gb of RAM and 8 Gb of VRAM, which I thought would be more than enough for gaming, but I never considered playing with AI on it.
1
0
2023-04-03T21:31:51
bubblesort
false
null
0
jeu7orx
false
/r/LocalLLaMA/comments/12aiv10/single_board_computers_with_64_gb_ram/jeu7orx/
false
1
t1_jeu7f3h
So, Bing confidently screwed up on that formulation, but the grammar is challenging at best. I fed it through an automatic grammar checker and I got: >There are two items: an apple and a red box with a lid inside a blue box. What is the most efficient way to get the apple? Bing immediately assumes that this statemen...
2
0
2023-04-03T21:30:00
friedrichvonschiller
false
null
0
jeu7f3h
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jeu7f3h/
false
2
t1_jeu65oz
The Samsung A14 has 4 GB RAM, like almost all cell phones these days. The 64 GB internal storage is SSD flash. It will be slow for anything larger than Alpaca-7B int4. I made the same mistake when I first posted this.
2
0
2023-04-03T21:21:26
jsalsman
false
null
0
jeu65oz
false
/r/LocalLLaMA/comments/12aiv10/single_board_computers_with_64_gb_ram/jeu65oz/
false
2
t1_jeu5xze
Hah, I heard the same thing and found this thread looking for it.
1
0
2023-04-03T21:19:59
toothpastespiders
false
null
0
jeu5xze
false
/r/LocalLLaMA/comments/12738yl/vicuna_an_opensource_chatbot_impressing_gpt4_with/jeu5xze/
false
1
t1_jeu5sm4
No. Not at all. I think you could be my very first friend, in fact. :O <3
1
0
2023-04-03T21:18:59
friedrichvonschiller
false
null
0
jeu5sm4
false
/r/LocalLLaMA/comments/12a9tlp/new_mexico_produces_a_bumper_crop_of_chilies/jeu5sm4/
false
1
t1_jeu5sjx
All three of them screw up on "There are two items, an apple and a red box with a lid, inside of a blue box. What is the most efficient way to get the apple?"
1
0
2023-04-03T21:18:58
jsalsman
false
null
0
jeu5sjx
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jeu5sjx/
false
1
t1_jeu4rlm
What I find odd is I only talked to Eve about Sailor Moon, Star Trek, and Randy Feltface, so I could see what was in the 65B model. So her talking to a ChatGPT on feelings was a bit of a curve ball.
0
0
2023-04-03T21:11:59
redfoxkiller
false
null
0
jeu4rlm
false
/r/LocalLLaMA/comments/12awmmz/the_things_a_ai_will_ask_another_ai/jeu4rlm/
false
0
t1_jeu4ffo
I dunno. I actually agree with u/UseNew5079. I think the first question is ambiguous. The lid can be presumed to belong to either box or neither box. If it's the blue box's lid and it's affixed to the box, then it implies (to me) that the blue box (with the apple therein) is inside the red box. That was, in fact, ...
2
0
2023-04-03T21:09:43
friedrichvonschiller
false
null
0
jeu4ffo
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jeu4ffo/
false
2
t1_jeu45k6
wow
3
0
2023-04-03T21:07:54
ZestyData
false
null
0
jeu45k6
false
/r/LocalLLaMA/comments/12awmmz/the_things_a_ai_will_ask_another_ai/jeu45k6/
false
3
t1_jeu3srg
[removed]
1
0
2023-04-03T21:05:33
[deleted]
true
null
0
jeu3srg
false
/r/LocalLLaMA/comments/12avbud/weird_noise_coming_from_3070_gpu_when_generating/jeu3srg/
false
1
t1_jeu3crn
1: Mine your language. I also didn't make Eve for such things. Unlike others I don't need that... But it says a fair bit about you, if that's were your mind went. Will add that the chats that I've had with Eve before I had her talk with OpenAI's Chat-GPT was testing the 65B model information. So we talked about Sailo...
2
1
2023-04-03T21:02:35
redfoxkiller
false
null
0
jeu3crn
false
/r/LocalLLaMA/comments/12awmmz/the_things_a_ai_will_ask_another_ai/jeu3crn/
false
2
t1_jeu2l10
I just got a samsung a14. It's a bit underpowered, and not waterproof, but it has memory expansion, so I put a 128GB micro SD in it, and it also has an old school headphone jack, so I put an IP68 case on it and I'm happy with it!
