wassname/Open-Assistant
1
0
Fork 0
mirror of https://github.com/wassname/Open-Assistant.git synced 2026-07-20 12:10:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'main' into 763-make-labels-required

Browse Source
This commit is contained in:
Keith Stevens
2023-01-17 08:58:29 +09:00
parent 7e1b57a93e ba68f721aa
commit cf29f544eb
39 changed files with 1033 additions and 602 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/deploy-to-node.yaml
+3
View File
@@ -33,6 +33,9 @@ jobs:
WEB_EMAIL_SERVER_PORT: ${{ secrets.DEV_WEB_EMAIL_SERVER_PORT }}
WEB_EMAIL_SERVER_USER: ${{ secrets.DEV_WEB_EMAIL_SERVER_USER }}
WEB_NEXTAUTH_SECRET: ${{ secrets.DEV_WEB_NEXTAUTH_SECRET }}
S3_BUCKET_NAME: ${{ secrets.S3_BUCKET_NAME }}
AWS_ACCESS_KEY: ${{ secrets.AWS_ACCESS_KEY }}
AWS_SECRET_KEY: ${{ secrets.AWS_SECRET_KEY }}
steps:
- name: Checkout
uses: actions/checkout@v2
ansible/deploy-to-node.yaml
+26 -1
View File
@@ -78,6 +78,29 @@
- name: backend
- name: web
- name: Copy pgbackrest.conf to managed node
ansible.builtin.copy:
src: ./pgbackrest.conf
dest: "./{{ stack_name }}/pgbackrest.conf"
mode: 0644
- name: Create pgbackrest container
community.docker.docker_container:
name: "oasst-{{ stack_name }}-pgbackrest"
image: woblerr/pgbackrest:2.43
state: "{{ 'stopped' if stack_name == 'production' else 'absent' }}"
network_mode: "oasst-{{ stack_name }}"
volumes:
- "./{{ stack_name }}/pgbackrest.conf:/etc/pgbackrest/pgbackrest.conf"
- "oasst-{{ stack_name }}-postgres-backend:/var/lib/postgresql/data"
env:
PGBACKREST_REPO1_S3_BUCKET:
"{{ lookup('ansible.builtin.env', 'S3_BUCKET_NAME') }}"
PGBACKREST_REPO1_S3_KEY:
"{{ lookup('ansible.builtin.env', 'AWS_ACCESS_KEY') }}"
PGBACKREST_REPO1_S3_KEY_SECRET:
"{{ lookup('ansible.builtin.env', 'AWS_SECRET_KEY') }}"
- name: Run the oasst oasst-backend
community.docker.docker_container:
name: "oasst-{{ stack_name }}-backend"
@@ -136,7 +159,9 @@
FASTAPI_KEY: "{{ web_api_key }}"
NEXTAUTH_SECRET:
"{{ lookup('ansible.builtin.env', 'WEB_NEXTAUTH_SECRET') }}"
NEXTAUTH_URL: http://web.{{ stack_name }}.open-assistant.io/
NEXTAUTH_URL:
"{{ 'https://open-assistant.io/' if stack_name == 'production' else
('https://web.' + stack_name + '.open-assistant.io/') }}"
ports:
- "{{ website_port }}:3000"
command: bash wait-for-postgres.sh node server.js
ansible/pgbackrest.conf
+24
View File
@@ -0,0 +1,24 @@
[oasst]
pg1-path=/var/lib/postgresql/data
[global]
