LLMs often return broken JSON, or valid JSON with the wrong answer. Here’s how we use a small model to fix the syntax without hiding bad reasoning.
#ai#smallmodels...Show more
Small models are the budget option? A fine-tuned 0.8B in production beat prompted GPT-5.4 on precision and recall at 16x lower serving cost.
The trick: a frontier model built the training data, then left the request path.
Full read on our LinkedIn.
#SmallModels#FineTuning
Shrink It, Train It Small, or Search for It: The Main Strategies for Small Models, Compared
A capable small model can be reached by compressing a large one, training a small one on purpose, searching…
Read more:
#SmallModels#ModelCompression
This is my average cost, not for an AI request. v4 pro and v4 flash (for auto context) (Rust @rustlang coding)
With Opus, GPT, or Flash, would have been ~0.40, 2m (2x faster, but 20x price).
Price/Limit no more
#BigApps#SmallModels
Microsoft’s new Phi-3.5-3.8B (Mini) beats LLaMA-3.1-8B, trained only on 3.4T tokens
Phi-3.5-16x3.8B (MoE) beats Gemini-Flash, trained only on 4.9T tokens
Phi-3.5-V-4.2B (Vision) beats GPT-4o, trained on 500B tokens.
AI efficiency game is 🔥 🚀 #SmallModels@Microsoft