Why did you use TRL 0.24.0?
#1
by PhilDevAi - opened
Why did you use trl version 0.24.0 instead of newer one, like 1.0.0?
Did you find to be more compatible with Gemma4? I was having a hard time with 1.0.0
yes watching "hidden feature of gemma4" youtube video by anythingllm creator, liteRT with google has multi-token processing, might be better than llama.cpp, a custom version i want to test out next but claude limits...xD Track 1: Clone TheTom/turboquant_plus, build with CUDA, run llama-server -m gemma4-legal.gguf -ctk turbo3 -ctv turbo3 --port 8090
Track 2: pip install litert-lm, download E2B model, run litert-lm