Yesterday, I developed a model merging program. This time I have no enough gpu memory to merge the models in only one time, so I tried to merge layer by layer. I found the memory of GPU is easily to release but CPU didn't.
Read More »[PyTorch] Release GPU / CPU Memory After Delete Model
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
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 Last night I tried to improve some code about merge two models. Because my poor device, I cannot merge all layers totally, but need to merge them layer by layer for reducing my memory cost.
Read More »[PyTorch] Delete Model And Free Memory (GPU / CPU)
Although I reach AI and research it all day, but there are too many new architectures, breakout, theories... many things I just know a rough concepts and put it away. The almost deeper understand need to until I must to be adjust the model architecture.
Read More »[Machine Learning] A LLM (Large Language Model) Visualization Website In early 2023, The pytorch 2.0 version added a new function that is torch.compile(), It cloud be accelerate the speed of model training or inference. By using precision training with compiling, it always make my training time reduce to half.
When using the transformers package, often we receive the following error message when we utilize different task models with various heads like AutoModelForSequenceClassification, AutoModelForSeq2SeqLM...
Read More »[Solved] Some weights of the model checkpoint at distilbert-base-multilingual-cased were not used when initializing DistilBertForSequenceClassification: ['vocab_projector.bias', 'vocab_layer_norm.bias', 'vocab_layer_norm.weight', 'vocab_transform.weight', 'vocab_transform.bias']
Moderation endpoint is a free tool that developed by OpenAI. It is design for detecting "Not Mode-rational Content" or "Sensitive Content". The more detailed list you can refer https://openai.com/policies/usage-policies.
Read More »Using OpenAI Moderation Endpoint To Dectect Not Moderation Content
Today when I training the model on my server, I write an GPU parallel training script, and input the newest training data; but in the process of training, I got an error message about "GPU not found". After I using torch.cuda.is_available() to check my device, I got another error message as following:
