It's just training a very small LoRA you can merge back into the base model, so it should retain the context window!
MACHINE LEARNING
-
Writing RL Agent from Scratch Without Prior Code
By
–
It was a fun exercise recently to just open up a completely blank file and write an RL agent from scratch, without looking at any of my prior code. There is a point of scale where rewriting things from scratch is a bad idea, but it is a blessing when you can! By “from scratch” I
-
Researchers solve hardest problems so engineers focus on implementation
By
–
Researchers solve the hardest problems so engineers don't have to.
-
Simulation: The Key to Rich, Diverse AI Training Data
By
–
But where does rich, diverse data come from —especially the rare, unpredictable data? That’s where simulation comes in.
-
Simulation trains autonomous vehicles for rare safety scenarios
By
–
Simulation and synthetic data help AVs train for rare events—like a helicopter landing on the road. Safety starts in a virtual space. Porsche uses high-fidelity simulation to test what’s too rare or risky to capture in the real world.
-
Data-Centric Approach Revolutionizes Autonomous Vehicle Safety
By
–
Why the shift?
Because real-world driving is too complex to pre-program. Data is the new engine. Billions of kilometers + labeled data + edge cases = safer autonomous vehicles. “We’ve moved away from the rules and handwritten components… to a data-centric approach,” Brian -
Automakers Shift from Rule-Based to Data-Driven Autonomous Systems
By
–
How are automakers innovating today? Automakers are shifting from rule-based data-driven autonomous systems.
-
How Close Are Cars to Human-Like AI Thinking?
By
–
How close are we to cars that think, reason, and communicate like humans? NVIDIA AI Podcast's new episode breakdown
-
Lead Author Discusses Research Paper on X Platform
By
–
I didn't realize the lead author was also on X, here is his thread on the paper (and links): https://
x.com/mpshanahan/sta
tus/1950587921666281595
… -
8B Model Multi-Turn Tool Calling Capabilities
By
–
I'd really love to see an 8B model with reliable multi-turn tool calling, if one exists I've not experienced it yet