import os
import tempfile
from pathlib import Path

"""
**Streamlit Space – minimal version (no explicit caching)**

Fixes the `StreamlitSetPageConfigMustBeFirstCommandError` **and** the
previous permission issues by:

1. Setting `HOME`, `HF_HOME`, and `HF_CACHE_DIR` **before** we import
   Streamlit – that way Streamlit creates `~/.streamlit` in a writable
   place instead of `/.streamlit`.
2. Moving the call to `st.set_page_config()` to be the **very first
   Streamlit command**.

You can re‑add `@st.cache_resource` once everything boots.
"""

# ----------------------------------------------------------------------------
# Writable environment — must be set *before* importing streamlit
# ----------------------------------------------------------------------------
HF_TMP = tempfile.mkdtemp(prefix="hf_")  # guaranteed‑writable temp dir
os.environ["HOME"] = HF_TMP  # for ~/.streamlit
os.environ["HF_HOME"] = HF_TMP
os.environ["HF_CACHE_DIR"] = HF_TMP

# ----------------------------------------------------------------------------
import streamlit as st
from llama_cpp import Llama

# ----- Streamlit page config (must be first Streamlit command) -------------
st.set_page_config(page_title="Sanskrit Transliteration", page_icon="📜")

# ----------------------------------------------------------------------------
# Model loader (no Streamlit cache for now)
# ----------------------------------------------------------------------------
llm = None  # global singleton

def get_model():
    global llm
    if llm is None:
        with st.spinner("Downloading & loading model – first run takes a bit…"):
            llm = Llama.from_pretrained(
                repo_id="13Aluminium/Sanskrit_transliteration_Qwen3",
                filename="unsloth.Q4_K_M.gguf",
                cache_dir=os.environ["HF_CACHE_DIR"],
            )
    return llm

# ----------------------------------------------------------------------------
# Inference helper
# ----------------------------------------------------------------------------

def transliterate(text: str, *, temperature: float = 0.1, max_tokens: int = 200) -> str:
    prompt = f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
Transliterate Sanskrit Shlok to Latin

### Input:
{text}

### Response:
"""
    model = get_model()
    out = model(prompt, max_tokens=max_tokens, temperature=temperature, echo=False)
    return out["choices"][0]["text"].strip()

# ----------------------------------------------------------------------------
# UI
# ----------------------------------------------------------------------------

st.title("📜 Sanskrit → Latin Transliteration (Qwen3)")

sample = (
    "भूरीणि भूरिकर्माणि श्रोतव्यानि विभागशः ।\n"
    "अतः साधोऽत्र यत्सारं समुद्धृत्य मनीषया ।\n"
    "ब्रूहि नः श्रद्दधानानां येनात्मा संप्रसीदति ॥ ११ ॥"
)

sanskrit_text = st.text_area("Sanskrit text:", value=sample, height=180)

temp = st.slider("Temperature", 0.0, 1.0, 0.1, 0.01)
max_tokens = st.number_input("Max tokens", 50, 512, 200, 1)

if st.button("🔤 Transliterate"):
    if not sanskrit_text.strip():
        st.warning("Please enter some Sanskrit text first.")
    else:
        with st.spinner("Transliterating…"):
            result = transliterate(sanskrit_text, temperature=temp, max_tokens=max_tokens)
        st.text(result)

# ----------------------------------------------------------------------------
# requirements.txt – keep the same:
# streamlit>=1.36
# llama-cpp-python>=0.2.11
# huggingface_hub>=0.24
# ----------------------------------------------------------------------------