tinyllama-mixpretrain-uniprottune

tinyllama-mixpretrain-uniprottune

This is an adapter of the monsoon-nlp/tinyllama-mixpretrain-quinoa-sciphi model on the GreenBeing dataset finetuning split (minus maize/corn/Zea, which I left for evaluation).

Usage

from peft import AutoPeftModelForCausalLM
from transformers import AutoTokenizer

# this model
model = AutoPeftModelForCausalLM.from_pretrained("monsoon-nlp/tinyllama-mixpretrain-uniprottune").to("cuda")
# base model for the tokenizer
tokenizer = AutoTokenizer.from_pretrained("monsoon-nlp/tinyllama-mixpretrain-quinoa-sciphi")

inputs = tokenizer("<sequence> Subcellular locations:", return_tensors="pt")
outputs = model.generate(input_ids=inputs["input_ids"].to("cuda"), max_new_tokens=50)
print(tokenizer.batch_decode(outputs, skip_special_tokens=True)[0])

Inference Notebook: https://colab.research.google.com/drive/1UTavcVpqWkp4C_GkkS_HxDQ0Orpw43iu?usp=sharing

It seems unreliable on the Zea proteins. Getting a lot of the same answers for Subcellular locations.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 20
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 10
• num_epochs: 1

Framework versions

• PEFT 0.10.0
• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.19.0
• Tokenizers 0.15.2