<![CDATA[Stories by Yos Vichet on Medium]]> https://medium.com/@sovannaro17?source=rss-16955a9ecc9e------2 https://cdn-images-1.medium.com/fit/c/150/150/1*wwGJs7LnDMscl_Rm0QcWRw.jpeg Stories by Yos Vichet on Medium https://medium.com/@sovannaro17?source=rss-16955a9ecc9e------2 Medium Thu, 05 Nov 2026 01:18:06 GMT <![CDATA[I Read the Docs Behind a “100 Tokens a Second” Claim. Here’s What They Actually Say]]>

The same graphics card shows 93 and 4 tokens a second in one project’s docs. Here’s why both are true, and how to tell which one you’ll…

Continue reading on Medium »

]]> https://medium.com/@sovannaro17/i-read-the-docs-behind-a-100-tokens-a-second-claim-heres-what-they-actually-say-1e2e37fb2153?source=rss-16955a9ecc9e------2 https://medium.com/p/1e2e37fb2153 Wed, 04 Nov 2026 23:01:02 GMT 2026-11-04T23:01:02.819Z <![CDATA[I Compared 5 Free AI Image Models So You Don’t Have to: Here’s What Your Card Can Run]]>

Five free models, five checks, from 8 GB cards to Macs. The real cost of an image model is hiding in files the label never counts.

Continue reading on Medium »

]]> https://medium.com/@sovannaro17/i-compared-5-free-ai-image-models-so-you-dont-have-to-here-s-what-your-card-can-run-f316a2344468?source=rss-16955a9ecc9e------2 https://medium.com/p/f316a2344468 Wed, 04 Nov 2026 11:01:01 GMT 2026-11-04T11:01:01.966Z <![CDATA[16GB VRAM vs the 24GB Upgrade: What Local AI Can Actually Do]]>

Six real jobs, from everyday chat to AI video, and when a 24GB upgrade is actually worth the money.

Continue reading on Medium »

]]> https://medium.com/@sovannaro17/16gb-vram-vs-the-24gb-upgrade-what-local-ai-can-actually-do-c69df27bb720?source=rss-16955a9ecc9e------2 https://medium.com/p/c69df27bb720 Tue, 03 Nov 2026 23:01:02 GMT 2026-11-03T23:01:02.762Z <![CDATA[I Compared a PS5 to an RTX 5060 Ti for Local AI. The Winner Surprised Me]]>

A PS5 can run a 27-billion-parameter AI model at readable speed. Whether you should buy one for that is a different question.

Continue reading on Medium »

]]> https://medium.com/@sovannaro17/i-compared-a-ps5-to-an-rtx-5060-ti-for-local-ai-the-winner-surprised-me-1b3a17b844f3?source=rss-16955a9ecc9e------2 https://medium.com/p/1b3a17b844f3 Tue, 03 Nov 2026 11:01:02 GMT 2026-11-03T11:01:02.441Z <![CDATA[I Studied the Best AI Models for 6GB GPUs. Here’s What I’d Buy]]>

Your small graphics card can run bigger AI than you think. Here is why system RAM matters more than the GPU, and which models, cards, and…

Continue reading on Medium »

]]> https://medium.com/@sovannaro17/i-studied-the-best-ai-models-for-6gb-gpus-heres-what-i-d-buy-f0e1721dac3b?source=rss-16955a9ecc9e------2 https://medium.com/p/f0e1721dac3b Mon, 02 Nov 2026 23:01:02 GMT 2026-11-02T23:01:02.282Z <![CDATA[I Compared Every Way to Get 64GB for Local AI. Here Is What Matters]]>

The same 64GB of memory can cost $31 or $154 per gigabyte. Here is how to choose without overpaying.

Continue reading on Medium »

]]> https://medium.com/@sovannaro17/i-compared-every-way-to-get-64gb-for-local-ai-here-is-what-matters-024b5ea2aa2e?source=rss-16955a9ecc9e------2 https://medium.com/p/024b5ea2aa2e Mon, 02 Nov 2026 11:01:02 GMT 2026-11-02T11:01:02.055Z <![CDATA[I Studied 6 Ways to Train a Small AI Model. They All Hit the Same Wall.]]>

Six training methods, one surprising ceiling, and what it reveals about what small AI models already know.

Continue reading on Medium »

]]> https://medium.com/@sovannaro17/i-studied-6-ways-to-train-a-small-ai-model-they-all-hit-the-same-wall-488f15dc356d?source=rss-16955a9ecc9e------2 https://medium.com/p/488f15dc356d Sun, 01 Nov 2026 23:01:02 GMT 2026-11-01T23:01:02.507Z <![CDATA[8 AI Video Models, 1 Prompt: Why the Cheaper One Won]]>

One $8 clip lost to a $1.60 clip. Here’s what a side-by-side test of eight models means for your budget.

Continue reading on Medium »

]]> https://medium.com/@sovannaro17/8-ai-video-models-1-prompt-why-the-cheaper-one-won-feabd0ad7012?source=rss-16955a9ecc9e------2 https://medium.com/p/feabd0ad7012 Sun, 01 Nov 2026 11:01:01 GMT 2026-11-01T11:01:01.852Z <![CDATA[I Went Through the Benchmarks: A Used RTX 4090 Can Match an RTX 5090 With One Setting]]>

Speculative decoding can nearly double local LLM speed on the hardware you already own. Here is how to test it before you buy.

Continue reading on Medium »

]]> https://medium.com/@sovannaro17/i-went-through-the-benchmarks-a-used-rtx-4090-can-match-an-rtx-5090-with-one-setting-826e66911178?source=rss-16955a9ecc9e------2 https://medium.com/p/826e66911178 Sat, 31 Oct 2026 23:01:02 GMT 2026-10-31T23:01:02.310Z <![CDATA[I Compared Goose and Rust’s Memory Models So You Don’t Have to Guess]]>

Memory safety is only the beginning. The real question is which responsibilities the language takes off your plate, and which ones it…

Continue reading on Medium »

]]> https://medium.com/@sovannaro17/i-compared-goose-and-rusts-memory-models-so-you-don-t-have-to-guess-240c93c043c1?source=rss-16955a9ecc9e------2 https://medium.com/p/240c93c043c1 Sat, 31 Oct 2026 11:01:01 GMT 2026-10-31T11:01:01.968Z