Daily reminder that GPT-4.5 is, in fact, really good!
@aymericroucher
-
Debating async vs thread for linear agent run
By
–
Haha we were discussing async this morning: the conclusion was since an agent's run is linear (chaining operation), it could just be put in a thread. Do you have use cases where async would be better?
-
smolagents reaches 15k GitHub stars, fastest-growing agentic framework
By
–
smolagents has reached 15k GitHub stars : it's one of the fastest-growing agentic frameworks ever Don't hesitate to try it out, we working with @avillanovamoral to add nice stuff like sandboxed code execution via E2B or Docker!
-
smolagents now supports code execution in e2b sandboxes
By
–
smolagents has code execution in @e2b sandboxes Docs here:
-
Gemini 2.0 Flash creates a very French version of Starship launch
By
–
"Gemini 2.0 Flash, here's a picture from the launch of Starship. Make a French version, the frenchiest version you can possibly french" It went quite well
-
Gemini 2.0 creates toy car blueprint from photo, measurements made up
By
–
Me: "Here's a car photo, please help me make a blueprint to print a toy model for my nephew". Gemini 2.0 Flash Experimental: "sure, here's your blueprint"
Measurements are made-up but this drawing precision -
ToolCallingAgent requires tool call; vLLM must integrate
By
–
More context: in ToolCallingAgent, smolagents always needs a tool call, so removing the necessity of calling a tool would just break the lib's logic. Indeed this is on vLLM's side to integrate!
-
smolagents 1.11.0 adds VLLMModel class for local inference
By
–
smolagents now support vLLM! As one of the most popular local inference solutions, the community had been asking us to integrate vLLM: after a heavy refactoring of our LLM classes, we've just released smolagents 1.11.0, with a brand new VLLMModel class. Go try it and tell us
-
New Agentic Leaderboard Ranks LLMs for Agents
By
–
Our new Agentic leaderboard is now live! I've long wanted a way to quickly know which LLM is best for powering agents. So we've just built a leaderboard with Albert Villanova! This ranks LLMs powering a smolagents CodeAgent on subsets of various benchmarks. GPT-4.5
-
Safe Execution of LLM-Generated Code by Agents
By
–
Agents executing LLM-generated code safely, it's possible!