repo1-retention-full=3 # keep last 3 backups
repo1-type=s3
repo1-path=/oasst-prod
repo1-s3-region=us-east-1
repo1-s3-endpoint=s3.amazonaws.com
# repo1-s3-bucket=$S3_BUCKET_NAME
# repo1-s3-key=$AWS_ACCESS_KEY
# repo1-s3-key-secret=$AWS_SECRET_KEY
# Force a checkpoint to start backup immediately.
start-fast=y
# Use delta restore.
delta=y
# Enable ZSTD compression.
compress-type=zst
compress-level=6
log-level-console=info
log-level-file=debug
backend/main.py
+78 -80
View File
@@ -20,6 +20,7 @@ from oasst_backend.models import message_tree_state
from oasst_backend.prompt_repository import PromptRepository, TaskRepository, UserRepository
from oasst_backend.tree_manager import TreeManager
from oasst_backend.user_stats_repository import UserStatsRepository, UserStatsTimeFrame
from oasst_backend.utils.database_utils import CommitMode, managed_tx_function
from oasst_shared.exceptions import OasstError, OasstErrorCode
from oasst_shared.schemas import protocol as protocol_schema
from pydantic import BaseModel
@@ -120,7 +121,8 @@ if settings.RATE_LIMIT:
if settings.DEBUG_USE_SEED_DATA:
@app.on_event("startup")
def seed_data():
@managed_tx_function(auto_commit=CommitMode.COMMIT)
def create_seed_data(session: Session):
class DummyMessage(BaseModel):
task_message_id: str
user_message_id: str
@@ -134,73 +136,73 @@ if settings.DEBUG_USE_SEED_DATA:
try:
logger.info("Seed data check began")
with Session(engine) as db:
api_client = api_auth(settings.OFFICIAL_WEB_API_KEY, db=db)
dummy_user = protocol_schema.User(id="__dummy_user__", display_name="Dummy User", auth_method="local")
ur = UserRepository(db=db, api_client=api_client)
tr = TaskRepository(db=db, api_client=api_client, client_user=dummy_user, user_repository=ur)
pr = PromptRepository(
db=db, api_client=api_client, client_user=dummy_user, user_repository=ur, task_repository=tr
)
tm = TreeManager(db, pr)
api_client = api_auth(settings.OFFICIAL_WEB_API_KEY, db=session)
dummy_user = protocol_schema.User(id="__dummy_user__", display_name="Dummy User", auth_method="local")
with open(settings.DEBUG_USE_SEED_DATA_PATH) as f:
dummy_messages_raw = json.load(f)
ur = UserRepository(db=session, api_client=api_client)
tr = TaskRepository(db=session, api_client=api_client, client_user=dummy_user, user_repository=ur)
pr = PromptRepository(
db=session, api_client=api_client, client_user=dummy_user, user_repository=ur, task_repository=tr
)
tm = TreeManager(session, pr)
dummy_messages = [DummyMessage(**dm) for dm in dummy_messages_raw]
with open(settings.DEBUG_USE_SEED_DATA_PATH) as f:
dummy_messages_raw = json.load(f)
for msg in dummy_messages:
task = tr.fetch_task_by_frontend_message_id(msg.task_message_id)
if task and not task.ack:
logger.warning("Deleting unacknowledged seed data task")
db.delete(task)
task = None
if not task:
if msg.parent_message_id is None:
task = tr.store_task(
protocol_schema.InitialPromptTask(hint=""), message_tree_id=None, parent_message_id=None
)
else:
parent_message = pr.fetch_message_by_frontend_message_id(
msg.parent_message_id, fail_if_missing=True
)
conversation_messages = pr.fetch_message_conversation(parent_message)
conversation = prepare_conversation(conversation_messages)
if msg.role == "assistant":
task = tr.store_task(
protocol_schema.AssistantReplyTask(conversation=conversation),
message_tree_id=parent_message.message_tree_id,
parent_message_id=parent_message.id,
)
else:
task = tr.store_task(
protocol_schema.PrompterReplyTask(conversation=conversation),
message_tree_id=parent_message.message_tree_id,
parent_message_id=parent_message.id,
)
tr.bind_frontend_message_id(task.id, msg.task_message_id)
message = pr.store_text_reply(
msg.text,
msg.task_message_id,
msg.user_message_id,
review_count=5,
review_result=True,
check_tree_state=False,
)
if message.parent_id is None:
tm._insert_default_state(
root_message_id=message.id, state=msg.tree_state or message_tree_state.State.GROWING
)
db.commit()
dummy_messages = [DummyMessage(**dm) for dm in dummy_messages_raw]
logger.info(
f"Inserted: message_id: {message.id}, payload: {message.payload.payload}, parent_message_id: {message.parent_id}"
for msg in dummy_messages:
task = tr.fetch_task_by_frontend_message_id(msg.task_message_id)
if task and not task.ack:
logger.warning("Deleting unacknowledged seed data task")
session.delete(task)
task = None
if not task:
if msg.parent_message_id is None:
task = tr.store_task(
protocol_schema.InitialPromptTask(hint=""), message_tree_id=None, parent_message_id=None
)
else:
logger.debug(f"seed data task found: {task.id}")
parent_message = pr.fetch_message_by_frontend_message_id(
msg.parent_message_id, fail_if_missing=True
)
conversation_messages = pr.fetch_message_conversation(parent_message)
conversation = prepare_conversation(conversation_messages)
if msg.role == "assistant":
task = tr.store_task(
protocol_schema.AssistantReplyTask(conversation=conversation),
message_tree_id=parent_message.message_tree_id,
parent_message_id=parent_message.id,
)
else:
task = tr.store_task(
protocol_schema.PrompterReplyTask(conversation=conversation),
message_tree_id=parent_message.message_tree_id,
parent_message_id=parent_message.id,
)
tr.bind_frontend_message_id(task.id, msg.task_message_id)
message = pr.store_text_reply(
msg.text,
msg.task_message_id,
msg.user_message_id,
review_count=5,
review_result=True,
check_tree_state=False,
)
if message.parent_id is None:
tm._insert_default_state(
root_message_id=message.id, state=msg.tree_state or message_tree_state.State.GROWING
)
session.flush()
logger.info("Seed data check completed")
logger.info(
f"Inserted: message_id: {message.id}, payload: {message.payload.payload}, parent_message_id: {message.parent_id}"
)
else:
logger.debug(f"seed data task found: {task.id}")
logger.info("Seed data check completed")
except Exception:
logger.exception("Seed data insertion failed")
@@ -220,48 +222,44 @@ def ensure_tree_states():
@app.on_event("startup")
@repeat_every(seconds=60 * settings.USER_STATS_INTERVAL_DAY, wait_first=False)
