Tagged “llm”

Fine-Tuning a Small Model to Do One Dull Thing Well

A Spare GPU and a Model That Lives in My House

The Helpful Assistant That Confidently Deleted My Branch

Running a Local LLM on Hardware That Should Have Said No

Running llama.cpp on a Machine That Should Know Better

I Wired an LLM Into My Shell, Here's Where It Went Wrong

a tiny model on a tinier machine

i let the agent type, but i keep my hand on the wheel

the llm in my shell, and where it drew blood

the cupboard model, one card and one prompt

a spare gpu, a quiet llm, and a weekend gone

a spare 3060 and an evening with ollama

a full walkthrough of fine-tuning a small model for one dull task

fine-tuning a tiny model to do one dull job well

search that understands what you meant, running on the laptop

running a 7b model on hardware that should know better

letting the robot drive, with my hand near the wheel

my first rag pipeline retrieved confidently and answered wrong

letting an agent touch the code, with the safety on

a small agent that books its own train tickets

putting a spare gpu to work as a private model server

a tiny model that does one dull thing well

which local model actually helps me write code

i built an agent that does the boring bits, and learned where the seams are

running a 7b model on a thin client i bought for a tenner

letting an agent touch the code, but never on its own

the second draft of my little agent

a small agent that does the boring bits

the smallest agent loop that earned its keep

my first rag pipeline retrieved confidently wrong things

a language model lives in my terminal now, and mostly that's fine

a tiny agent, a tool loop, and a healthy fear of let-it-rip

the spare gpu finally earns its keep

letting an agent touch the code, with the safety on

a small agent that earns its keep

what local code models are actually good for, after a month of trying

an llm in my shell, and the handful of times it bit me

my first rag pipeline was confidently useless

my first rag pipeline was confidently wrong

putting a spare 3090 to work as a local llm box

a language model in my terminal, and the days it turned on me

my first rag pipeline was a confident liar

on-device embeddings, or how i stopped paying per search

i didn't need a bigger model, i needed a smaller, dumber one

a spare 3060 and an evening with ollama

a spare gpu, a quantised model, and lower expectations

a spare gpu, ollama, and a model that runs on my own metal

the llm in my terminal is brilliant until it confidently isn't

my first rag pipeline retrieved confidently wrong things

my first rag pipeline retrieved confidently wrong answers

which local model actually helps me write code

semantic search for my notes, all on the laptop

running a 7b model on a laptop that should know better

running a 13b model on a laptop that cost less than the gpu i wanted

which local model is any good at code, actually

a small agent that does the boring bit for me

semantic search that never leaves the laptop

a language model on the gpu that was gathering dust

search that understands me, running on my own laptop

running code models on my own gpu, a fair-ish comparison

i wired an llm to my shell and it mostly behaved

a language model on the spare gpu, and what it's actually good for

running a model on the spare 3090, and what it actually costs

the spare gpu earns its keep

running a local llm on a spare gpu

putting a language model on the spare gpu in the cupboard

running a local llm on a spare gpu

putting a language model on the gpu that was just sitting there

a language model of my very own

a language model on the spare GPU, and what it was actually good for

getting gpt-j to talk on an old gpu

running a local llm on a spare gpu

a language model on the spare gpu in the cupboard

a spare gpu and a language model that runs in my own house

a language model on the gpu i forgot i had

getting gpt-j to talk on a spare gpu, and why it's harder than the demos suggest