lamini/instruct-peft-tuned-12b

---

license: cc-by-4.0

Lamini’ instruct-peft-tuned-12b, an instruction-following large language model trained on the Lamini machine learning platform. The dataset includes 71k instruction fined-tuned samples, created by using Lamini's data generator pipeline, which you can find it here: https://github.com/lamini-ai/lamini. This model is trained using Low-Rank Adaptation of Large Language Models (LoRA) technique which reduces computational complexity and improve efficiency. This integration allows users to train larger models more efficiently, consuming fewer time and hardware resources. We have used our custom dataset and fine-tuned it on top of EleutherAI's pythia-12b-deduped model to create instruct-peft-tuned-12b model.

You can try this model using Lamini's Hugging Face spaces hosted here, https://huggingface.co/spaces/lamini/instruct-playground-12b

Also, Its super easy to integrate this model to your code pipeline using Lamini's library: (https://lamini-ai.github.io/)

• Install Lamini library:

  pip install lamini

• To generate LAMINI-KEY log-in to https://app.lamini.ai/ and then navigate to Account tab

• Below is a sample code:

  
  from llama import Type, Context, LLM
  class Question(Type):
      question: str = Context("a question")
  
  class Response(Type):
      response: str = Context("the response to the question")
  
  def lamini(input):
      llm=LLM(name="lamini-playground-instruct-12b", 
              config={
              "production":{
                  "key": [LAMINI-KEY]
              }
          })
      user_query_text=Question(question=input)
      result = llm(
              input=user_query_text,
              output_type=Response,
              model_name="lamini/instruct-peft-tuned-12b",
              is_peft_model=True
          )
      return result.response
  prompt=input("Enter your prompt: ")
  print(lamini(prompt))