File size: 10,763 Bytes
914512c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
"""bioai.agent.fireworks_client -- thin real Fireworks AI API client.

Uses ``requests`` only (no fireworks-ai SDK dependency -- keeps the install
small and the code portable to the ROCm container). Caches every response to
``.fireworks_cache/`` under the project root, keyed by ``sha256(prompt)`` with a
24-hour TTL so demo runs don't re-spend credits when the prompts are
identical.

If ``FIREWORKS_API_KEY`` is not set in the environment, the constructor
raises a clear ``RuntimeError("Set FIREWORKS_API_KEY env var")`` -- the
orchestrator catches this and falls back to a degraded-mode response so
the demo still runs end-to-end without an API key.
"""

from __future__ import annotations

import hashlib
import json
import os
import time
from pathlib import Path
from typing import Dict, List, Optional

import requests

# --------------------------------------------------------------------------- #
# Constants
# --------------------------------------------------------------------------- #
FIREWORKS_ENDPOINT = "https://api.fireworks.ai/inference/v1/chat/completions"
DEFAULT_MODEL = "accounts/fireworks/models/llama-v3p1-70b-instruct"
CACHE_DIR = Path(__file__).resolve().parents[2] / ".fireworks_cache"
CACHE_TTL_SECONDS = 24 * 60 * 60  # 24 hours


# --------------------------------------------------------------------------- #
# FireworksClient
# --------------------------------------------------------------------------- #
class FireworksClient:
    """Real Fireworks AI chat-completions client with disk caching.

    Parameters
    ----------
    model:
        Fireworks model id. Defaults to Llama-3.1-70B-Instruct.
    api_key:
        Optional explicit API key. If ``None``, reads ``FIREWORKS_API_KEY``
        from the environment and raises if missing.
    cache_dir:
        Where to store cached responses.
    cache_ttl:
        Cache time-to-live in seconds (default 24 hours).
    timeout:
        HTTP timeout per request, in seconds.
    """

    def __init__(
        self,
        model: Optional[str] = None,
        api_key: Optional[str] = None,
        cache_dir: Path | str = CACHE_DIR,
        cache_ttl: int = CACHE_TTL_SECONDS,
        timeout: int = 60,
    ):
        self.model = model or DEFAULT_MODEL
        self.api_key = api_key or os.environ.get("FIREWORKS_API_KEY")
        if not self.api_key:
            raise RuntimeError("Set FIREWORKS_API_KEY env var")
        self.cache_dir = Path(cache_dir)
        self.cache_dir.mkdir(parents=True, exist_ok=True)
        self.cache_ttl = cache_ttl
        self.timeout = timeout
        # Reuse a session for connection pooling across calls.
        self._session = requests.Session()
        self._session.headers.update({
            "Authorization": f"Bearer {self.api_key}",
            "Content-Type": "application/json",
        })

    # ------------------------------------------------------------------ #
    # Low-level chat
    # ------------------------------------------------------------------ #
    def chat(
        self,
        messages: List[Dict[str, str]],
        temperature: float = 0.7,
        max_tokens: int = 2048,
    ) -> str:
        """Send a chat-completions request. Returns the assistant message text.

        ``messages`` is the standard OpenAI-style list of
        ``{"role": ..., "content": ...}`` dicts.
        """
        cache_key = self._cache_key(messages, temperature, max_tokens)
        cached = self._cache_get(cache_key)
        if cached is not None:
            print(
                f"[fireworks] CACHE HIT  model={self.model} "
                f"prompt_len={sum(len(m['content']) for m in messages)} "
                f"response_len={len(cached)}"
            )
            return cached

        payload = {
            "model": self.model,
            "messages": messages,
            "temperature": temperature,
            "max_tokens": max_tokens,
        }
        prompt_len = sum(len(m["content"]) for m in messages)
        print(f"[fireworks] API CALL  model={self.model} prompt_len={prompt_len}")
        t0 = time.time()
        resp = self._session.post(
            FIREWORKS_ENDPOINT, data=json.dumps(payload), timeout=self.timeout
        )
        dt = time.time() - t0
        if resp.status_code != 200:
            # Surface the error body so the caller can log it.
            raise RuntimeError(
                f"Fireworks API returned {resp.status_code}: {resp.text[:500]}"
            )
        data = resp.json()
        text = (
            data.get("choices", [{}])[0]
            .get("message", {})
            .get("content", "")
        )
        # Cache and log
        self._cache_set(cache_key, text)
        print(
            f"[fireworks] API OK   model={self.model} "
            f"response_len={len(text)} elapsed={dt:.2f}s cached=False"
        )
        return text

    # ------------------------------------------------------------------ #
    # High-level helpers
    # ------------------------------------------------------------------ #
    def parse_pest_report(self, user_text: str) -> Dict:
        """Extract pest species, crop, severity, location from free text.

