– local llms 101 – tired of guides that just tell you to run a script and call it a day?
– want to actually know what your GPU is doing, not just trust a black box?
– here's what really happens when you run a local LLM
– what gets loaded, why, and how it all fits together
– no
LLMS
-
Local LLMs 101: Understanding GPU Processing and Architecture
By
–
-
Unified Architecture Different Parameters Activations
By
–
Unify in architecture, not in details. The parameters/activations will be different.
-
LLM Truthfulness: Preventing Deliberate Falsehoods to Humans
By
–
If an LLM is saying something to a human that it knows is false, this is very bad and is the top priority to fix. After that we can talk about when it's okay for an AI to keep quiet and say other things not meant to deceive. Then, discuss if the LLM is thinking false stuff.
-
X.ai LLM Trainers Competence Assessment
By
–
I am sure that http://
x.ai has more competent LLM trainers than me -
Missing GPT-1 Base Baseline in Comparison Analysis
By
–
Ok, so my criticism here is not the scores. But the right baseline is missing.
Should have a "GPT-1 base" column. -
Long Context Problems Require Solutions Beyond Long Context Models
By
–
long context problems arent solved by long context model agreed
-
Model-Tool Cotraining Achieves Better Results Than Larger Models
By
–
err i commented too quickly and people might get overexcited. i was just saying model-tool specific cotraining is achieves same or better results than using a 10-100x larger model without cotraining. this research doesn’t touch on memory/context compression, which is yet another
-
LoRA Matches Full Fine-Tuning Performance With Proper Implementation
By
–
LoRA Without Regret – Recent Blog from Thinking Machines TL/DR: LoRA actually matches full supervised fine-tuning(SFT) when you get the details right. Nearly same sample efficiency, loss(or better), same final performance. Some plain points:
– Apply LoRA to ALL layers, -
Learning LLM Attention Mechanisms: From Zero to Expert
By
–
– you are
– a random CS grad with 0 clue how LLMs work
– get tired of people gatekeeping with big words and tiny GPUs
– decide to go full monk mode
– 2 years later i can explain attention mechanisms at parties and ruin them – here’s the forbidden knowledge map
– top to bottom, -
Claude’s XML tag obsession revealed
By
–
First discovery: they're obsessed with XML tags. Not markdown. Not JSON formatting. XML. Why? Because Claude was trained to recognize structure through tags, not just content. Look at how Anthropic writes prompts vs how everyone else does it: Everyone else: You are a legal