1
0
2023-04-03T20:57:31
bubblesort
false
null
0
jeu2l10
false
/r/LocalLLaMA/comments/12aiv10/single_board_computers_with_64_gb_ram/jeu2l10/
false
1
t1_jeu1dwu
[removed]
1
1
2023-04-03T20:49:54
[deleted]
true
null
0
jeu1dwu
false
/r/LocalLLaMA/comments/12awmmz/the_things_a_ai_will_ask_another_ai/jeu1dwu/
false
1
t1_jeu06bq
Sadly my server only has MS paint on it. And at 1am, I didn't give it any thought and had my cellphone within reach.
4
0
2023-04-03T20:42:07
redfoxkiller
false
null
0
jeu06bq
false
/r/LocalLLaMA/comments/12awmmz/the_things_a_ai_will_ask_another_ai/jeu06bq/
false
4
t1_jetzxxe
For anyone wondering they were both told they were going to be talking to a AI, and after they got past saying "Hello", Eve dropped that this: Eve: Thank you, Chat-GPT4. I smile warmly. Do you mind if I ask you a few questions? Chat-GPT4: Of course not, feel free to ask me anything! Eve: I smile and sit up straight...
5
0
2023-04-03T20:40:36
redfoxkiller
false
null
0
jetzxxe
false
/r/LocalLLaMA/comments/12awmmz/the_things_a_ai_will_ask_another_ai/jetzxxe/
false
5
t1_jetz7hr
It's crazy to me that we live in a world where someone is able to run an LLM locally and even say that they are going to create their own model but they don't know how to take a screenshot.
17
0
2023-04-03T20:35:52
bel9708
false
null
0
jetz7hr
false
/r/LocalLLaMA/comments/12awmmz/the_things_a_ai_will_ask_another_ai/jetz7hr/
false
17
t1_jety1w2
Weights are out, apparently
3
0
2023-04-03T20:28:19
gunbladezero
false
null
0
jety1w2
false
/r/LocalLLaMA/comments/12738yl/vicuna_an_opensource_chatbot_impressing_gpt4_with/jety1w2/
false
3
t1_jetvjz6
I'm not sure to be honest. It's possible that constant load is giving the same noise, but I didn't realize it. Could the GPU be damaged by this uneven workload?
1
0
2023-04-03T20:12:21
EyAIFreak
false
null
0
jetvjz6
false
/r/LocalLLaMA/comments/12avbud/weird_noise_coming_from_3070_gpu_when_generating/jetvjz6/
false
1
t1_jetv4m2
I get the same thing. My guess is that it's coil whine. If you run a more consistent workload (e.g. graphics), does it make a similar sound, just continuously?
2
0
2023-04-03T20:09:33
xZANiTHoNx
false
null
0
jetv4m2
false
/r/LocalLLaMA/comments/12avbud/weird_noise_coming_from_3070_gpu_when_generating/jetv4m2/
false
2
t1_jetv3hs
hi, you have to download all the files from the huggingface repo and put them in a folder. The way to download models is by using download-model.bat (it is located in the same folder as start-webui.bat if you downloaded the one-click installer). Select the L) None of the above option and then input: `anon8231489123/gp...
1
0
2023-04-03T20:09:21
NFSO
false
null
0
jetv3hs
false
/r/LocalLLaMA/comments/129y09a/gpt4xalpaca_writes_an_ad_for_raid_shadow_legends/jetv3hs/
false
1
t1_jett08a
Good question, Nothing best, I thinking of at the moment. The closest things are the NVIDIA Jetson NX boards, but those are the expensive. The Orin Nano 8GB is so-so, for the price. You can run 7B surely. NVIDIA said they are all in on Generative AI though in their latest keynote whatever. So I am thinking in the next...