def update_leader_board_day() -> None:
@managed_tx_function(auto_commit=CommitMode.COMMIT)
def update_leader_board_day(session: Session) -> None:
try:
with Session(engine) as session:
usr = UserStatsRepository(session)
usr.update_stats(time_frame=UserStatsTimeFrame.day)
session.commit()
usr = UserStatsRepository(session)
usr.update_stats(time_frame=UserStatsTimeFrame.day)
except Exception:
logger.exception("Error during leaderboard update (daily)")
@app.on_event("startup")
@repeat_every(seconds=60 * settings.USER_STATS_INTERVAL_WEEK, wait_first=False)
def update_leader_board_week() -> None:
@managed_tx_function(auto_commit=CommitMode.COMMIT)
def update_leader_board_week(session: Session) -> None:
try:
with Session(engine) as session:
usr = UserStatsRepository(session)
usr.update_stats(time_frame=UserStatsTimeFrame.week)
session.commit()
usr = UserStatsRepository(session)
usr.update_stats(time_frame=UserStatsTimeFrame.week)
except Exception:
logger.exception("Error during user states update (weekly)")
@app.on_event("startup")
@repeat_every(seconds=60 * settings.USER_STATS_INTERVAL_MONTH, wait_first=False)
def update_leader_board_month() -> None:
@managed_tx_function(auto_commit=CommitMode.COMMIT)
def update_leader_board_month(session: Session) -> None:
try:
with Session(engine) as session:
usr = UserStatsRepository(session)
usr.update_stats(time_frame=UserStatsTimeFrame.month)
session.commit()
usr = UserStatsRepository(session)
usr.update_stats(time_frame=UserStatsTimeFrame.month)
except Exception:
logger.exception("Error during user states update (monthly)")
@app.on_event("startup")
@repeat_every(seconds=60 * settings.USER_STATS_INTERVAL_TOTAL, wait_first=False)
def update_leader_board_total() -> None:
@managed_tx_function(auto_commit=CommitMode.COMMIT)
def update_leader_board_total(session: Session) -> None:
try:
with Session(engine) as session:
usr = UserStatsRepository(session)
usr.update_stats(time_frame=UserStatsTimeFrame.total)
session.commit()
usr = UserStatsRepository(session)
usr.update_stats(time_frame=UserStatsTimeFrame.total)
except Exception:
logger.exception("Error during user states update (total)")
backend/oasst_backend/api/v1/tasks.py
+22 -1
View File
@@ -1,4 +1,4 @@
from typing import Any
from typing import Any, Optional
from uuid import UUID
from fastapi import APIRouter, Depends
@@ -48,6 +48,27 @@ def request_task(
return task
@router.post("/availability", response_model=dict[protocol_schema.TaskRequestType, int])
def tasks_availability(
*,
user: Optional[protocol_schema.User] = None,
db: Session = Depends(deps.get_db),
api_key: APIKey = Depends(deps.get_api_key),
):
api_client = deps.api_auth(api_key, db)
try:
pr = PromptRepository(db, api_client, client_user=user)
tm = TreeManager(db, pr)
return tm.determine_task_availability()
except OasstError:
raise
except Exception:
logger.exception("Task availability query failed.")
raise OasstError("Task availability query failed.", OasstErrorCode.TASK_AVAILABILITY_QUERY_FAILED)
@router.post("/{task_id}/ack", response_model=None, status_code=HTTP_204_NO_CONTENT)
def tasks_acknowledge(
*,
backend/oasst_backend/config.py
+3
View File
@@ -55,6 +55,8 @@ class TreeManagerConfiguration(BaseModel):
mandatory_labels_prompter_reply: Optional[list[protocol_schema.TextLabel]] = [protocol_schema.TextLabel.spam]
"""Mandatory labels in text-labeling tasks for prompter replies."""
rank_prompter_replies: bool = False
class Settings(BaseSettings):
PROJECT_NAME: str = "open-assistant backend"
@@ -67,6 +69,7 @@ class Settings(BaseSettings):
POSTGRES_PASSWORD: str = "postgres"
POSTGRES_DB: str = "postgres"
DATABASE_URI: Optional[PostgresDsn] = None
DATABASE_MAX_TX_RETRY_COUNT: int = 3
RATE_LIMIT: bool = True
REDIS_HOST: str = "localhost"
backend/oasst_backend/tree_manager.py
+351 -261
View File
@@ -9,14 +9,14 @@ import pydantic
from loguru import logger
from oasst_backend.api.v1.utils import prepare_conversation, prepare_conversation_message_list
from oasst_backend.config import TreeManagerConfiguration, settings
from oasst_backend.models import Message, MessageReaction, MessageTreeState, TextLabels, message_tree_state
from oasst_backend.models import Message, MessageReaction, MessageTreeState, Task, TextLabels, message_tree_state
from oasst_backend.prompt_repository import PromptRepository
from oasst_backend.utils.database_utils import CommitMode, async_managed_tx_method, managed_tx_method
from oasst_backend.utils.hugging_face import HfClassificationModel, HfEmbeddingModel, HfUrl, HuggingFaceAPI
from oasst_shared.exceptions.oasst_api_error import OasstError, OasstErrorCode
from oasst_shared.schemas import protocol as protocol_schema
from sqlalchemy.sql import text
from sqlmodel import Session, func
from sqlmodel import Session, func, not_
class TaskType(Enum):
@@ -49,6 +49,7 @@ class ActiveTreeSizeRow(pydantic.BaseModel):
class ExtendibleParentRow(pydantic.BaseModel):
parent_id: UUID
parent_role: str
depth: int
message_tree_id: UUID
active_children_count: int
@@ -59,6 +60,7 @@ class ExtendibleParentRow(pydantic.BaseModel):
class IncompleteRankingsRow(pydantic.BaseModel):
parent_id: UUID
role: str
children_count: int
child_min_ranking_count: int
@@ -70,21 +72,23 @@ class TreeManager:
_all_text_labels = list(map(lambda x: x.value, protocol_schema.TextLabel))
def __init__(
self, db: Session, prompt_repository: PromptRepository, cfg: Optional[TreeManagerConfiguration] = None
self,
db: Session,
prompt_repository: PromptRepository,
cfg: Optional[TreeManagerConfiguration] = None,
):
self.db = db
self.cfg = cfg or settings.tree_manager
self.pr = prompt_repository
def _task_selection(
def _random_task_selection(
self,
desired_task_type: protocol_schema.TaskRequestType,
num_ranking_tasks: int,
num_replies_need_review: int,
num_prompts_need_review: int,
num_missing_prompts: int,
num_missing_replies: int,
) -> Tuple[TaskType, TaskRole]:
) -> TaskType:
"""
Determines which task to hand out to human worker.
