import streamlit as st from transformers import AutoTokenizer, AutoModelForCausalLM from transformers import pipeline import torch import json import pandas as pd import requests GITHUB_CODE = "https://huggingface.co/datasets/lvwerra/github-code" INCODER_IMG = "https://huggingface.co/datasets/loubnabnl/repo-images/raw/main/incoder.png" @st.cache(allow_output_mutation=True) def load_tokenizer(model_ckpt): return AutoTokenizer.from_pretrained(model_ckpt) @st.cache(allow_output_mutation=True) def load_model(model_ckpt): model = AutoModelForCausalLM.from_pretrained(model_ckpt, low_cpu_mem_usage=True) return model @st.cache() def load_examples(): with open("utils/examples.json", "r") as f: examples = json.load(f) return examples st.set_page_config(page_icon=":laptop:", layout="wide") st.sidebar.header("Models") models = ["CodeParrot", "InCoder"] selected_models = st.sidebar.multiselect("Select code generation models to compare", models, default=["CodeParrot"]) st.sidebar.header("Tasks") tasks = [" ", "Pretraining datasets", "Model architecture", "Model evaluation", "Code generation"] selected_task = st.sidebar.selectbox("Select a task", tasks) if selected_task == " ": st.title("Code Generation Models") with open("utils/intro.txt", "r") as f: intro = f.read() st.markdown(intro) elif selected_task == "Pretraining datasets": st.title("Pretraining datasets 📚") st.markdown(f"Preview of some code files from Github repositories in [Github-code dataset]({GITHUB_CODE}):") df = pd.read_csv("utils/data_preview.csv") st.dataframe(df) for model in selected_models: with open(f"datasets/{model.lower()}.txt", "r") as f: text = f.read() st.markdown(f"### {model}") st.markdown(text) elif selected_task == "Model architecture": st.title("Model architecture 🔨") for model in selected_models: with open(f"architectures/{model.lower()}.txt", "r") as f: text = f.read() st.markdown(f"## {model}") st.markdown(text) if model == "InCoder": st.image(INCODER_IMG, caption="Figure 1: InCoder training", width=700) elif selected_task == "Model evaluation": st.title("Code models evaluation 📊") with open("evaluation/intro.txt", "r") as f: intro = f.read() st.markdown(intro) elif selected_task == "Code generation": st.title("Code generation 💻") st.sidebar.header("Examples") examples = load_examples() example_names = [example["name"] for example in examples] name2id = dict([(name, i) for i, name in enumerate(example_names)]) selected_example = st.sidebar.selectbox("Select one of the following examples or implement yours", example_names) example_text = examples[name2id[selected_example]]["value"] default_length = examples[name2id[selected_example]]["length"] st.sidebar.header("Generation settings") temperature = st.sidebar.slider("Temperature:", value=0.2, min_value=0.0, step=0.1, max_value=2.0) max_new_tokens = st.sidebar.slider("Number of tokens to generate:", value=default_length, min_value=8, step=8, max_value=256) seed = st.sidebar.slider("Random seed:", value=42, min_value=0, step=1, max_value=1000) gen_prompt = st.text_area("Generate code with prompt:", value=example_text, height=220,).strip() if st.button("Generate code!"): with st.spinner("Generating code..."): for model in selected_models: url = f'https://hf.space/embed/loubnabnl/{model.lower()}-subspace/+/api/predict/' r = requests.post(url=url, json={"data": [gen_prompt, max_new_tokens, temperature, seed]}) generated_text = r.json()['data'][0] st.markdown(f"{model}") st.code(generated_text)