LLMatic/mutation_models.py

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, CodeGenForCausalLM
#import openai
import os
import ray

from conf.config import Config

cfg = Config()


def codex_mutate(cfg: Config, prompt, model="code-davinci-002",temperature=0.5):
    
    os.environ['OPENAI_API_KEY'] = ""# ENETR OPENAI API KEY HERE
    openai.api_key = os.getenv("OPENAI_API_KEY")
    return openai.Completion.create(
                                    model=model,
                                    prompt=prompt,
                                    max_tokens=300,
                                    temperature=temperature,
                                    ) 


#['codegen-350M-multi', 'codegen-2B-multi', 'codegen-6B-multi', 'codegen-16B-multi', 'codegen-350M-mono', 'codegen-2B-mono', 'codegen-6B-mono', 'codegen-16B-mono']
#@ray.remote(num_cpus=cfg.NUM_CPUS)  # FIXME: Number of parallel processes should be configured globally.
@ray.remote(num_gpus=cfg.NUM_GPUS)
def codegen_mutate(cfg: Config, prompt, temperature):
    diff = False

    if diff:
        # model = "/mnt/lustre/users/mnasir/NAS-LLM/diff-codegen-6B"
        model = os.path.join(cfg.SAVE_DIR, "diff-codegen-6B")
        tokenizer = AutoTokenizer.from_pretrained(model)#.to(cfg.DEVICE)
        tokenizer.padding_side = 'left'
        tokenizer.pad_token = tokenizer.eos_token
        model = CodeGenForCausalLM.from_pretrained(model).to(cfg.DEVICE)
        inputs = tokenizer(prompt, return_tensors='pt', padding=True)#.to(device)
        #model.config.use_cache = True
        sample = model.generate(**inputs, temperature=0.5, max_length=len(inputs[0]) + 300)
        
        return tokenizer.decode(sample[0][len(inputs[0]):])
    else:
        model = f'Salesforce/{cfg.MUTATION}'
        # model = "/mnt/lustre/users/mnasir/NAS-LLM/codegen-6B"
        #model = "/mnt/lustre/users/mnasir/NAS-LLM/diff-codegen-6B"
        #model = os.path.join(cfg.SAVE_DIR, "codegen-6B")

        # TODO: Should be doing something like this to download the model automatically.
        # model = os.path.join(cfg.SAVE_DIR, cfg.MUTATION)
        # if not os.path.exists(model):
        #     # TODO: `Salesforce` part should not be hardcoded / should be configurable so that we can download models 
        #     # from other sources.
        #     model = f'Salesforce/{cfg.MUTATION}'

        tokenizer = AutoTokenizer.from_pretrained(model)
        model = AutoModelForCausalLM.from_pretrained(model).to(cfg.DEVICE)

        # TODO: Above lines may have downloaded a fresh model if it was not already present. Now, copy the model file 
        # to the desired location if necessary.

        inputs = tokenizer(prompt, return_tensors="pt").to(cfg.DEVICE)
        sample = model.generate(**inputs, max_length=350 + len(inputs[0]),temperature=temperature,num_beams=1,do_sample=True)
        return tokenizer.decode(sample[0][len(inputs[0]):], truncate_before_pattern=[r"\n\n^#", "^'''", "\n\n\n"])


def replace_word_mutation(sentence):
    if "Add" in sentence:
        return sentence.replace("Add", "Delete")
    elif "Delete" in sentence:
        return sentence.replace("Delete", "Add")
    else:
        return sentence