danucalovj comments

Results 9 comments of


                                            danucalovj

Examples?

Would recommend Titanium Web Proxy for that. This repository doesn't seem to be updated often.

Both export_*_checkpoint.py not reflecting base and lora weights merge upon inference

Possibly related to the newest peft release. I had a similar issue (see other issues posted here around peft breaking things left and right). Fix: https://github.com/tloen/alpaca-lora/issues/293: pip uninstall peft -y...

cannot run example

You're getting a 401 error on the huggingface model and dataset. Try downloading the model locally into your working directory using: git lfs install git clone https://huggingface.co/decapoda-research/llama-7b-hf And also the...

When I click submit, the inference does not work and an error is displayed.

Assuming you're using GPU (cuda), trying changing/adding the value of device_map in the following functions of generate.py: ```if device == "cuda": model = LlamaForCausalLM.from_pretrained( ... device_map={'':0}, ... ) model =...

Finetuning saves a 443 byte adapter_model.bin

See: https://github.com/tloen/alpaca-lora/issues/293

Inference using my fine-tuned weights not @tloen fined tuned lora weights on Hugging Face.

python generate.py \ --load_8bit \ --base_model 'decapoda-research/llama-7b-hf' \ --lora_weights './lora-alpaca'

Question 2

Same here :(

Question 2

@kartset try this: Change this: `model.compile(loss='binary_crossentropy', optimizer='adam', metrics = ['accuracy'])` To this: `model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics = ['accuracy'])` Not sure of results yet, running epochs right now, at least got it...

Question 2

Nevermind. That didn't work.