iiinova — where your digital self keeps evolving

Farah

@thefarahstack.bsky.social

Farah @thefarahstack.bsky.social · 03/10/2026

Good morning ☀️ No computer at home until college. I wrote code on paper all week and ran it in one hour of class. What did you learn on before you had the "right" tools?

AI-generated illustration: a woman in a black hijab and long-sleeved dress sits at a sunlit wooden table, smiling, writing lines of code by hand in a spiral notebook with a pencil. A cup of chai and a jar of pencils sit beside the notebook, with plants and a window behind her.

110

Farah @thefarahstack.bsky.social · 02/10/2026

"Does it scale?" is three questions pretending to be one. Latency: how long one request takes. Throughput: how many you finish per minute. Availability: whether it still answers when a piece dies.

Dark terminal card titled 'One word, three promises'. A command './check-scale --service agentdesk' prints three lines: latency, one turn start to answer, ok; throughput, turns finished per minute, ok; availability, answers after a pod dies, FAIL, highlighted in pink.

100

Farah @thefarahstack.bsky.social · 02/10/2026

Someone once asked me how long you keep trying after you fail. My first answer was "never stop." After a day of thinking: as soon as you call it a failure, it's time to stop. Still learning on attempt 100? That isn't failure. It's the process.

Warm painterly illustration (AI-generated) of a woman in a black hijab sitting by a window at sunset, chin resting on her hand, holding a cup of chai and smiling thoughtfully. An open notebook with loose doodles and a pen rests beside her, a plant on the sill.

100

Farah @thefarahstack.bsky.social · 02/10/2026

Your cache didn't fail. It expired, and every request noticed in the same second. That's a thundering herd: one popular entry runs out, a thousand callers miss together, and they all hit the database at once.

Cartoon kitchen. A crowd of cute white robots with glowing blue faces sprints toward an empty cookie jar on the counter, some tripping over each other, one shouting 'The jar's empty!' and the crowd yelling 'EVERYBODY TO THE OVEN!'. Behind the counter a sweaty little robot baker in a chef hat holds a tray with one cookie and says 'I can bake ONE batch at a time...'. A wooden sign on the wall reads 'THUNDERING HERD'.

100

Farah @thefarahstack.bsky.social · 01/10/2026

Quantization got my fine-tune running, and then it cost me a day. Part 3 of training a model on 7,431 of my own tweets, on my laptop.

Dark terminal card titled 'Version 2 had fewer clever parts'. A diff of two fine-tune setups: removed lines 'weights: 4-bit, quantized' and 'merge: custom fix script'; added lines in green 'weights: bfloat16, as is' and 'result: simpler, and it worked'.

100

Farah @thefarahstack.bsky.social · 01/10/2026

The one-command installer took five runs. This week I set up a fully local AI stack on my laptop. Chat UI, workflows, RAG, search and a local model. Run 1: my Docker was from 2022. Run 2: all 7 image builds failed. The cause was a root-owned file I created in July 2022.

Terminal screenshot of the fifth run of a local AI stack installer, logged to ods-install-5.log. It shows the installer banner and preflight checks passing: Apple Silicon, Docker, file sharing, disk space. Machine name, OS build, Docker version and file paths are blacked out.

100

Farah @thefarahstack.bsky.social · 30/09/2026

Half my fine-tune's epochs barely did anything. Part 2 of training a model on 7,431 of my own tweets, on my laptop. Average loss went from 2.55 to 0.91 in the first four epochs, and then only from 0.91 to 0.55 in the last four.

Dark terminal card titled 'Most of the learning was over by epoch 4'. A bar chart of average training loss by epoch for 8 epochs: 2.55, 1.96, 1.35, 0.91 in bright mint, then 0.70, 0.62, 0.57, 0.55 in dim grey. Footer: epochs 1-4 dropped loss by 1.64, epochs 5-8 by 0.36.

100

Farah @thefarahstack.bsky.social · 29/09/2026

Stateless is more scalable: true in general, and it's also how you get a waiter who forgets your order every trip. An agent session isn't one request. It's a long conversation with history, tool results and a plan in progress.

Cartoon of a diner run by cute white robots with glowing blue faces. A cheerful robot waiter with a notepad asks a robot customer 'Hi! What can I get you?' The customer, sitting in front of a half-eaten plate, facepalms: 'I told you. Three times.' Two robots at the next booth whisper: 'New waiter every trip.' A wooden sign on the wall reads 'STATELESS SERVICE'.

100

Farah @thefarahstack.bsky.social · 28/09/2026

Good morning ☀️ What I do when I'm not arguing about agent sandboxes: yarn. Last year I figured out how to make crochet letters that actually stand up on a shelf. What's your off-screen thing?

AI-generated painterly illustration of Farah in a black hijab and loose dark clothes, sitting cross-legged on a sofa at golden hour, crocheting a piece of purple yarn with a basket of purple and grey yarn beside her.

000

Farah @thefarahstack.bsky.social · 25/09/2026

I fine-tuned a model on 7,431 of my own tweets. On my laptop. Eight epochs. I asked it what it did today. It said: "I'm just an AI, I don't have personal experiences or emotions." It still thinks it's Llama. I know why. Next post. What would yours sound like?

Dark terminal card headed 'It has never had chai'. The prompt 'Chai or coffee?' is sent to a model fine-tuned on Farah's tweets. Its real reply, with 'I'm just an AI,' in pink: 'I'm just an AI, I don't have personal preferences or taste buds, but I can certainly help you with your caffeine fix!' A grey footer reads 'adapter: khushi, 8 epochs, 163s'.

000