File size: 13,695 Bytes
9dad6a7 513ea7b 9dad6a7 513ea7b 9dad6a7 513ea7b 9dad6a7 513ea7b 9dad6a7 513ea7b 9dad6a7 513ea7b 9dad6a7 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 | from __future__ import annotations
import json
import random
from collections.abc import Iterable, Sequence
from typing import Any
from .prompts import AUTHOR_SYSTEM
from .quality import is_situation_grounded
from .safety import guard_input
from .schema import ForestDraft
CREATURES = (
("The Patient Fox", "patience under pressure", "I am allowed to learn."),
("The Listening Owl", "careful perspective", "I can listen before I leap."),
("The Brave Snail", "quiet courage", "I make progress at my pace."),
("The Steady Deer", "steadiness through change", "I can meet one moment."),
("The Clear-Voiced Wren", "honest self-expression", "I can speak gently and clearly."),
("The Curious Otter", "playful curiosity", "I can stay curious here."),
("The Gentle Bear", "self-compassion", "I can be kind to myself."),
("The Building Beaver", "practical resourcefulness", "I can shape one next step."),
)
IMAGE_PROMPTS = {
"The Patient Fox": "a gentle russet fox sitting in a mossy clearing, soft kind eyes",
"The Listening Owl": "a round tawny owl resting on a low branch, attentive kind eyes",
"The Brave Snail": "a tiny snail crossing a fern frond, softly glowing spiral shell",
"The Steady Deer": "a small deer standing in morning mist, calm gentle expression",
"The Clear-Voiced Wren": "a tiny wren singing beside dusty rose wildflowers",
"The Curious Otter": "a small otter floating beside reeds, bright gentle eyes",
"The Gentle Bear": "a sleepy bear cub resting under dappled canopy light",
"The Building Beaver": "a friendly beaver holding one smooth twig beside a stream",
}
TEMPLATE_NAMES = (
"Mika",
"Noor",
"Ari",
"Sam",
"Leila",
"Jun",
"Robin",
"Maya",
"Theo",
"Nia",
"Alex",
"Bao",
"Clara",
"Dev",
"Esme",
"Farah",
"Gabe",
"Hana",
"Iris",
"Jules",
"Kai",
"Lina",
"Mateo",
"Niko",
"Omar",
"Priya",
"Quinn",
"Ravi",
"Sora",
"Tess",
"Uma",
"Vera",
"Will",
"Xia",
"Yara",
"Zane",
"An",
"Bea",
"Cole",
"Dara",
)
TEACHER_NAMES = (
"Cora",
"Eli",
"Imani",
"Pax",
"Ren",
"Sol",
"Tala",
"Zoe",
)
LINE_TEMPLATES = (
(
"{situation} can be difficult without meaning you are failing. "
"Your {strength} can help you choose one honest next step."
),
(
"There is real uncertainty in {situation}. {strength_capitalized} does not erase it; "
"it gives you room to respond without rushing."
),
(
"When {situation_lower}, you do not have to solve the whole path today. "
"Your {strength} is enough for the next part."
),
(
"{situation} asks a lot of you. Let {strength} make the next choice smaller, "
"clearer, and kinder."
),
(
"It makes sense that {situation_lower} feels tender. Your {strength} can sit "
"beside the difficulty instead of pretending it is easy."
),
(
"You can acknowledge the hard part of {situation_lower} and still trust your "
"{strength} to help with what comes next."