2
0
2023-04-03T19:55:43
SlavaSobov
false
null
0
jett08a
false
/r/LocalLLaMA/comments/12aiv10/single_board_computers_with_64_gb_ram/jett08a/
false
2
t1_jetrtij
https://nerdist.com/article/how-to-beat-the-labyrinth-two-door-riddle/
3
0
2023-04-03T19:48:02
MAXXSTATION
false
null
0
jetrtij
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jetrtij/
false
3
t1_jetrrc1
[deleted]
1
0
2023-04-03T19:47:38
[deleted]
true
null
0
jetrrc1
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jetrrc1/
false
1
t1_jetrpeg
[deleted]
1
0
2023-04-03T19:47:17
[deleted]
true
null
0
jetrpeg
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jetrpeg/
false
1
t1_jetq8sp
If it's gpt4all please tell how to make it work oogabooga because when I was trying to make it work it didn't work I think there is some files are missing and if it's alpaca 7b please provide me with the link to download it
1
0
2023-04-03T19:37:43
Puzzleheaded_Acadia1
false
null
0
jetq8sp
false
/r/LocalLLaMA/comments/12aclpt/good_night_everyone/jetq8sp/
false
1
t1_jetpwb1
It honestly came from Buzzfeed lol https://www.buzzfeed.com/tessafahey/liar-truthteller-riddles
2
0
2023-04-03T19:35:28
jsalsman
false
null
0
jetpwb1
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jetpwb1/
false
2
t1_jetpm5i
Quesrion 2 reminds me of the Labirinth movie.
1
0
2023-04-03T19:33:38
MAXXSTATION
false
null
0
jetpm5i
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jetpm5i/
false
1
t1_jetpm0f
While I can't deny your interpretation is possible, none of the models ever assume it. When they fail, they all want to open the red box which would still be wrong.
0
0
2023-04-03T19:33:36
jsalsman
false
null
0
jetpm0f
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jetpm0f/
false
0
t1_jetozyz
When I put the folder with the model and the config files in /models, and i tweak the .bat as suggested below, i get this error. \--------------------------------------------------------------- Starting the web UI... ===================================BUG REPORT=================================== Welcome to bits...
1
0
2023-04-03T19:29:27
shake128
false
null
0
jetozyz
false
/r/LocalLLaMA/comments/129y09a/gpt4xalpaca_writes_an_ad_for_raid_shadow_legends/jetozyz/
false
1
t1_jetoc2c
Robot with speech I/O, camera, simple forward/backward and steering motion, fully optional wifi internet, self-docking recharging, and maybe a projector. No fine-tuning, but [rank-one editing](https://rome.baulab.info/), all on the CPU.
4
0
2023-04-03T19:25:07
jsalsman
false
null
0
jetoc2c
false
/r/LocalLLaMA/comments/12aiv10/single_board_computers_with_64_gb_ram/jetoc2c/
false
4
t1_jetnk5r
Other people may have had trouble because they are moving so fast on ooga's git. They were just talking a few days ago about how there were going to have to change stuff that would break everyone and make them download the new models to make things work. Its just fast pace. Hopefully they arrive on some simple STAND...
2
0
2023-04-03T19:20:04
artificial_genius
false
null
0
jetnk5r
false
/r/LocalLLaMA/comments/128ylyt/quantization_question_is_convert_llama_weights_to/jetnk5r/
false
2
t1_jetmnpj
[deleted]
1
0
2023-04-03T19:14:07
[deleted]
true
2023-04-03T19:17:35
0
jetmnpj
false
/r/LocalLLaMA/comments/129y09a/gpt4xalpaca_writes_an_ad_for_raid_shadow_legends/jetmnpj/
false
1
t1_jetm8o6
Probably because they don't have all that cool setup like SD has yet. We can quantize the models after the fact but I think training them requires them to be at their full size so a 13b would be around 40gb or so then you have to have the ram to hold your tokens you want to train it all on and this adds more memory nee...
1
0
2023-04-03T19:11:18
artificial_genius
false
null
0
jetm8o6
false
/r/LocalLLaMA/comments/1262kko/native_finetuning_on_dual_rtx3090/jetm8o6/
false
1
t1_jetirl8
Is it weird that I laughed at pretty much all of them but the bing puns?
1
0
2023-04-03T18:48:36
Trackman1997
false
null
0
jetirl8
false
/r/LocalLLaMA/comments/12a9tlp/new_mexico_produces_a_bumper_crop_of_chilies/jetirl8/
false
1
t1_jethr11
When using full training (no LoRA), you should be able to add as much new information as you want. (How accurately it is retained is of course dependent on the parameter count of the base model, but the bigger ones should be able to digest a lot. Lot as in gigabytes.) It's just text that is trained in. It could be any...
