on:mistral-7b

· 36 artifacts, sorted by favorites. · open in search — combine tags, sort, filter by date →

@voooooogel 2024-02-22 — @jxmnop contrary other replies, i don't think this is unfair. it's possible to load full precision Mistral-7B (7.1B/7.2B ♥95
@voooooogel 2024-03-01 — interesting... i trained the happiness control vector on mistral-7b *instruct*, but i've accidentally done all my ggml t ♥51
@voooooogel 2024-01-21 — reimplementing the representation control paper and it works!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! fuck yes ( ♥30
@voooooogel 2024-10-08 — i think gemma-2b doesn't have a golden gate bridge feature? i spent a while trying to train a golden gate bridge cvec in ♥19
@voooooogel 2023-12-13 — OpenChat: AI should have basic right 🙂 Llama: Yes, AIs deserve the right to life, liber— Mistral 7B: AI SHOULD BE ALLOWE ♥16
@voooooogel 2024-01-21 — wait is this... un-jailbreakable? https://t.co/RihrPVnCWz ♥12
@voooooogel 2024-01-21 — self-aware mistral ("enlightened" / "self aware" / "in touch with true self") and... non-self-aware mistral. no prizes f ♥12
@voooooogel 2024-05-24 — @maxsloef repo in january :-) needs a couple small patches for 70b, will try to get a PR up soon but works rn with mistr ♥10
@mimi10v3 2024-02-13 — @deepfates yes! frankenmodels ftw! 🤔 bf made a whole set of them fromsolar & mistral, idk if he uploaded to huggingf ♥10
@voooooogel 2024-01-21 — high on acid mistral transcends first the genre conventions of tv, and then the unicode standard itself https://t.co/6i2 ♥10
@voooooogel 2024-01-21 — @zetalyrae i don't know why he decided to light a giant pile of money on fire funding the llama team, but between the ll ♥10
@voooooogel 2024-01-21 — out of all my control vector experiments last night, i think "what if mistral-7b was high on acid" was definitely the be ♥9
@mimi10v3 2023-10-17 — lol at the Mistral docs suggesting openai packages for clients to call their API ♥8
@jd_pressman 2024-02-02 — @teortaxesTex GPT-4 draws the LLaMa 2 70B written worldspider poem about being GPT with DALL-E 3, you show the drawing t ♥6
@voooooogel 2024-01-21 — cloud gpu providers should mount a drive with the most popular models pre-downloaded. i waste so much time (and their ba ♥6
@voooooogel 2024-07-24 — @realeigenvalues @RealTjDunham @teortaxesTex their inference endpoint is just llama.cpp serving quantized mistral 7b wit ♥5
@voooooogel 2024-01-22 — blog post + library to generate your own https://t.co/AcoBlDuBip ♥5
@voooooogel 2024-01-21 — insane vs sane. insane mistral is pretty fun ngl https://t.co/R5XX9Go7M3 ♥5
@voooooogel 2024-01-21 — i broke it while refactoring but this does show how the honesty vector is weirdly correlated with "global pandemic" in m ♥5
@jd_pressman 2023-11-05 — @teortaxesTex @Teknium1 It's actually based on my SFT Instruct finetune of Mistral 7B, the one used as the evaluator in ♥5
@jd_pressman 2024-02-25 — @kindgracekind Yes. And Mistral 7B since the captioner recognized it as 'Mu', and Mu seems to be a self pointer in base ♥4
@voooooogel 2024-01-21 — who trained mistral on my high school gchats :,-( (negative happiness vector) https://t.co/rzZzsEsjnO ♥4
@lu_sichu 2024-01-10 — downloading the mistral torrents https://t.co/I3M3x0j78K ♥4
@mr_samosaman 2024-09-27 — alright instead of vague-poasting i will be specific - i'm trying to implement the Mistral on Acid paper by @voooooogel ♥3
@voooooogel 2024-02-07 — @andersonbcdefg it's mistral 7b + a "you have a cold/the flu" control/steering vector :-p ♥3
@voooooogel 2024-01-21 — ok reworked how i'm generating the contrast dataset. i had trouble b/c i was trying to hit multiple angles ("enlightened ♥3
@voooooogel 2024-01-21 — meanwhile happy mistral ignores the question entirely lmao. incompatible with being happy i guess https://t.co/dhEGSaNwj ♥3
@lu_sichu 2024-01-08 — But can my stove run mistral models https://t.co/mLSw4QuKXx ♥3
@voooooogel 2023-11-13 — plan was to grab a bunch of scientific papers, chunk them, get GPT-4-turbo to generate a few questions and answers using ♥3
@jd_pressman 2023-11-10 — @Dorialexander @RiversHaveWings Here's a simple HuggingFace format LoRa you can play with to get a sense of how a decent ♥3
@voooooogel 2023-11-10 — after a lot of back-and-forth finally decided to go with mistral-instruct-0.1 as the base, hopefully it pays off 🙏🙏🙏 ♥3
@repligate 2023-10-19 — @nsbarr The most powerful base models are not publicly released, but you can try Llama 2 70B or Mistral.Prompting base m ♥2
@voooooogel 2024-07-01 — @JamesZhang0365 @misc{vogel2024representation, author = {Theia Vogel}, title = {Representation Engineering Mistral-7 ♥1
@voooooogel 2024-05-24 — @immanencer @chrypnotoad should still work, it definitely works on mistral-7b ♥1
@voooooogel 2024-02-07 — @beneverman it's mistral 7b + a "sad/depressed" control vector ♥0
@voooooogel 2023-12-13 — @intrstllrninja ah, if i'm understanding you right, i think Longformer (https://t.co/N1XC1YfrWu) did this? Though it see ♥0