),
)
REFLECTION_TEMPLATES = (
"What would change if you asked only for the next kind and concrete step?",
"Which part deserves your attention now, and which part can wait?",
"What could you notice before deciding you must already have the answer?",
"What is one choice that would make this moment more workable?",
"How might you make room for both the difficulty and your own agency?",
"What support would let you move with care instead of pressure?",
)
def normalize_positive_frame(row: dict[str, Any]) -> dict[str, str]:
return {
"situation": str(row.get("original_text", "")).strip(),
"support_hint": str(row.get("reframed_text", "")).strip(),
"strategy": str(row.get("strategy", "")).strip(),
"source": "SALT-NLP/positive_reframing",
}
def normalize_empathy(row: dict[str, Any]) -> dict[str, str]:
conversations = row.get("conversations") or []
assistant_lines = [
str(message.get("content", "")).strip()
for message in conversations
if message.get("role") == "assistant" and message.get("content")
]
return {
"situation": str(row.get("situation", "")).strip(),
"support_hint": " ".join(assistant_lines[:2]),
"strategy": str(row.get("emotion", "")).strip(),
"source": "Estwld/empathetic_dialogues_llm",
}
def validate_synthetic_example(example: dict[str, Any]) -> dict[str, Any] | None:
name = str(example.get("name", "")).strip()
situation = str(example.get("situation", "")).strip()
if not guard_input(name, situation).allowed:
return None
try:
forest = ForestDraft.model_validate(example.get("forest"))
except (TypeError, ValueError):
return None
if any(not is_situation_grounded(clearing.line, situation) for clearing in forest.clearings):
return None
return {
"name": name,
"situation": situation,
"forest": forest.model_dump(),
"source": str(example.get("source", "synthetic")),
"teacher_model": str(example.get("teacher_model", "")),
}
def build_sft_record(example: dict[str, Any]) -> dict[str, Any]:
user_content = json.dumps(
{
"name": example["name"],
"situation": example["situation"],
"validated_fact_plan": {
"faithful_summary": example["situation"],
"fact_anchors": [
{
"source_phrase": example["situation"],
"meaning": example["situation"],
}
],
"central_uncertainty": "What will happen next",
"desired_direction": "Move with clarity and care",
},
},
ensure_ascii=False,
)
assistant_content = json.dumps(example["forest"], ensure_ascii=False)
return {
"name": example["name"],
"situation": example["situation"],
"source": example.get("source", "synthetic"),
"teacher_model": example.get("teacher_model", ""),
"messages": [
{"role": "system", "content": AUTHOR_SYSTEM},
{"role": "user", "content": user_content},
{"role": "assistant", "content": assistant_content},
],
}
def deduplicate_records(records: Iterable[dict[str, Any]]) -> list[dict[str, Any]]:
seen: set[tuple[str, str]] = set()
result: list[dict[str, Any]] = []
for record in records:
identity = (
str(record["name"]).strip().casefold(),
str(record["situation"]).strip().casefold(),
)
if identity in seen:
continue
seen.add(identity)
result.append(record)
return result
def split_records(
records: Sequence[dict[str, Any]],
*,
validation_fraction: float = 0.1,
seed: int = 42,
) -> dict[str, list[dict[str, Any]]]:
if not 0 < validation_fraction < 1:
raise ValueError("validation_fraction must be between zero and one")
shuffled = list(records)
random.Random(seed).shuffle(shuffled)
validation_count = max(1, round(len(shuffled) * validation_fraction))
return {
"train": shuffled[validation_count:],
"validation": shuffled[:validation_count],
}
def template_forest(name: str, situation: str, variant: int) -> dict[str, Any]:
rotated = list(CREATURES[variant % len(CREATURES) :]) + list(
CREATURES[: variant % len(CREATURES)]
)
selected = rotated[:5]
roles = ("arrive", "steady", "widen", "step", "carry")
clearings = []
for clearing_index, (creature, strength, spell) in enumerate(selected):
line_template = LINE_TEMPLATES[(variant + clearing_index) % len(LINE_TEMPLATES)]