        Returns a dict with keys ``pest_species``, ``crop``, ``severity``,
        ``location``. On parse failure, returns a dict with ``_raw`` set to
        the raw model output and best-effort defaults.
        """
        system_prompt = (
            "You are an agricultural pest identification assistant. "
            "Extract structured data from the user's pest report and return "
            "STRICT JSON ONLY (no markdown fences, no commentary) with keys: "
            '"pest_species" (string), "crop" (string), "severity" (one of '
            '"low","moderate","high","severe"), "location" (string), '
            '"notes" (string, optional). If a field is unknown, use null.'
        )
        messages = [
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": user_text},
        ]
        raw = self.chat(messages, temperature=0.1, max_tokens=512)
        try:
            parsed = json.loads(raw)
        except json.JSONDecodeError:
            # Try to find a JSON block in the response.
            import re
            m = re.search(r"\{.*\}", raw, re.DOTALL)
            if m:
                try:
                    parsed = json.loads(m.group(0))
                except json.JSONDecodeError:
                    parsed = {}
            else:
                parsed = {}
        # Ensure all expected keys exist
        for key in ("pest_species", "crop", "severity", "location"):
            parsed.setdefault(key, None)
        parsed["_raw"] = raw
        return parsed

    def generate_safety_card(
        self,
        sirna_seq: str,
        offtarget_risks: Dict[str, float],
        half_life_hours: float,
    ) -> str:
        """Generate a markdown safety card for one siRNA candidate."""
        ot_str = "\n".join(
            f"  - {sp}: {risk:.3f}" for sp, risk in offtarget_risks.items()
        ) or "  (no off-target hits detected)"
        system_prompt = (
            "You are a regulatory toxicology writer. Produce a concise "
            "markdown SAFETY CARD for a dsRNA-based biopesticide siRNA "
            "candidate. Use only the data provided. Do not invent numbers. "
            "Sections: Sequence, Off-Target Profile, Environmental Fate, "
            "Overall Risk Tier (low/moderate/high). Keep it under 200 words."
        )
        user_prompt = (
            f"siRNA sequence (21 nt): {sirna_seq}\n\n"
            f"Off-target risks per species (0..1, fraction of 21-mers hit):\n{ot_str}\n\n"
            f"Predicted environmental half-life: {half_life_hours:.2f} hours\n"
        )
        messages = [
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": user_prompt},
        ]
        return self.chat(messages, temperature=0.3, max_tokens=800)

    def generate_regulatory_memo(
        self,
        pest_species: str,
        candidates: List[Dict],
    ) -> str:
        """Generate an EPA-style regulatory memo summarising the top candidates.

        Each candidate dict should contain at least ``sirna_seq``,
        ``efficacy``, ``offtarget_max``, ``half_life_hours``, ``final_score``.
        """
        cand_lines = []
        for i, c in enumerate(candidates, start=1):
            cand_lines.append(
                f"  {i}. {c.get('sirna_seq', '?')}  "
                f"efficacy={c.get('efficacy', 0):.3f}  "
                f"offtarget_max={c.get('offtarget_max', 0):.3f}  "
                f"half_life={c.get('half_life_hours', 0):.2f}h  "
                f"score={c.get('final_score', 0):.3f}"
            )
        cand_block = "\n".join(cand_lines) or "  (no candidates provided)"

        system_prompt = (
            "You are an EPA FIFRA regulatory affairs consultant. Produce a "
            "concise markdown MEMO (under 400 words) recommending whether "
            "the listed dsRNA biopesticide candidates are suitable for an "
            "experimental use permit against the named pest. Sections: "
            "Pest & Crop, Candidate Summary, Risk Assessment, Recommendation. "
            "Be conservative; if any candidate has high off-target risk or "
            "very short half-life, flag it."
        )
        user_prompt = (
            f"Pest species: {pest_species}\n\n"
            f"Top candidates (sorted by final_score):\n{cand_block}\n"
        )
        messages = [
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": user_prompt},
        ]
        return self.chat(messages, temperature=0.3, max_tokens=1500)

    # ------------------------------------------------------------------ #
    # Cache helpers
    # ------------------------------------------------------------------ #
    def _cache_key(
        self,
        messages: List[Dict[str, str]],
        temperature: float,
        max_tokens: int,
    ) -> str:
        blob = json.dumps(
            {"model": self.model, "messages": messages,
             "temperature": temperature, "max_tokens": max_tokens},
            sort_keys=True,
        )
        return hashlib.sha256(blob.encode("utf-8")).hexdigest()

    def _cache_get(self, key: str) -> Optional[str]:
        path = self.cache_dir / f"{key}.txt"
        if not path.exists():
            return None
        age = time.time() - path.stat().st_mtime
        if age > self.cache_ttl:
            return None
        return path.read_text(encoding="utf-8")

    def _cache_set(self, key: str, value: str) -> None:
        path = self.cache_dir / f"{key}.txt"
        path.write_text(value, encoding="utf-8")