Today I was reading the training source code of DreamBooth, and I found the tempfile built-in module; I happened to reconstruct a script for model's layer merging, suddenly thought that the module would make the code more elegant, so I made this note.
DPO (Direct Preference Optimization) is a fine-tuning method that want to replace RLHF (Reinforcement Learning from Human Feedback).
Read More »Direct Preference Optimization (DPO) Training Note
HuggingFace has published an article stating that the current LLM is best trained according to the ChatML format. In normal case, it will be generated according to three different roles of system, user and assistant. The format is as follows:
Read More »[Solved] Mistral Cannot Generate eos_token `<|im_end|>` After SFTTrainer Fine-tuning Today I installed pyaudio on my new Linux laptop (a python package for sound recording), I encountered an error I never met:
Read More »[Solved] fatal error: portaudio.h: No such file or directory 9 | #include "portaudio.h" | ^~~~~~~~~~~~~ compilation terminated IPython is a system that provided a interactive interface, it could be integrated with shells or graphical interface; for example we can use ipython (if you had been installed it), or you can use some GUI editor/IDE such as VS Code and PyCharm.
Design a food rating system that can do the following:
Given two strings s and t, return true if t is an anagram of s, and false otherwise.
You are given the array paths, where paths[i] = [cityAi, cityBi] means there exists a direct path going from cityAi to cityBi. Return the destination city, that is, the city without any path outgoing to another city.
vLLM is a large language model (LLM) acceleration framework that developed by the research team of University of California, Berkeley. It is used PagedAttention to increase the usage rate of GPU VRAM, and this method do not change the model architecture.
Read More »Using vLLM To Accelerate The Decoding Of Large Language Model
