get_topic.py

import json

with open("issues_dict.json", "r") as f:
    issues = json.load(f)

topic_maintainers_map ={
    "text models": ["@ArthurZucker", "@younesbelkada"],
    "vision models": "@amyeroberts",
    "speech models": "@sanchit-gandhi",
    "graph models": "@clefourrier",
    "flax": "@sanchit-gandhi",
    "generate": "@gante",
    "pipelines": "@Narsil",
    "tensorflow": ["@gante", "@Rocketknight1"],
    "tokenizers": "@ArthurZucker",
    "trainer": ["@muellerzr", "@pacman100"],
    "deepspeed": "@pacman100",
    "ray/raytune": ["@richardliaw", "@amogkam"],
    "Big Model Inference": "@SunMarc",
    "quantization (bitsandbytes, autogpt)": ["@SunMarc", "@younesbelkada"],
    "Documentation": ["@stevhliu", "@MKhalusova"],
    "accelerate": "different repo",
    "datasets": "different repo",
    "diffusers": "different repo",
    "rust tokenizers": "different repo",
    "Flax examples": "@sanchit-gandhi",
    "PyTorch vision examples": "@amyeroberts",
    "PyTorch text examples": "@ArthurZucker",
    "PyTorch speech examples": "@sanchit-gandhi",
    "PyTorch generate examples": "@gante",
    "TensorFlow": "@Rocketknight1",
    "Research projects and examples": "not maintained",
}


issue_no = 2781
issue = issues[str(issue_no)]


from transformers import AutoTokenizer, LlamaForCausalLM

model = LlamaForCausalLM.from_pretrained("meta-llama/Llama-2-7b-hf")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")

# prompt = f"Which of the following topics {list(topic_maintainers_map.keys())} is this issue about:\n{issue['body']}"
prompt = f"QUESTION: What is the provided issue about? Pick up to 3 topics from the following list: {list(topic_maintainers_map.keys())} \nISSUE START:\n{issue['body']} \n ISSUE END. \n ANSWER:"
inputs = tokenizer(prompt, return_tensors="pt")

prefix_len = inputs.input_ids.shape[1]

# Generate
generate_ids = model.generate(inputs.input_ids, max_length=30 + prefix_len)
outputs = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
print(outputs[prefix_len:])

print("TITLE", issue["number"] + " " + issue["title"])