File size: 10,451 Bytes
23cdeed 66ad25b b5e99b3 66ad25b 2fe3d02 66ad25b 2fe3d02 66ad25b b5e99b3 66ad25b a70f2fd 66ad25b a70f2fd 66ad25b b5e99b3 2fe3d02 23cdeed 2fe3d02 23cdeed 66ad25b 23cdeed 2fe3d02 23cdeed 66ad25b 23cdeed 66ad25b 23cdeed | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 | # -*- coding: utf-8 -*-
"""
pluto/modes.py — Real mode switching engine.
Groq primary:
- MODE_QUICK: llama-3.1-8b-instant (fast, lightweight)
- MODE_REASONING: llama-3.3-70b-versatile (deep, accurate)
- MODE_VISION: llama-3.1-8b-instant (text/doc understanding)
Mistral fallback (if Groq fails or no key):
- All modes: mistral-small-latest
Real switching = True because MODE_QUICK uses 8b and MODE_REASONING uses 70b.
"""
from __future__ import annotations
import os
from dataclasses import dataclass
from dotenv import load_dotenv
load_dotenv()
def _clean_api_key(api_key: str | None) -> str:
cleaned = str(api_key or "").strip().strip('"').strip("'")
if cleaned.lower().startswith("bearer "):
cleaned = cleaned[7:].strip()
return cleaned
def _looks_like_nvidia_key(api_key: str) -> bool:
return _clean_api_key(api_key).startswith("nvapi-")
@dataclass(frozen=True)
class ModeConfig:
"""Concrete model configuration for a single processing mode."""
mode_name: str
model_id: str
temperature: float
max_tokens: int
compute_profile: str
provider: str # "nvidia" | "groq" | "mistral"
def to_log_dict(self) -> dict:
return {
"mode_name": self.mode_name,
"model_id": self.model_id,
"temperature": self.temperature,
"max_tokens": self.max_tokens,
"compute_profile": self.compute_profile,
"provider": self.provider,
}
def _build_registry() -> dict[str, ModeConfig]:
"""
NVIDIA NIM model stack — Pluto v2.
Roles:
MODE_QUICK → Nemotron Nano 8B (high-volume: extraction workers, critic, judge)
MODE_REASONING → Nemotron Super 49B (synthesis, strategist audit, debate responder)
MODE_VISION → Nemotron Nano VL (doc parsing: tables, figures, scanned PDFs)
MODE_ULTRA → Nemotron Ultra 253B (escalation only: confidence < 0.6)
Embedding + reranking are handled separately in embedder.py and dispatcher.py
(they use /v1/embeddings and scoring endpoints, not chat completions).
Fallback: if NVIDIA_API_KEY absent, fall back to Groq or Mistral.
"""
# Check for any NVIDIA key
nvidia_keys = [
"NVIDIA_API_KEY", "NVIDIA_API_KEY_NANO", "NVIDIA_API_KEY_SUPER",
"NVIDIA_API_KEY_VL", "NVIDIA_API_KEY_EMBED", "NVIDIA_API_KEY_RERANK",
"NVIDIA_API_KEY_ULTRA"
]
nvidia_ready = any(_clean_api_key(os.getenv(k)) for k in nvidia_keys)
groq_key = _clean_api_key(os.getenv("GROQ_API_KEY", ""))
mistral_key = _clean_api_key(os.getenv("MISTRAL_API_KEY", ""))
if nvidia_ready:
return {
"MODE_QUICK": ModeConfig(
mode_name="MODE_QUICK",
model_id="meta/llama-3.2-3b-instruct",
temperature=0.1,
max_tokens=1024,
compute_profile="low-latency",
provider="nvidia",
),
"MODE_REASONING": ModeConfig(
mode_name="MODE_REASONING",
model_id="nvidia/nemotron-3-nano-omni-30b-a3b-reasoning",
temperature=0.3,
max_tokens=4096,
compute_profile="high-reasoning",
provider="nvidia",
),
"MODE_VISION": ModeConfig(
mode_name="MODE_VISION",
model_id="nvidia/llama-3.1-nemotron-nano-vl-8b-v1",
temperature=0.1,
max_tokens=4096,
compute_profile="vision-capable",
provider="nvidia",
),
"MODE_ULTRA": ModeConfig(
mode_name="MODE_ULTRA",
model_id="nvidia/llama-3.1-nemotron-ultra-253b-v1",
temperature=0.2,
max_tokens=4096,
compute_profile="deep-reasoning",
provider="nvidia",
),
# Keep MODE_GEMINI name for backward compat — maps to Super
"MODE_GEMINI": ModeConfig(
mode_name="MODE_GEMINI",
model_id="nvidia/nemotron-3-nano-omni-30b-a3b-reasoning",
temperature=0.0,
max_tokens=4096,
compute_profile="high-throughput",
provider="nvidia",
),
}
elif groq_key:
# Groq fallback — same size tiers
return {
"MODE_QUICK": ModeConfig(
mode_name="MODE_QUICK",
model_id="llama-3.1-8b-instant",
temperature=0.1,
max_tokens=1024,
compute_profile="low-latency",
provider="groq",
),
"MODE_REASONING": ModeConfig(
mode_name="MODE_REASONING",
model_id="llama-3.3-70b-versatile",
temperature=0.3,
max_tokens=4096,
compute_profile="high-reasoning",
provider="groq",
),
"MODE_VISION": ModeConfig(
mode_name="MODE_VISION",
model_id="llama-3.1-8b-instant",
temperature=0.1,
max_tokens=4096,
compute_profile="vision-capable",
provider="groq",
),
"MODE_ULTRA": ModeConfig(
mode_name="MODE_ULTRA",
model_id="llama-3.3-70b-versatile",
temperature=0.2,
max_tokens=4096,
compute_profile="deep-reasoning",
provider="groq",
),
"MODE_GEMINI": ModeConfig(
mode_name="MODE_GEMINI",
model_id="llama-3.3-70b-versatile",
temperature=0.0,
max_tokens=4096,
compute_profile="high-throughput",
provider="groq",
),
}
if mistral_key and not _looks_like_nvidia_key(mistral_key):
return _build_mistral_registry()
return _build_unconfigured_registry()
def _build_mistral_registry() -> dict[str, ModeConfig]:
"""Use Mistral for every mode when it is the only configured chat provider."""
