app.py

import os
import re
import json
from functools import lru_cache

import gradio as gr
import torch

# -------------------
# Writable caches for HF + Gradio (fixes PermissionError in Spaces)
# -------------------
os.environ.setdefault("HF_HOME", "/data/.cache/huggingface")
os.environ.setdefault("HF_HUB_CACHE", "/data/.cache/huggingface/hub")
os.environ.setdefault("GRADIO_TEMP_DIR", "/data/gradio")
os.environ.setdefault("GRADIO_CACHE_DIR", "/data/gradio")

for p in [
    "/data/.cache/huggingface/hub",
    "/data/gradio",
]:
    try:
        os.makedirs(p, exist_ok=True)
    except Exception:
        pass

# Timezone (Python 3.9+)
try:
    from zoneinfo import ZoneInfo
except Exception:
    ZoneInfo = None

# Cohere SDK (hosted path)
try:
    import cohere
    _HAS_COHERE = True
except Exception:
    _HAS_COHERE = False

from transformers import AutoTokenizer, AutoModelForCausalLM
from huggingface_hub import login

# -------------------
# NEW: Safety imports
# -------------------
from safety import safety_filter, refusal_reply

# -------------------
# NEW: Augmentation imports
# -------------------
from retriever import init_retriever, retrieve_context
from decision_math import compute_operational_numbers
from prompt_templates import build_system_preamble

# -------------------
# Config
# -------------------
MODEL_ID = os.getenv("MODEL_ID", "CohereLabs/c4ai-command-r7b-12-2024")
HF_TOKEN = os.getenv("HUGGINGFACE_HUB_TOKEN") or os.getenv("HF_TOKEN")
COHERE_API_KEY = os.getenv("COHERE_API_KEY")
USE_HOSTED_COHERE = bool(COHERE_API_KEY and _HAS_COHERE)

# -------------------
# Helpers
# -------------------
def pick_dtype_and_map():
    if torch.cuda.is_available():
        return torch.float16, "auto"
    if torch.backends.mps.is_available():
        return torch.float16, {"": "mps"}
    return torch.float32, "cpu"

def is_identity_query(message, history):
    patterns = [
        r"\bwho\s+are\s+you\b",
        r"\bwhat\s+are\s+you\b",
        r"\bwhat\s+is\s+your\s+name\b",
        r"\bwho\s+is\s+this\b",
        r"\bidentify\s+yourself\b",
        r"\btell\s+me\s+about\s+yourself\b",
        r"\bdescribe\s+yourself\b",
        r"\band\s+you\s*\?\b",
        r"\byour\s+name\b",
        r"\bwho\s+am\s+i\s+chatting\s+with\b"
    ]
    def match(t):
        return any(re.search(p, (t or "").strip().lower()) for p in patterns)
    if match(message):
        return True
    if history:
        last_user = history[-1][0] if isinstance(history[-1], (list, tuple)) else None
        if match(last_user):
            return True
    return False

def _iter_user_assistant(history):
    for item in (history or []):
        if isinstance(item, (list, tuple)):
            u = item[0] if len(item) > 0 else ""
            a = item[1] if len(item) > 1 else ""
            yield u, a

def _history_to_prompt(message, history):
    parts = []
    for u, a in _iter_user_assistant(history):
        if u:
            parts.append(f"User: {u}")
        if a:
            parts.append(f"Assistant: {a}")
    parts.append(f"User: {message}")
    parts.append("Assistant:")
    return "\n".join(parts)

# -------------------
# Cohere Hosted
# -------------------
_co_client = None
if USE_HOSTED_COHERE:
    _co_client = cohere.Client(api_key=COHERE_API_KEY)

def cohere_chat(message, history):
    try:
        prompt = _history_to_prompt(message, history)
        resp = _co_client.chat(
            model="command-r7b-12-2024",
            message=prompt,
            temperature=0.3,
            max_tokens=350,
        )
        if hasattr(resp, "text") and resp.text:
            return resp.text.strip()
        if hasattr(resp, "reply") and resp.reply:
            return resp.reply.strip()
        if hasattr(resp, "generations") and resp.generations:
            return resp.generations[0].text.strip()
        return "Sorry, I couldn't parse the response from Cohere."
    except Exception as e:
        return f"Error calling Cohere API: {e}"