The task type is drawn with relative weight (e.g. ranking has highest priority)
@@ -92,75 +96,97 @@ class TreeManager:
"""
logger.debug(
f"TreeManager._task_selection({num_ranking_tasks=}, {num_replies_need_review=}, "
f"TreeManager._random_task_selection({num_ranking_tasks=}, {num_replies_need_review=}, "
f"{num_prompts_need_review=}, {num_missing_prompts=}, {num_missing_replies=})"
)
task_type = TaskType.NONE
task_role = TaskRole.ANY
if desired_task_type == protocol_schema.TaskRequestType.random:
task_weights = [0] * 5
task_weights = [0] * 5
if num_ranking_tasks > 0:
task_weights[TaskType.RANKING.value] = 10
if num_ranking_tasks > 0:
task_weights[TaskType.RANKING.value] = 10
if num_replies_need_review > 0:
task_weights[TaskType.LABEL_REPLY.value] = 5
if num_replies_need_review > 0:
task_weights[TaskType.LABEL_REPLY.value] = 5
if num_prompts_need_review > 0:
task_weights[TaskType.LABEL_PROMPT.value] = 5
if num_prompts_need_review > 0:
task_weights[TaskType.LABEL_PROMPT.value] = 5
if num_missing_replies > 0:
task_weights[TaskType.REPLY.value] = 2
if num_missing_replies > 0:
task_weights[TaskType.REPLY.value] = 2
if num_missing_prompts > 0:
task_weights[TaskType.PROMPT.value] = 1
if num_missing_prompts > 0:
task_weights[TaskType.PROMPT.value] = 1
task_weights = np.array(task_weights)
weight_sum = task_weights.sum()
if weight_sum < 1e-8:
task_type = TaskType.NONE
else:
task_weights = task_weights / weight_sum
task_type = TaskType(np.random.choice(a=len(task_weights), p=task_weights))
else:
match desired_task_type:
case protocol_schema.TaskRequestType.initial_prompt:
if num_missing_prompts > 0:
task_type = TaskType.PROMPT
case protocol_schema.TaskRequestType.label_initial_prompt:
if num_prompts_need_review > 0:
task_type = TaskType.LABEL_PROMPT
case protocol_schema.TaskRequestType.assistant_reply | protocol_schema.TaskRequestType.prompter_reply:
if num_missing_replies > 0:
task_role = (
TaskRole.ASSISTANT
if desired_task_type == protocol_schema.TaskRequestType.assistant_reply
else TaskRole.PROMPTER
)
task_type = TaskType.REPLY
case protocol_schema.TaskRequestType.label_assistant_reply | protocol_schema.TaskRequestType.label_prompter_reply:
if num_replies_need_review > 0:
task_role = (
TaskRole.ASSISTANT
if desired_task_type == protocol_schema.TaskRequestType.label_assistant_reply
else TaskRole.PROMPTER
)
task_type = TaskType.LABEL_REPLY
case protocol_schema.TaskRequestType.rank_assistant_replies | protocol_schema.TaskRequestType.rank_prompter_replies:
if num_ranking_tasks > 0:
task_role = (
TaskRole.ASSISTANT
if desired_task_type == protocol_schema.TaskRequestType.rank_assistant_replies
else TaskRole.PROMPTER
)
task_type = TaskType.RANKING
task_weights = np.array(task_weights)
weight_sum = task_weights.sum()
if weight_sum > 1e-8:
task_weights = task_weights / weight_sum
task_type = TaskType(np.random.choice(a=len(task_weights), p=task_weights))
logger.debug(f"Selected {task_type=}, {task_role=}")
return task_type, task_role
logger.debug(f"Selected {task_type=}")
return task_type
def _determine_task_availability_internal(
self,
num_active_trees: int,
extensible_parents: list[ExtendibleParentRow],
prompts_need_review: list[Message],
replies_need_review: list[Message],
incomplete_rankings: list[IncompleteRankingsRow],
) -> dict[protocol_schema.TaskRequestType, int]:
task_count_by_type: dict[protocol_schema.TaskRequestType, int] = {t: 0 for t in protocol_schema.TaskRequestType}
num_missing_prompts = max(0, self.cfg.max_active_trees - num_active_trees)
task_count_by_type[protocol_schema.TaskRequestType.initial_prompt] = num_missing_prompts
task_count_by_type[protocol_schema.TaskRequestType.prompter_reply] = len(
list(filter(lambda x: x.parent_role == "assistant", extensible_parents))
)
task_count_by_type[protocol_schema.TaskRequestType.assistant_reply] = len(
list(filter(lambda x: x.parent_role == "prompter", extensible_parents))
)
task_count_by_type[protocol_schema.TaskRequestType.label_initial_prompt] = len(prompts_need_review)
task_count_by_type[protocol_schema.TaskRequestType.label_assistant_reply] = len(
list(filter(lambda m: m.role == "assistant", replies_need_review))
)
task_count_by_type[protocol_schema.TaskRequestType.prompter_reply] = len(
list(filter(lambda m: m.role == "prompter", replies_need_review))
)
if self.cfg.rank_prompter_replies:
task_count_by_type[protocol_schema.TaskRequestType.rank_prompter_replies] = len(
list(filter(lambda r: r.role == "prompter", incomplete_rankings))
)
task_count_by_type[protocol_schema.TaskRequestType.rank_assistant_replies] = len(
list(filter(lambda r: r.role == "assistant", incomplete_rankings))
)
task_count_by_type[protocol_schema.TaskRequestType.random] = sum(
task_count_by_type[t] for t in protocol_schema.TaskRequestType if t in task_count_by_type
)
return task_count_by_type
def determine_task_availability(self) -> dict[protocol_schema.TaskRequestType, int]:
num_active_trees = self.query_num_active_trees()
extensible_parents = self.query_extendible_parents()
prompts_need_review = self.query_prompts_need_review()
replies_need_review = self.query_replies_need_review()
incomplete_rankings = self.query_incomplete_rankings()
return self._determine_task_availability_internal(
num_active_trees=num_active_trees,
extensible_parents=extensible_parents,
prompts_need_review=prompts_need_review,