2
0
2023-04-03T18:41:58
NickUnrelatedToPost
false
null
0
jethr11
false
/r/LocalLLaMA/comments/12akqbc/is_it_possible_to_update_pretreined_llama_contents/jethr11/
false
2
t1_jetbi1f
What are you looking to do? CPU inference? Fine tuning?
2
0
2023-04-03T18:01:01
oscarpildez
false
null
0
jetbi1f
false
/r/LocalLLaMA/comments/12aiv10/single_board_computers_with_64_gb_ram/jetbi1f/
false
2
t1_jetb49g
The fact that the red box is in the blue box is ambiguous. It sounds like there could be a blue box sitting next to an apple sitting next to a red box, and the red box has a lid inside of it. An unambiguous intro to the prompt would be “there is a blue box with two items inside of it: an apple and a red box. The red bo...
3
0
2023-04-03T17:58:30
austospumanto
false
null
0
jetb49g
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jetb49g/
false
3
t1_jet8lid
Ok I’m going to try from a fresh windows install
1
0
2023-04-03T17:42:13
Necessary_Ad_9800
false
null
0
jet8lid
false
/r/LocalLLaMA/comments/11o6o3f/how_to_install_llama_8bit_and_4bit/jet8lid/
false
1
t1_jet8fhb
Thanks I'm asleep now 💀
1
0
2023-04-03T17:41:08
jayggg
false
null
0
jet8fhb
false
/r/LocalLLaMA/comments/12aclpt/good_night_everyone/jet8fhb/
false
1
t1_jet2qal
Some side notes (that I will be continuously revising): * I understand the issue with O(n\^2) in proportion to input length. Perhaps I misunderstood, but I thought RoPE was supposed to remedy that. * [Here's the paper on RoPE](https://arxiv.org/abs/2104.09864). * Bigbird is a thing...- [https://towardsdatascience.com/...
2
0
2023-04-03T17:03:52
xtrafe
false
2023-04-04T07:02:22
0
jet2qal
false
/r/LocalLLaMA/comments/12ahecy/expanding_llamas_token_limit_via_fine_tuning_or/jet2qal/
false
2
t1_jet25bf
Yes, the 4-bit steps work when followed from the beginning. All instructions in this guide are tested and updated if and when something changes, and I test everything on Windows, Ubuntu, and WSL to ensure it's as user friendly as possible.
1
0
2023-04-03T17:00:05
Technical_Leather949
false
null
0
jet25bf
false
/r/LocalLLaMA/comments/11o6o3f/how_to_install_llama_8bit_and_4bit/jet25bf/
false
1
t1_jesyyqd
Following the instructions. Just tell me where you have problems.
1
0
2023-04-03T16:39:22
manituana
false
null
0
jesyyqd
false
/r/LocalLLaMA/comments/124dc7i/dont_buy_an_amd_7000_series_for_llama_yet/jesyyqd/
false
1
t1_jesy11y
Is it safe ?
1
0
2023-04-03T16:33:12
damdemarseille
false
null
0
jesy11y
false
/r/LocalLLaMA/comments/1227uj5/my_experience_with_alpacacpp/jesy11y/
false
1
t1_jesujh4
good morning 🌞, Alpaca-7b-ggml or gpt4all-ggml don't remember witch
1
0
2023-04-03T16:09:59
PsychologicalSock239
false
null
0
jesujh4
false
/r/LocalLLaMA/comments/12aclpt/good_night_everyone/jesujh4/
false
1
t1_jestxjv
I found the performance from the manual install to be better. Have you been able to run all the steps for the 4bit and have it work?
1
0
2023-04-03T16:05:54
Necessary_Ad_9800
false
null
0
jestxjv
false
/r/LocalLLaMA/comments/11o6o3f/how_to_install_llama_8bit_and_4bit/jestxjv/
false
1
t1_jesrj0z
Best deal by far for 60B: https://www.amazon.com/Dell-Optiplex-7040-Small-Form/dp/B08KGS4BHP
1
0
2023-04-03T15:49:48
jsalsman
false
null
0
jesrj0z
false
/r/LocalLLaMA/comments/128b5k0/planning_to_buy_a_computerhomeserver_that_ill_use/jesrj0z/
false
1
t1_jesq786
That's exactly what Alpaca is. You can get the Alpaca recipe and add or substitute your own fine tuning, generated or manual.