# -------------------
# Local HF Model
# -------------------
@lru_cache(maxsize=1)
def load_local_model():
    if not HF_TOKEN:
        raise RuntimeError("HUGGINGFACE_HUB_TOKEN is not set.")
    login(token=HF_TOKEN, add_to_git_credential=False)
    dtype, device_map = pick_dtype_and_map()
    tok = AutoTokenizer.from_pretrained(
        MODEL_ID,
        token=HF_TOKEN,
        use_fast=True,
        model_max_length=4096,
        padding_side="left",
        trust_remote_code=True,
    )
    mdl = AutoModelForCausalLM.from_pretrained(
        MODEL_ID,
        token=HF_TOKEN,
        device_map=device_map,
        low_cpu_mem_usage=True,
        torch_dtype=dtype,
        trust_remote_code=True,
    )
    if mdl.config.eos_token_id is None and tok.eos_token_id is not None:
        mdl.config.eos_token_id = tok.eos_token_id
    return mdl, tok

def build_inputs(tokenizer, message, history):
    msgs = []
    for u, a in _iter_user_assistant(history):
        if u:
            msgs.append({"role": "user", "content": u})
        if a:
            msgs.append({"role": "assistant", "content": a})
    msgs.append({"role": "user", "content": message})
    return tokenizer.apply_chat_template(
        msgs, tokenize=True, add_generation_prompt=True, return_tensors="pt"
    )

def local_generate(model, tokenizer, input_ids, max_new_tokens=350):
    input_ids = input_ids.to(model.device)
    with torch.no_grad():
        out = model.generate(
            input_ids=input_ids,
            max_new_tokens=max_new_tokens,
            do_sample=True,
            temperature=0.3,
            top_p=0.9,
            repetition_penalty=1.15,
            pad_token_id=tokenizer.eos_token_id,
            eos_token_id=tokenizer.eos_token_id,
        )
    gen_only = out[0, input_ids.shape[-1]:]
    return tokenizer.decode(gen_only, skip_special_tokens=True).strip()

# -------------------
# Snapshot Loader
# -------------------
def _load_snapshot(path="snapshots/current.json"):
    try:
        with open(path, "r", encoding="utf-8") as f:
            return json.load(f)
    except Exception:
        return {
            "timestamp": None,
            "beds_total": 400,
            "staffed_ratio": 1.0,
            "occupied_pct": 0.97,
            "ed_census": 62,
            "ed_admits_waiting": 19,
            "avg_ed_wait_hours": 8,
            "discharge_ready_today": 11,
            "discharge_barriers": {"allied_health": 7, "placement": 4},
            "rn_shortfall": {"med_ward_A": 1, "med_ward_B": 1},
            "forecast_admits_next_24h": {"respiratory": 14, "other": 9},
            "isolation_needs_waiting": {"contact": 3, "airborne": 1},
            "telemetry_needed_waiting": 5
        }

# Init retriever once
init_retriever()

# -------------------
# Chat Function (with Augmentation + Safety)
# -------------------
def chat_fn(message, history, user_tz):
    try:
        # ---- INPUT SAFETY ----
        safe_in, blocked_in, reason_in = safety_filter(message, mode="input")
        if blocked_in:
            return refusal_reply(reason_in)

        # Identity short-circuit
        if is_identity_query(safe_in, history):
            return "I am ClarityOps, your strategic decision making AI partner."