replies_need_review=replies_need_review,
incomplete_rankings=incomplete_rankings,
)
def next_task(
self, desired_task_type: protocol_schema.TaskRequestType
self, desired_task_type: protocol_schema.TaskRequestType = protocol_schema.TaskRequestType.random
) -> Tuple[protocol_schema.Task, Optional[UUID], Optional[UUID]]:
logger.debug("TreeManager.next_task()")
@@ -168,148 +194,195 @@ class TreeManager:
num_active_trees = self.query_num_active_trees()
prompts_need_review = self.query_prompts_need_review()
replies_need_review = self.query_replies_need_review()
extensible_parents = self.query_extendible_parents()
incomplete_rankings = self.query_incomplete_rankings()
if not self.cfg.rank_prompter_replies:
incomplete_rankings = list(filter(lambda r: r.role == "assistant", incomplete_rankings))
active_tree_sizes = self.query_extendible_trees()
# determine type of task to generate
num_missing_replies = sum(x.remaining_messages for x in active_tree_sizes)
task_type, task_role = self._task_selection(
desired_task_type,
num_ranking_tasks=len(incomplete_rankings),
num_replies_need_review=len(replies_need_review),
num_prompts_need_review=len(prompts_need_review),
num_missing_prompts=max(0, self.cfg.max_active_trees - num_active_trees),
num_missing_replies=num_missing_replies,
)
if task_type == TaskType.NONE:
raise OasstError(
f"No tasks of type '{desired_task_type.value}' are currently available.",
OasstErrorCode.TASK_REQUESTED_TYPE_NOT_AVAILABLE,
HTTPStatus.SERVICE_UNAVAILABLE,
task_role = TaskRole.ANY
if desired_task_type == protocol_schema.TaskRequestType.random:
task_type = self._random_task_selection(
num_ranking_tasks=len(incomplete_rankings),
num_replies_need_review=len(replies_need_review),
num_prompts_need_review=len(prompts_need_review),
num_missing_prompts=max(0, self.cfg.max_active_trees - num_active_trees),
num_missing_replies=num_missing_replies,
)
if task_role != TaskRole.ANY:
# Todo: Allow role specific message selection...
raise OasstError(
f"No tasks of type '{desired_task_type.value}' are currently available.",
OasstErrorCode.TASK_REQUESTED_TYPE_NOT_AVAILABLE,
HTTPStatus.SERVICE_UNAVAILABLE,
if task_type == TaskType.NONE:
raise OasstError(
f"No tasks of type '{protocol_schema.TaskRequestType.random.value}' are currently available.",
OasstErrorCode.TASK_REQUESTED_TYPE_NOT_AVAILABLE,
HTTPStatus.SERVICE_UNAVAILABLE,
)
else:
task_count_by_type = self._determine_task_availability_internal(
num_active_trees=num_active_trees,
extensible_parents=extensible_parents,
prompts_need_review=prompts_need_review,
replies_need_review=replies_need_review,
incomplete_rankings=incomplete_rankings,
)
available_count = task_count_by_type.get(desired_task_type)
if not available_count:
raise OasstError(
f"No tasks of type '{desired_task_type.value}' are currently available.",
OasstErrorCode.TASK_REQUESTED_TYPE_NOT_AVAILABLE,
HTTPStatus.SERVICE_UNAVAILABLE,
)
task_type_role_map = {
protocol_schema.TaskRequestType.initial_prompt: (TaskType.PROMPT, TaskRole.ANY),
protocol_schema.TaskRequestType.prompter_reply: (TaskType.REPLY, TaskRole.PROMPTER),
protocol_schema.TaskRequestType.assistant_reply: (TaskType.REPLY, TaskRole.ASSISTANT),
protocol_schema.TaskRequestType.rank_prompter_replies: (TaskType.RANKING, TaskRole.PROMPTER),
protocol_schema.TaskRequestType.rank_assistant_replies: (TaskType.RANKING, TaskRole.ASSISTANT),
protocol_schema.TaskRequestType.label_initial_prompt: (TaskType.LABEL_PROMPT, TaskRole.ANY),
protocol_schema.TaskRequestType.label_assistant_reply: (TaskType.LABEL_REPLY, TaskRole.ASSISTANT),
protocol_schema.TaskRequestType.label_prompter_reply: (TaskType.LABEL_REPLY, TaskRole.PROMPTER),
}
task_type, task_role = task_type_role_map[desired_task_type]
message_tree_id = None
parent_message_id = None
logger.debug(f"selected {task_type=}")
match task_type:
case TaskType.RANKING:
assert len(incomplete_rankings) > 0
ranking_parent_id = random.choice(incomplete_rankings).parent_id
if task_role == TaskRole.PROMPTER:
incomplete_rankings = list(filter(lambda m: m.role == "prompter", incomplete_rankings))
elif task_role == TaskRole.ASSISTANT:
incomplete_rankings = list(filter(lambda m: m.role == "assistant", incomplete_rankings))
messages = self.pr.fetch_message_conversation(ranking_parent_id)
assert len(messages) > 1 and messages[-1].id == ranking_parent_id
ranking_parent = messages[-1]
assert not ranking_parent.deleted and ranking_parent.review_result
conversation = prepare_conversation(messages)
replies = self.pr.fetch_message_children(ranking_parent_id, reviewed=True, exclude_deleted=True)
if len(incomplete_rankings) > 0:
ranking_parent_id = random.choice(incomplete_rankings).parent_id
assert len(replies) > 1
random.shuffle(replies) # hand out replies in random order
reply_messages = prepare_conversation_message_list(replies)
replies = [p.text for p in replies]
messages = self.pr.fetch_message_conversation(ranking_parent_id)
assert len(messages) > 1 and messages[-1].id == ranking_parent_id
ranking_parent = messages[-1]
assert not ranking_parent.deleted and ranking_parent.review_result
conversation = prepare_conversation(messages)
replies = self.pr.fetch_message_children(ranking_parent_id, reviewed=True, exclude_deleted=True)
if messages[-1].role == "assistant":
logger.info("Generating a RankPrompterRepliesTask.")