return {
"MODE_QUICK": ModeConfig(
mode_name="MODE_QUICK",
model_id="mistral-small-latest",
temperature=0.1,
max_tokens=1024,
compute_profile="fallback",
provider="mistral",
),
"MODE_REASONING": ModeConfig(
mode_name="MODE_REASONING",
model_id="mistral-small-latest",
temperature=0.3,
max_tokens=4096,
compute_profile="fallback",
provider="mistral",
),
"MODE_VISION": ModeConfig(
mode_name="MODE_VISION",
model_id="mistral-small-latest",
temperature=0.1,
max_tokens=4096,
compute_profile="fallback",
provider="mistral",
),
"MODE_ULTRA": ModeConfig(
mode_name="MODE_ULTRA",
model_id="mistral-small-latest",
temperature=0.2,
max_tokens=4096,
compute_profile="fallback",
provider="mistral",
),
"MODE_GEMINI": ModeConfig(
mode_name="MODE_GEMINI",
model_id="mistral-small-latest",
temperature=0.0,
max_tokens=4096,
compute_profile="fallback",
provider="mistral",
),
}
def _build_unconfigured_registry() -> dict[str, ModeConfig]:
"""Return placeholder modes so imports work without provider credentials."""
return {
"MODE_QUICK": ModeConfig(
mode_name="MODE_QUICK",
model_id="unconfigured/MODE_QUICK",
temperature=0.1,
max_tokens=1024,
compute_profile="unconfigured",
provider="unconfigured",
),
"MODE_REASONING": ModeConfig(
mode_name="MODE_REASONING",
model_id="unconfigured/MODE_REASONING",
temperature=0.3,
max_tokens=4096,
compute_profile="unconfigured",
provider="unconfigured",
),
"MODE_VISION": ModeConfig(
mode_name="MODE_VISION",
model_id="unconfigured/MODE_VISION",
temperature=0.1,
max_tokens=4096,
compute_profile="unconfigured",
provider="unconfigured",
),
"MODE_ULTRA": ModeConfig(
mode_name="MODE_ULTRA",
model_id="unconfigured/MODE_ULTRA",
temperature=0.2,
max_tokens=4096,
compute_profile="unconfigured",
provider="unconfigured",
),
"MODE_GEMINI": ModeConfig(
mode_name="MODE_GEMINI",
model_id="unconfigured/MODE_GEMINI",
temperature=0.0,
max_tokens=4096,
compute_profile="unconfigured",
provider="unconfigured",
),
}
MODE_REGISTRY: dict[str, ModeConfig] = _build_registry()
def _missing_provider_error() -> EnvironmentError:
return EnvironmentError("None of NVIDIA_API_KEY, GROQ_API_KEY, or MISTRAL_API_KEY is set.")
def _is_unconfigured() -> bool:
return any(mode.provider == "unconfigured" for mode in MODE_REGISTRY.values())
def _refresh_mode_registry() -> None:
"""Refresh mode config in place so imported MODE_REGISTRY references stay valid."""
MODE_REGISTRY.clear()
MODE_REGISTRY.update(_build_registry())
def is_real_switching() -> bool:
"""True if MODE_QUICK and MODE_REASONING use DIFFERENT model_ids."""
if _is_unconfigured():
_refresh_mode_registry()
if _is_unconfigured():
return False
quick = MODE_REGISTRY["MODE_QUICK"].model_id
reasoning = MODE_REGISTRY["MODE_REASONING"].model_id
return quick != reasoning
def get_mode(mode_name: str) -> ModeConfig:
"""Look up a mode config by name."""
if mode_name not in MODE_REGISTRY:
raise ValueError(f"Unknown mode: {mode_name}. Valid: {list(MODE_REGISTRY)}")
mode = MODE_REGISTRY[mode_name]
if mode.provider == "unconfigured":
_refresh_mode_registry()
mode = MODE_REGISTRY.get(mode_name)
if mode is None:
raise ValueError(f"Unknown mode: {mode_name}. Valid: {list(MODE_REGISTRY)}")
if mode.provider == "unconfigured":
raise _missing_provider_error()
return mode
|