        # --- Load snapshot + policies + numbers
        snapshot = _load_snapshot()
        policy_context = retrieve_context(
            "bed management huddle discharge acceleration bed leveling ambulance offload"
        )
        computed = compute_operational_numbers(snapshot)
        system_preamble = build_system_preamble(snapshot, policy_context, computed)

        # Augmented input
        augmented_user = (
            system_preamble
            + "\n\nUser question:\n"
            + safe_in
        )

        # ---- GENERATION ----
        if USE_HOSTED_COHERE:
            out = cohere_chat(augmented_user, history)
        else:
            model, tokenizer = load_local_model()
            inputs = build_inputs(tokenizer, augmented_user, history)
            out = local_generate(model, tokenizer, inputs, max_new_tokens=350)

        # Tidy echoes
        if isinstance(out, str):
            for tag in ("Assistant:", "System:", "User:"):
                if out.startswith(tag):
                    out = out[len(tag):].strip()

        # ---- OUTPUT SAFETY ----
        safe_out, blocked_out, reason_out = safety_filter(out, mode="output")
        if blocked_out:
            return refusal_reply(reason_out)
        return safe_out
    except Exception as e:
        return f"Error: {e}"

# -------------------
# Theme & CSS
# -------------------
theme = gr.themes.Soft(
    primary_hue="teal",
    neutral_hue="slate",
    radius_size=gr.themes.sizes.radius_lg,
)

custom_css = """
:root {
  --brand-bg: #e6f7f8;
  --brand-accent: #0d9488;
  --brand-text: #0f172a;
  --brand-text-light: #ffffff;
}

.gradio-container { background: var(--brand-bg); }

h1 {
  color: var(--brand-text);
  font-weight: 700;
  font-size: 28px !important;
}

.chatbot header, .chatbot .label, .chatbot .label-wrap, .chatbot .top, .chatbot .header, .chatbot > .wrap > header {
  display: none !important;
}

.message.user, .message.bot {
  background: var(--brand-accent) !important;
  color: var(--brand-text-light) !important;
  border-radius: 12px !important;
  padding: 8px 12px !important;
}

textarea, input, .gr-input { border-radius: 12px !important; }

.examples, .examples .grid {
  display: flex !important;
  justify-content: center !important;
  text-align: center !important;
}
"""

# -------------------
# UI
# -------------------
with gr.Blocks(theme=theme, css=custom_css) as demo:
    tz_box = gr.Textbox(visible=False)
    demo.load(
        lambda tz: tz,
        inputs=[tz_box],
        outputs=[tz_box],
        js="() => Intl.DateTimeFormat().resolvedOptions().timeZone",
    )

    hide_label_sink = gr.HTML(visible=False)
    demo.load(
        fn=lambda: "",
        inputs=None,
        outputs=hide_label_sink,
        js="""
        () => {
          const sel = [
            '.chatbot header',
            '.chatbot .label',
            '.chatbot .label-wrap',
            '.chatbot .top',
            '.chatbot .header',
            '.chatbot > .wrap > header'
          ];
          sel.forEach(s => document.querySelectorAll(s).forEach(el => el.style.display = 'none'));
          return "";
        }
        """,
    )

    gr.Markdown("# ClarityOps Augmented Decision AI")

    gr.ChatInterface(
        fn=chat_fn,
        type="messages",
        additional_inputs=[tz_box],
        chatbot=gr.Chatbot(
            label="",
            show_label=False,
            type="messages",
            height=700,
        ),
        examples=[
            ["What are the symptoms of hypertension?"],
            ["What are common drug interactions with aspirin?"],
            ["What are the warning signs of diabetes?"],
        ],
        cache_examples=False,
        submit_btn="Submit",
        retry_btn="Retry",
        clear_btn="Clear",
        undo_btn=None,
    )

if __name__ == "__main__":
    port = int(os.environ.get("PORT", "7860"))
    demo.launch(
        server_name="0.0.0.0",
        server_port=port,
        show_api=False,
        max_threads=8,
    )