task = protocol_schema.RankPrompterRepliesTask(
conversation=conversation,
replies=replies,
reply_messages=reply_messages,
ranking_parent_id=ranking_parent.id,
message_tree_id=ranking_parent.message_tree_id,
)
else:
logger.info("Generating a RankAssistantRepliesTask.")
task = protocol_schema.RankAssistantRepliesTask(
conversation=conversation,
replies=replies,
reply_messages=reply_messages,
ranking_parent_id=ranking_parent.id,
message_tree_id=ranking_parent.message_tree_id,
)
assert len(replies) > 1
random.shuffle(replies) # hand out replies in random order
reply_messages = prepare_conversation_message_list(replies)
replies = [p.text for p in replies]
parent_message_id = ranking_parent_id
message_tree_id = messages[-1].message_tree_id
if messages[-1].role == "assistant":
logger.info("Generating a RankPrompterRepliesTask.")
task = protocol_schema.RankPrompterRepliesTask(
conversation=conversation,
replies=replies,
reply_messages=reply_messages,
ranking_parent_id=ranking_parent.id,
message_tree_id=ranking_parent.message_tree_id,
)
else:
logger.info("Generating a RankAssistantRepliesTask.")
task = protocol_schema.RankAssistantRepliesTask(
conversation=conversation,
replies=replies,
reply_messages=reply_messages,
ranking_parent_id=ranking_parent.id,
message_tree_id=ranking_parent.message_tree_id,
)
parent_message_id = ranking_parent_id
message_tree_id = messages[-1].message_tree_id
case TaskType.LABEL_REPLY:
assert len(replies_need_review) > 0
random_reply_message_id = random.choice(replies_need_review)
messages = self.pr.fetch_message_conversation(random_reply_message_id)
if task_role == TaskRole.PROMPTER:
replies_need_review = list(filter(lambda m: m.role == "prompter", replies_need_review))
elif task_role == TaskRole.ASSISTANT:
replies_need_review = list(filter(lambda m: m.role == "assistant", replies_need_review))
conversation = prepare_conversation(messages[:-1])
message = messages[-1]
if len(replies_need_review) > 0:
random_reply_message = random.choice(replies_need_review)
messages = self.pr.fetch_message_conversation(random_reply_message)
self.cfg.p_full_labeling_review_reply_prompter: float = 0.1
conversation = prepare_conversation(messages[:-1])
message = messages[-1]
label_mode = protocol_schema.LabelTaskMode.full
valid_labels = self._all_text_labels
self.cfg.p_full_labeling_review_reply_prompter: float = 0.1
if message.role == "assistant":
if random.random() > self.cfg.p_full_labeling_review_reply_assistant:
valid_labels = list(map(lambda x: x.value, self.cfg.mandatory_labels_assistant_reply))
label_mode = protocol_schema.LabelTaskMode.simple
logger.info(f"Generating a LabelAssistantReplyTask. ({label_mode=:s})")
task = protocol_schema.LabelAssistantReplyTask(
message_id=message.id,
conversation=conversation,
reply=message.text,
valid_labels=valid_labels,
mandatory_labels=list(map(lambda x: x.value, self.cfg.mandatory_labels_assistant_reply)),
mode=label_mode,
)
else:
if random.random() > self.cfg.p_full_labeling_review_reply_prompter:
valid_labels = list(map(lambda x: x.value, self.cfg.mandatory_labels_prompter_reply))
label_mode = protocol_schema.LabelTaskMode.simple
logger.info(f"Generating a LabelPrompterReplyTask. ({label_mode=:s})")
task = protocol_schema.LabelPrompterReplyTask(
message_id=message.id,
conversation=conversation,
reply=message.text,
valid_labels=valid_labels,
mandatory_labels=list(map(lambda x: x.value, self.cfg.mandatory_labels_prompter_reply)),
mode=label_mode,
)
label_mode = protocol_schema.LabelTaskMode.full
valid_labels = self._all_text_labels
parent_message_id = message.id
message_tree_id = message.message_tree_id
if message.role == "assistant":
if (
desired_task_type == protocol_schema.TaskRequestType.random
and random.random() > self.cfg.p_full_labeling_review_reply_assistant
):
valid_labels = list(map(lambda x: x.value, self.cfg.mandatory_labels_assistant_reply))
label_mode = protocol_schema.LabelTaskMode.simple
logger.info(f"Generating a LabelAssistantReplyTask. ({label_mode=:s})")
task = protocol_schema.LabelAssistantReplyTask(
message_id=message.id,
conversation=conversation,
reply=message.text,
valid_labels=valid_labels,
mandatory_labels=list(map(lambda x: x.value, self.cfg.mandatory_labels_assistant_reply)),
mode=label_mode,
)
else:
if (
desired_task_type == protocol_schema.TaskRequestType.random
and random.random() > self.cfg.p_full_labeling_review_reply_prompter
):
valid_labels = list(map(lambda x: x.value, self.cfg.mandatory_labels_prompter_reply))
label_mode = protocol_schema.LabelTaskMode.simple
logger.info(f"Generating a LabelPrompterReplyTask. ({label_mode=:s})")
task = protocol_schema.LabelPrompterReplyTask(
message_id=message.id,
conversation=conversation,
reply=message.text,
