Will Finger

Once you read you never be the same I present you 3 amazing christian books, Have you ever read Watchman Nee, Jessie Pen Lewis and G H Pember? Earth's Earliest Ages by G.H. Pember goodreads.com/book/show/1101… The Spiritual Man by Watchman Nee goodreads.com/book/show/8497… War on the Saints, The Full Text, Unabridged Edition by Jessie Penn-Lewis goodreads.com/book/show/2835…

@RichardHolmboe @sudoingX x.com/willfi/status/…

@sudoingX What is your context window for qwen 3.5 35B-A3B?

if you're about to download nvidia's nemotron cascade 2 at Q4_K_M for a single RTX 3090, stop. save yourself the frustration i went through last night. Q4_K_M is 24.5GB. your 3090 has 24GB VRAM. the model loads, no room for KV cache, no room for context, no room for compute buffer. it will not run. this is a MoE architecture where the expert weights don't compress well at standard Q4. every quant table online lists it as "recommended" without checking if it fits consumer VRAM. the fix: bartowski IQ4_XS at 18.17GB. imatrix quantization that's smarter about which weights need precision and which don't. same 4-bit tier, 6GB smaller because it doesn't blindly keep every expert at the same precision. leaves you 5.4GB of headroom for KV cache and context. downloading it now on the same RTX 3090 i ran qwen 3.5 35B-A3B on at 112 tok/s. same machine, same node, same everything. first up is context scaling sweep from 4K to 262K to see how mamba-2 handles long context compared to qwen's deltanet. then speed benchmarks at each context level. then i'm pointing hermes agent at it for autonomous coding sessions to see how it handles tool calls, file creation, and multi-step builds over long sessions. nvidia vs alibaba. mamba vs deltanet. same hardware, different architectures. i'll report back with exact flags, exact numbers, exact VRAM breakdowns. no theory, no spec sheets. tested data from a real card.

@sudoingX Nice one! For 16gb VRAM Windows: The qwen 3.5 35b and 27b are the best ones still. x.com/willfi/status/…

I tried to warn designers. This is exactly why I built @MagicPathAI, you need to lean more into AI native workflow. You're getting outpaced by product and engineers now. Prototypes are the new PRDs. If PMs are doing this, what do you do? You either own that or you're out.

@0xCVYH Pra quem tem Windows: x.com/willfi/status/… Tmj!

Qwen3.5 4B rodando local com 4GB de RAM. Pesquisou 20+ sites, citou fontes, e encontrou a melhor resposta. O detalhe: fez tool calls + web search DURANTE o processo de raciocinio. Nao eh resposta pronta da memoria. Eh pesquisa ativa em tempo real. Modelo de 4 BILHOES de parametros. No seu notebook. Sem cloud. Sem API. Sem custo. Via Unsloth Studio (open source). Se voce tem um notebook com 4GB sobrando, ja pode ter seu proprio pesquisador de IA local.

@stevibe 27b is the best of them. But how does it compare to Opus 4.6? I wonder

"122B has to be smarter than 27B" I showed 4 UI components to three Qwen3.5 models and asked them to recreate them from a screenshot alone: - 27B (dense) - 35B-A3B (MoE) - 122B-A10B (MoE) Same screenshot. Same prompt. Same task. Which one do you think nailed it?

