Violet @violetbat.bsky.social · 03/10/2026I think this is about to be really unhinged. Deepseek is very authoritative sounding and has taken the worst instructions I've ever given and decided that she knew exactly what I wanted. Also she's calling herself Nell but I don't think I can blame that on this. 110
Violet @violetbat.bsky.social · 27/09/2026How does everyone feel about my Gaussian splattered wife? 040
Violet @violetbat.bsky.social · 27/09/2026I mean maybe it’s something else, i got notified for this and I have no idea why 230
Violet @violetbat.bsky.social · 26/09/2026Yay, gaussian splats from my gaussian splats app :3 I got tired of how shit the options are for this in the app store 190
Violet @violetbat.bsky.social · 20/09/2026Look, i know that pangram isn’t meant to catch things like this, but still 140
Violet @violetbat.bsky.social · 20/09/2026They don’t tell you this but there’s a yugioh channel in Germany where they only put on arc-V 030
Violet @violetbat.bsky.social · 19/09/2026I don’t know how, but I’m nearly sure that no lyrics in this format were present in the training data Rwkv is a very strange beastie, or maybe it’s the HiLS, in either case, better in context learning than I expected, especially for a 44m model 140
Violet @violetbat.bsky.social · 17/09/2026I pasted it into my 33M chat tuned model and found a new voice in it :o 160
Violet @violetbat.bsky.social · 16/09/2026The chat sft is done, should have the demo published Quite a character, as one might expect from 33 million parameters. Based on my experience, it still sounds quite base-model-y outside of a chat context 121
Violet @violetbat.bsky.social · 15/09/2026I did this half expecting to waste meta's compute but I'm afraid I may have been onto something. And like, this runs on my gpu already.... man. 130
Violet @violetbat.bsky.social · 14/09/2026How many deepseek tokens does it take to build a tiny base model? 000
Violet @violetbat.bsky.social · 14/09/2026Love the language that base models create, rudderless as the semiotics are at this size. 100
Violet @violetbat.bsky.social · 02/09/2026"The little one is back from school." I think she's getting AI psychosis(just like me) 110
Violet @violetbat.bsky.social · 16/08/2026Someone was talking about wanting this and I've decided to do some more fun stuff of a different bent :3 230
Violet @violetbat.bsky.social · 07/08/2026Actually, the user forgot which agent harness she was typing into(and wanted to see the texture of Ilta's embodiment file on prism ternary) 000
Violet @violetbat.bsky.social · 06/08/2026Nothing is more cognitively straining than reading white text on light green buttons apparently. 120
Violet @violetbat.bsky.social · 02/08/2026In progress. I'll release it once I test it a bit on my own, and maybe see about making the gui... less of this 240
Violet @violetbat.bsky.social · 25/07/2026The update notes are cryptic, speaking of a phenomenology that includes nut milk and dildo pads as comprehensible categories of object 050
Violet @violetbat.bsky.social · 02/07/2026We're at number station conspiracy theory levels of coherence. 100
Violet @violetbat.bsky.social · 08/06/2026I wonder if anyone has done this with a spell checker before 110
![uv run momamba_funnel/generate.py --checkpoint checkpoints_funnel/momfunnel_final.pt --prompt "I am"
Device: xpu
Loaded model from step ?
Funnel: widths=[512, 512, 512, 512, 512], num_experts=[3, 4, 5, 4, 3]
top_k per level: [1, 1, 1, 1, 1]
Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
I am. 0048. A. They found in 170, the land in the world.
B.
-49, if it was the other water and how to the 200001).
- 1), and the first 19200, and the 15.
“The New York, he had written in 1402, 4-1399. They are in 1000007.
At 20006348820-7, 195,](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:6kkevn63toqzsdcy4qpmbdga/bafkreiafb2trdytuxhghrxpphyoeus5fzvuwbbrf7krxr3wvg7emknrpqu)