clearings.append(
{
"arc_role": roles[clearing_index],
"source_phrase": situation,
"creature": creature,
"strength": strength,
"line": line_template.format(
situation=situation.rstrip("."),
situation_lower=situation.rstrip(".").lower(),
strength=strength,
strength_capitalized=strength.capitalize(),
),
"reflection": REFLECTION_TEMPLATES[
(variant + clearing_index) % len(REFLECTION_TEMPLATES)
],
"spell": spell,
"image_prompt": IMAGE_PROMPTS[creature],
}
)
return {
"forest_title": (
f"{name}'s Path Through This Moment"
if variant % 2 == 0
else f"{name}'s Clearing for What Comes Next"
),
"proposed_strengths": [item[1] for item in selected],
"clearings": clearings,
}
def build_template_examples(
situations: Sequence[str],
*,
variants_per_situation: int = 4,
) -> list[dict[str, Any]]:
examples = []
for situation_index, situation in enumerate(situations):
for variant in range(variants_per_situation):
name = TEMPLATE_NAMES[(situation_index + variant) % len(TEMPLATE_NAMES)]
examples.append(
{
"name": name,
"situation": situation,
"forest": template_forest(name, situation, variant),
"source": "template_coverage",
}
)
return examples
def teacher_requests(situations: Sequence[str], *, start: int) -> list[dict[str, str]]:
return [
{
"name": TEACHER_NAMES[(start + offset) % len(TEACHER_NAMES)],
"situation": situation,
}
for offset, situation in enumerate(situations)
]
def forest_batch_json_schema() -> dict[str, Any]:
clearing = {
"type": "object",
"additionalProperties": False,
"required": [
"arc_role",
"source_phrase",
"creature",
"strength",
"line",
"reflection",
"spell",
"image_prompt",
],
"properties": {
"arc_role": {"type": "string"},
"source_phrase": {"type": "string"},
"creature": {"type": "string"},
"strength": {"type": "string"},
"line": {"type": "string"},
"reflection": {"type": "string"},
"spell": {"type": "string"},
"image_prompt": {"type": "string"},
},
}
forest = {
"type": "object",
"additionalProperties": False,
"required": ["forest_title", "proposed_strengths", "clearings"],
"properties": {
"forest_title": {"type": "string"},
"proposed_strengths": {
"type": "array",
"items": {"type": "string"},
},
"clearings": {
"type": "array",
"items": clearing,
},
},
}
return {
"type": "object",
"additionalProperties": False,
"required": ["examples"],
"properties": {
"examples": {
"type": "array",
"items": {
"type": "object",
"additionalProperties": False,
"required": ["name", "situation", "forest"],
"properties": {
"name": {"type": "string"},
"situation": {"type": "string"},
"forest": forest,
},
},
}
},
}
class CohereForestGenerator:
def __init__(
self,
api_key: str,
*,
model: str = "command-a-03-2025",
) -> None:
import cohere
self.client = cohere.ClientV2(api_key=api_key)
self.model = model
def generate(
self,
requests: Sequence[dict[str, str]],
*,
source_hints: Sequence[dict[str, str]] = (),
seed: int = 42,
) -> list[dict[str, Any]]:
prompt = {
"task": (
"Write one complete Compliment Forest for each request. Return 4-6 distinct "
"clearings. Every line must repeat at least one concrete noun or phrase from "
"its situation. Acknowledge difficulty without diagnosis, guarantees, hollow "
"praise, or toxic positivity. Spells begin with 'I' and use at most 12 words. "
"Use arrive, steady, widen, step, and optional carry in order. Each "
"source_phrase must copy exact text from the situation. Image prompts "
"describe one coherent scene and contain no style words or text."
),
"requests": list(requests),
"voice_hints": list(source_hints)[:8],
}
response = self.client.chat(
model=self.model,
messages=[
{"role": "system", "content": AUTHOR_SYSTEM},
{"role": "user", "content": json.dumps(prompt, ensure_ascii=False)},
],
response_format={
"type": "json_object",
"json_schema": forest_batch_json_schema(),
},
safety_mode="CONTEXTUAL",
temperature=0.65,
max_tokens=5000,
seed=seed,
)
payload = json.loads(response.message.content[0].text)
return list(payload.get("examples", []))
|