valid_labels=valid_labels,
mandatory_labels=list(map(lambda x: x.value, self.cfg.mandatory_labels_prompter_reply)),
mode=label_mode,
)
parent_message_id = message.id
message_tree_id = message.message_tree_id
case TaskType.REPLY:
# select a tree with missing replies
extensible_parents = self.query_extendible_parents()
assert len(extensible_parents) > 0
if task_role == TaskRole.PROMPTER:
extensible_parents = list(filter(lambda x: x.parent_role == "assistant", extensible_parents))
elif task_role == TaskRole.ASSISTANT:
extensible_parents = list(filter(lambda x: x.parent_role == "prompter", extensible_parents))
# fetch random conversation to extend
random_parent = random.choice(extensible_parents)
logger.debug(f"selected {random_parent=}")
messages = self.pr.fetch_message_conversation(random_parent.parent_id)
assert all(m.review_result for m in messages) # ensure all messages have positive review
conversation = prepare_conversation(messages)
if len(extensible_parents) > 0:
random_parent = random.choice(extensible_parents)
# generate reply task depending on last message
if messages[-1].role == "assistant":
logger.info("Generating a PrompterReplyTask.")
task = protocol_schema.PrompterReplyTask(conversation=conversation)
else:
logger.info("Generating a AssistantReplyTask.")
task = protocol_schema.AssistantReplyTask(conversation=conversation)
# fetch random conversation to extend
logger.debug(f"selected {random_parent=}")
messages = self.pr.fetch_message_conversation(random_parent.parent_id)
assert all(m.review_result for m in messages) # ensure all messages have positive review
conversation = prepare_conversation(messages)
parent_message_id = messages[-1].id
message_tree_id = messages[-1].message_tree_id
# generate reply task depending on last message
if messages[-1].role == "assistant":
logger.info("Generating a PrompterReplyTask.")
task = protocol_schema.PrompterReplyTask(conversation=conversation)
else:
logger.info("Generating a AssistantReplyTask.")
task = protocol_schema.AssistantReplyTask(conversation=conversation)
parent_message_id = messages[-1].id
message_tree_id = messages[-1].message_tree_id
case TaskType.LABEL_PROMPT:
assert len(prompts_need_review) > 0
message = self.pr.fetch_message(random.choice(prompts_need_review))
message = random.choice(prompts_need_review)
label_mode = protocol_schema.LabelTaskMode.full
valid_labels = self._all_text_labels
@@ -337,6 +410,13 @@ class TreeManager:
case _:
task = None
if task is None:
raise OasstError(
f"No task of type '{desired_task_type.value}' is currently available.",
OasstErrorCode.TASK_REQUESTED_TYPE_NOT_AVAILABLE,
HTTPStatus.SERVICE_UNAVAILABLE,
)
logger.info(f"Generated {task=}.")
return task, message_tree_id, parent_message_id
@@ -515,7 +595,8 @@ class TreeManager:
logger.debug(f"False {mts.active=}, {mts.state=}")
return False
rankings_by_message = self.query_tree_ranking_results(message_tree_id)
ranking_role_filter = None if self.cfg.rank_prompter_replies else "assistant"
rankings_by_message = self.query_tree_ranking_results(message_tree_id, role_filter=ranking_role_filter)
for parent_msg_id, ranking in rankings_by_message.items():
if len(ranking) < self.cfg.num_required_rankings:
logger.debug(f"False {parent_msg_id=} {len(ranking)=}")
@@ -528,68 +609,59 @@ class TreeManager:
# calculate acceptance based on spam label
return np.mean([1 - l.labels[protocol_schema.TextLabel.spam] for l in labels])
_sql_find_prompts_need_review = """
-- find initial prompts that need more reviews
SELECT m.id
FROM message_tree_state mts
LEFT JOIN message m ON mts.message_tree_id = m.id
WHERE mts.active
AND mts.state = :state
AND NOT m.review_result
AND NOT m.deleted
AND m.review_count < :num_reviews_initial_prompt
AND m.parent_id is NULL
AND (:excluded_user_id IS NULL OR m.user_id != :excluded_user_id)
"""
def query_prompts_need_review(self) -> list[UUID]:
def query_prompts_need_review(self) -> list[Message]:
"""
Select id of initial prompts with less then required rankings in active message tree
Select initial prompt messages with less then required rankings in active message tree
(active == True in message_tree_state)
"""
r = self.db.execute(
text(self._sql_find_prompts_need_review),
{
"state": message_tree_state.State.INITIAL_PROMPT_REVIEW,
"num_reviews_initial_prompt": self.cfg.num_reviews_initial_prompt,
"excluded_user_id": None if settings.DEBUG_ALLOW_SELF_LABELING else self.pr.user_id,
},
qry = (
self.db.query(Message)
.select_from(MessageTreeState)
.outerjoin(Message, MessageTreeState.message_tree_id == Message.message_tree_id)
.filter(
MessageTreeState.active,
MessageTreeState.state == message_tree_state.State.INITIAL_PROMPT_REVIEW,
not_(Message.review_result),
not_(Message.deleted),