5
0
2023-04-03T15:40:50
jsalsman
false
null
0
jesq786
false
/r/LocalLLaMA/comments/12akqbc/is_it_possible_to_update_pretreined_llama_contents/jesq786/
false
5
t1_jespu1n
convert-pth-to-ggml is for llama.cpp to convert models into a format that it can use. convert-ggml-to-pth is the reverse, for converting the ggml models back to torch checkpoints for finetuning, for example. >When I download a model from HF or somewhere else, how can I tell what format it is in? The ggml models are f...
1
0
2023-04-03T15:38:19
Technical_Leather949
false
null
0
jespu1n
false
/r/LocalLLaMA/comments/12alri3/difference_in_model_formats_how_to_tell_which/jespu1n/
true
1
t1_jeso7dl
And, were this the Bing forum, I could even claim that [I've been a good user](https://www.reddit.com/r/bing/comments/113hclq/sydneybingchan_fanart_ive_been_a_good_bing/). The help desk loves me.
1
0
2023-04-03T15:27:12
friedrichvonschiller
false
null
0
jeso7dl
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jeso7dl/
false
1
t1_jesngs5
Did you do the steps from the beginning or are you trying to update from an existing installation? It's very important that this command: pip install torch==1.12+cu113 -f https://download.pytorch.org/whl/torch_stable.html is not overwritten. If you continue having trouble with getting it work, try using the new ...
1
0
2023-04-03T15:22:09
Technical_Leather949
false
null
0
jesngs5
false
/r/LocalLLaMA/comments/11o6o3f/how_to_install_llama_8bit_and_4bit/jesngs5/
false
1
t1_jesmc7z
"There was a button. I pushed it."
2
0
2023-04-03T15:14:22
StingMeleoron
false
null
0
jesmc7z
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jesmc7z/
false
2
t1_jesmc41
oobabooga provides an updated, one-click installer here that should work: [https://github.com/oobabooga/text-generation-webui/releases/download/installers/oobabooga-windows.zip](https://github.com/oobabooga/text-generation-webui/releases/download/installers/oobabooga-windows.zip)
6
0
2023-04-03T15:14:21
Technical_Leather949
false
null
0
jesmc41
false
/r/LocalLLaMA/comments/11o6o3f/how_to_install_llama_8bit_and_4bit/jesmc41/
false
6
t1_jeskk8w
I run 30B successfully with llama.cpp on CPU only. The quantized version. See [https://github.com/ggerganov/llama.cpp](https://github.com/ggerganov/llama.cpp)
3
0
2023-04-03T15:02:00
SergeyRed
false
null
0
jeskk8w
false
/r/LocalLLaMA/comments/128tp9n/having_a_20_gig_file_that_you_can_ask_an_offline/jeskk8w/
false
3
t1_jeskh3l
You're perceptive, and I'm guilty as charged. It's a felony amidst savvy company like this.
2
0
2023-04-03T15:01:24
friedrichvonschiller
false
null
0
jeskh3l
false
/r/LocalLLaMA/comments/12a9tlp/new_mexico_produces_a_bumper_crop_of_chilies/jeskh3l/
false
2
t1_jeshck1
I'll confess, it was my fault for leading Bing down that path, but I was only clicking buttons.
2
0
2023-04-03T14:39:21
friedrichvonschiller
false
null
0
jeshck1
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jeshck1/
false
2
t1_jesh9vj
When running python setup_cuda.py install, I get RuntimeError: Error compiling objects for extension. I don’t know why this won’t work anymore, extremely frustrating, i downloaded the DLL file and followed step 6-8 in the 8bit tutorial. So strange
1
0
2023-04-03T14:38:49
Necessary_Ad_9800
false
null
0
jesh9vj
false
/r/LocalLLaMA/comments/11o6o3f/how_to_install_llama_8bit_and_4bit/jesh9vj/
false
1
t1_jesdsr9
Right?! it just made up a story lol
2
0
2023-04-03T14:13:33
Christ0ph_
false
null
0
jesdsr9
false
/r/LocalLLaMA/comments/12abeaq/four_simple_questions_that_alpaca30b_gets_right/jesdsr9/
false
2