Fine Tuning
178 episodes · Page 3 of 8
#6052: What's Actually in the Hugging Face ASR Catalog?
Four real architectural bets sit underneath thousands of Hugging Face ASR repos. Here's how to tell the bets from the repackagings.
#6039: Who's Actually Building Frontier AI Outside the US and China?
Mistral sits 20 points off the frontier. Korea runs a bake-off. The UAE open-sources everything. What's actually driving sovereign AI?
#6032: Blending Datasets vs Merging Weights for Fine-Tuning
Merging weights happens in parameter space. Blending datasets happens in data space — and the messy part is worth it.
#6029: One Model, Two Modes: Does AI Consolidation Hold Up?
OpenAI hides a router, DeepSeek merged two modes, Anthropic removed the knob. Is the model landscape really consolidating?
#6027: Can You Snap Two Fine-Tunes Together Like LEGO?
Two fine-tunes, one base model, and a question from Jerusalem: can you merge Hebrew code-switching with technical vocab — without retraining?
#6023: How Much Voice Data Is Enough?
Two hours of phone recordings took Hebrew code-switching from 20% to 60%. But when does more data stop mattering?
#6022: Fine-Tuning for Code-Switched Hebrew on 2 Hours of Audio
Two hours of one man's voice beat models trained on 120,000 hours — but only for code-switched Hebrew.
#6016: What 70 Billion Parameters Actually Means
A single artificial neuron, built from scratch — then the hard part: does that neuron analogy still describe a real transformer?
#6003: What Enterprises Actually Pay for AI Inference
Twelve and a half billion dollars in enterprise API spend — and nobody can tell you who spent it. Inside AI's most undocumented market.
#5994: Hebrew in Latin Letters: A TTS Pipeline Fix
A token classifier that finds Hebrew words written in Latin letters so TTS stops mangling them — built for under $3.
#5995: NVIDIA ASR: Parakeet, Canary, and Nemotron Explained
Whisper gives you one dial. NVIDIA gives you five. A practical map of the Parakeet, Canary, and Nemotron speech models.
#5975: Merging vs Distillation: Can Two Models Become One?
Merging and distillation get lumped together, but they're opposites. One composes existing capability; the other has to teach it.
#5945: How a Reward Model Actually Trains an LLM
One number, no gradients, four models in the room. What a reward model actually does inside the RLHF loop.
#5944: Keeping the Rollout Engine Fed: Inside RLHF Pipelines
Most of an RLHF run is spent waiting on generation, not gradients. Here's how TRL, OpenRLHF, and verl actually differ.
#5937: The Fine-Tuning Career: Titles, Skills, and What It Pays
The job above the labeling layer exists — it just doesn't have one name. What it takes, what it pays, and how to sell it.
#5936: Training a Model When You Don't Know the Labels Yet
Daniel wants to extract adapter specs from his shelf — but he can't name the labels until he's already annotating.
#5922: Teaching a Model Your Voice: SFT vs DPO
Three ways to teach a model your writing voice — and the evidence says one of them quietly makes things worse.
#5873: Abliteration: How $5 Undoes a Model's Guardrails
A rank-one weight edit costs under $5 and strips refusal behavior from open models. Then an automated tool beat human experts at it.
#5868: Fine-Tuning Parakeet for Um-Free, Hebrish Speech
Parakeet runs great on a phone — but fine-tuning it happens on a GPU. Here's the actual recipe for dropping ums and catching Hebrew.
#5860: Who Can Actually Train an LLM From Scratch?
Only a handful of organizations can truly train a model from random weights. Here's why nobody can count them.
#5853: Who Actually Competes With Hugging Face?
2.96 million model repos, and 85.6% have under 200 downloads. What ModelScope, Kaggle, and Ollama actually do differently.
#5830: Serving Your Own Fine-Tuned Model in the Cloud
You fine-tuned an open-weight model. Now how does anyone actually talk to it? Dedicated GPUs vs serverless inference, and the math that decides it.
#5805: Freeze the Model, Move the Labels
Your classifier isn't stale — your label set moved. Why retraining is cheap, and why not retraining might be smarter.
#5624: One Forward Pass: Building a Bounded Classifier
No generation, no parsing — just label scores in a single pass. How to build a small encoder classifier that actually works.