app.py

import os
import streamlit as st
import pandas as pd
from datetime import datetime

from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from datasets import load_dataset
from huggingface_hub import login

# ––––– Auth –––––

HF_TOKEN = os.getenv(“HF_TOKEN”)
if HF_TOKEN:
login(token=HF_TOKEN)

# ––––– Page Config –––––

st.set_page_config(
page_title=“Code Assistant”,
page_icon=“🧠”,
layout=“wide”
)

# ––––– Session State –––––

if “history” not in st.session_state:
st.session_state.history = []

if “datasets” not in st.session_state:
st.session_state.datasets = {}

if “pipelines” not in st.session_state:
st.session_state.pipelines = {}

# ––––– Model Loader –––––

@st.cache_resource
def load_textgen_pipeline(model_id: str):
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
device_map=“auto”,
torch_dtype=“auto”
)
return pipeline(
“text-generation”,
model=model,
tokenizer=tokenizer,
max_new_tokens=256,
do_sample=True,
temperature=0.3
)

# ––––– Generation –––––

def generate_code(prompt, model_id, dataset_name=None):
if model_id not in st.session_state.pipelines:
st.session_state.pipelines[model_id] = load_textgen_pipeline(model_id)

```
pipe = st.session_state.pipelines[model_id]

context = ""
if dataset_name:
    dataset = st.session_state.datasets[dataset_name]
    context = f"\n# Dataset preview:\n{dataset[:3]}\n"

full_prompt = f"""You are a precise coding assistant.
```

Write clean, correct code.
{context}
Prompt:
{prompt}
“””

```
result = pipe(full_prompt)[0]["generated_text"]
return result
```

# ––––– Sidebar –––––

with st.sidebar:
st.header(“⚙️ Control Plane”)

```
model_id = st.selectbox(
    "Model",
    [
        "microsoft/phi-2",
        "codellama/CodeLlama-7b-hf",
        "bigcode/starcoder2-3b",
    ]
)

st.divider()

st.subheader("📦 Dataset")

dataset_source = st.radio(
    "Dataset source",
    ["None", "Hugging Face Hub", "Upload file"]
)

dataset_name = None

if dataset_source == "Hugging Face Hub":
    hf_dataset_id = st.text_input(
        "Dataset repo (e.g. squad, openwebtext)"
    )

    if st.button("Load dataset") and hf_dataset_id:
        ds = load_dataset(hf_dataset_id, split="train[:100]")
        st.session_state.datasets[hf_dataset_id] = ds
        dataset_name = hf_dataset_id
        st.success(f"Loaded {hf_dataset_id}")

elif dataset_source == "Upload file":
    uploaded = st.file_uploader("Upload CSV", type=["csv"])
    if uploaded:
        df = pd.read_csv(uploaded)
        st.session_state.datasets[uploaded.name] = df.to_dict(orient="records")
        dataset_name = uploaded.name
        st.success(f"Loaded {uploaded.name}")

if st.session_state.datasets:
    dataset_name = st.selectbox(
        "Active dataset",
        options=[None] + list(st.session_state.datasets.keys())
    )
```

# ––––– Main UI –––––

st.title(“🧠 Code Assistant”)
st.caption(“Transformers + Datasets + Hugging Face Hub”)

prompt = st.text_area(
“Coding prompt”,
height=150,
placeholder=“Generate a Python function that validates the dataset schema”
)

if st.button(“Generate”, type=“primary”):
with st.spinner(“Thinking…”):
output = generate_code(prompt, model_id, dataset_name)

```
st.session_state.history.append({
    "time": datetime.now().strftime("%H:%M:%S"),
    "model": model_id,
    "dataset": dataset_name,
    "prompt": prompt,
    "output": output
})
```

# ––––– Output –––––

for item in reversed(st.session_state.history):
label = f”[{item[‘time’]}] {item[‘model’]}”
if item[“dataset”]:
label += f” | {item[‘dataset’]}”

```
with st.expander(label):
    st.code(item["output"], language="python")
```