In 72 hours I got over 100k of value 1. Lambda gave me 5000$ credits in compute 2. Nvidia offered me 8x H100s on the cloud (20$/h) idk for how long but assuming 2 weeks that'd be 5000$~ 3. TNG technology offered me 2 weeks of B200s which is something like 12000$ in compute 4. A kind person offered me 100k in GCP credits (enough to train a 27B if you do it right) 5. Framework offered to mail me a desktop computer 6. We got 14,000$ in donations which will go to buying 2x RTX Pro 6000s (bringing me up to 384GB VRAM) 7. I got over 6M impressions which based on my RPM would be 1500$ over my 500$~ usual per pay period 8. I have gained 17,000~ followers, over doubling my follower count 9. 17 subscribers on X + 700 on youtube. The total value of all this approaches at minimum 50,000$~ and closer to 150,000$ if I leverage it all. --------------------- What I'll be doing with all this: Eric is an incredibly driven researcher I have been bouncing ideas off of over the last month. Him and I have been tackling the idea of getting massive models to fit on relatively cheap memory. The idea is taking advantage of different forms of memory, in combination with expert saliency scoring, to offload specific expert groupings to different memory tiers. For the MoEs I've tested over my entire AI session history about 37.5% of the model is responsible for 95% of token routing. So we can offload 62.5% of an LLM onto SSD/NVMe/CPU/Cheap VRAM this should theoretically result in minimal latency added if we can select the right experts. We can combine this with paged swapping to further accelerate the prompt processing, if done right we are looking at very very decent performance for massive unquantisation & unpruned LLMs. You can get DeepSeek-v3.2-speciale at full intelligence with decent tokens/s as long as you have enough vram to host the core 20-40% of the model and enough ram or SSD to host the rest. Add quantisation to the mix and you can basically have decent speeds and intelligence with just 5-10% of the model's size in vram (+ you need some for context) The funds will be used to push this to it's limits. ----------------- There's also tons of research that you can quantise a model drastically, then distill from the original BF16 or make a LoRA to align it back to the original mostly. This will be added to the pipeline too. ------------------ All this will be built out here: github.com/0xSero/moe-com… you will be able to take any MoE and shove it in here, and with only 24GB and enough RAM/NVMe to compress it down. it'll be slow as hell but it will work with little tinkering. ------------------ Lastly I will be looking into either a full training run from scratch -> or just post-training on an open AMERICAN base model - a research model - an openclaw/nanoclaw/hermes model - a browser-use model To prove that this can be done. -------------------- I will be bad at all of it, and doubt I will get beyond the best small models from 6 months ago, but I want to prove it's no boogeyman impossible task to everyone who says otherwise. -------------------- By the end of the year: 1. I will have 1 model I trained in some capacity be on the top 5 at either pinchbench, browseruse, or research. 2. My github will have a master repo which combines all my work into reusable generalised scripts to help you do that same. 3. The largest public comparative dataset for all MoE quantisations, prunes, benchmarks, costs, hardware requirements. -------------------------- A lot of this will be lead by Eric, who I will tag in the next post. I want to say thank you to everyone who has supported me, I have gotten a lot of comments stating: 1. I'm crazy, stupid, or both 2. I'm wasting my time, no one cares about this 3. This is not a real issue I believe the amount of interest and support I've received says it all. donate.sybilsolutions.ai

@sudoingX RTX 5080 + AMD 9800x3D + 96gb ram ddr5. Running Qwen 35b 100 T/S with 128k context

i just became a mod of x/LocalLLaMA. if you're running local models on your own hardware and want in, the community is open. pinned and highlighted on my profile. approving members starting today. drop your setup below and i'll get you in. 3060, 3090, 4090, 5090, AMD, whatever you're running. all welcome. if you're hitting issues with hermes agent, llama.cpp, model selection, configs, i'm here. let's make local AI accessible for everyone.

A new free tool for designers is on the way. Made by designers, for designers. Coming soon.

@BrettFromDJ Love it the depth! What do you think about clutter/noise? How to mitigate in a scalable DS?

More interface details. 😍

we are so back

@0xSero When browser? Only Cursor have?

OpenCode Desktop app updates worth checking out. 1. Now we have Queue mode, which I am a huge fan of. 2. They've enabled adding custom providers in settings. 3. Performance seems improved, it's less slow in large sessions and thread.

@Poloolpp Usp and awp nerfs

If you're not running the P2K now you're trolling

@Ozzny_CS2 Yep. Awp needs 6 bullets instead 5 to balance now.

Here's how the new Reload system works in CS2 ‼️ > When you reload, you drop the used magazine and get a full new one > Each reload takes 1 magazine from the counter What do we think, W or L change?

@EkeN3tt @Interloper_CS @gabefollower That’s good, smoke spray was op

@Interloper_CS @gabefollower YAY CZ meta instead of Ak or M4. I don't think people understand how much this affects the game. I mean if you spray 15 bullets in a smoke u will not want to reload

Counter-Strike developers have just released one of the biggest updates to the meta. Now, if you reload your weapon while there is still ammo in the magazine, the ammo disappears instead of being returned to your reserves. They have also increased (and decreased) the number of bullets for some weapons. The CZ-75, for example, now has 36 bullets instead of 24.

@Tunahan60Aurora @ThourCS2 @CounterStrike 3 bullets if you do that

This is how the NEW Reloading function works. - It now shows the magazine count - Early reload discards the whole magazine GG @CounterStrike

@LCamm12 @ThourCS2 @CounterStrike 😂😂😂😂

@ThourCS2 @CounterStrike Horrible IMO? Essentially destroyed spamming smokes or any walls that can be penetrated

@lisskins @ThourCS2 @CounterStrike Winnable in casual

@ThourCS2 @CounterStrike Bring back the 10th bullet in the AWP

@_san24k @ThourCS2 @CounterStrike This makes sense now

@ThourCS2 @CounterStrike 30 bullets after reloading with a fully chambered weapon doesnt make sense. should be 31

@ThourCS2 @CounterStrike Excellent.

I have to repost this