Message.review_count < self.cfg.num_reviews_initial_prompt,
Message.parent_id.is_(None),
)
)
return [x["id"] for x in r.all()]
_sql_find_replies_need_review = """
SELECT m.id
FROM message_tree_state mts
LEFT JOIN message m ON mts.message_tree_id = m.message_tree_id
WHERE mts.active
AND mts.state = :breeding_state
AND NOT m.review_result
AND NOT m.deleted
AND m.review_count < :num_required_reviews
AND m.parent_id is NOT NULL
AND (:excluded_user_id IS NULL OR m.user_id != :excluded_user_id)
"""
if not settings.DEBUG_ALLOW_SELF_LABELING:
qry = qry.filter(Message.user_id != self.pr.user_id)
def query_replies_need_review(self) -> list[UUID]:
return qry.all()
def query_replies_need_review(self) -> list[Message]:
"""
Select ids of child messages (parent_id IS NOT NULL) with less then required rankings
Select child messages (parent_id IS NOT NULL) with less then required rankings
in active message tree (active == True in message_tree_state)
"""
r = self.db.execute(
text(self._sql_find_replies_need_review),
{
"breeding_state": message_tree_state.State.GROWING,
"num_required_reviews": self.cfg.num_reviews_reply,
"excluded_user_id": None if settings.DEBUG_ALLOW_SELF_LABELING else self.pr.user_id,
},
qry = (
self.db.query(Message)
.select_from(MessageTreeState)
.outerjoin(Message, MessageTreeState.message_tree_id == Message.message_tree_id)
.filter(
MessageTreeState.active,
MessageTreeState.state == message_tree_state.State.GROWING,
not_(Message.review_result),
not_(Message.deleted),
Message.review_count < self.cfg.num_reviews_reply,
Message.parent_id.is_not(None),
)
)
return [x["id"] for x in r.all()]
if not settings.DEBUG_ALLOW_SELF_LABELING:
qry = qry.filter(Message.user_id != self.pr.user_id)
return qry.all()
_sql_find_incomplete_rankings = """
-- find incomplete rankings
SELECT m.parent_id, COUNT(m.id) children_count, MIN(m.ranking_count) child_min_ranking_count,
SELECT m.parent_id, m.role, COUNT(m.id) children_count, MIN(m.ranking_count) child_min_ranking_count,
COUNT(m.id) FILTER (WHERE m.ranking_count >= :num_required_rankings) as completed_rankings
FROM message_tree_state mts
LEFT JOIN message m ON mts.message_tree_id = m.message_tree_id
@@ -598,7 +670,7 @@ WHERE mts.active -- only consider active trees
AND m.review_result -- must be reviewed
AND NOT m.deleted -- not deleted
AND m.parent_id IS NOT NULL -- ignore initial prompts
GROUP BY m.parent_id
GROUP BY m.parent_id, m.role
HAVING COUNT(m.id) > 1 and MIN(m.ranking_count) < :num_required_rankings
"""
@@ -616,10 +688,10 @@ HAVING COUNT(m.id) > 1 and MIN(m.ranking_count) < :num_required_rankings
_sql_find_extendible_parents = """
-- find all extendible parent nodes
SELECT m.id as parent_id, m.depth, m.message_tree_id, COUNT(c.id) active_children_count
SELECT m.id as parent_id, m.role as parent_role, m.depth, m.message_tree_id, COUNT(c.id) active_children_count
FROM message_tree_state mts
LEFT JOIN message m ON mts.message_tree_id = m.message_tree_id -- all elements of message tree
LEFT JOIN message c ON m.id = c.Id -- child nodes
LEFT JOIN message c ON m.id = c.parent_id -- child nodes
WHERE mts.active -- only consider active trees
AND mts.state = :growing_state -- message tree must be growing
AND NOT m.deleted -- ignore deleted messages as parents
@@ -627,7 +699,7 @@ WHERE mts.active -- only consider active trees
AND m.review_result -- parent node must have positive review
AND NOT c.deleted -- don't count deleted children
AND (c.review_result OR c.review_count < :num_reviews_reply) -- don't count children with negative review but count elements under review
GROUP BY m.id, m.depth, m.message_tree_id, mts.max_children_count
GROUP BY m.id, m.role, m.depth, m.message_tree_id, mts.max_children_count
HAVING COUNT(c.id) < mts.max_children_count -- below maximum number of children
"""
@@ -636,10 +708,7 @@ HAVING COUNT(c.id) < mts.max_children_count -- below maximum number of children
r = self.db.execute(
text(self._sql_find_extendible_parents),
{
"growing_state": message_tree_state.State.GROWING,
"num_reviews_reply": self.cfg.num_reviews_reply,
},
{"growing_state": message_tree_state.State.GROWING, "num_reviews_reply": self.cfg.num_reviews_reply},
)
return [ExtendibleParentRow.from_orm(x) for x in r.all()]
@@ -671,21 +740,27 @@ HAVING COUNT(m.id) < mts.goal_tree_size
)
return [ActiveTreeSizeRow.from_orm(x) for x in r.all()]
_sql_get_tree_size = """
SELECT mts.message_tree_id, mts.goal_tree_size, COUNT(m.id) AS tree_size
FROM message_tree_state mts
LEFT JOIN message m ON mts.message_tree_id = m.message_tree_id
WHERE mts.active
AND NOT m.deleted
AND m.review_result
AND mts.message_tree_id = :message_tree_id
GROUP BY mts.message_tree_id, mts.goal_tree_size
"""
def query_tree_size(self, message_tree_id: UUID) -> ActiveTreeSizeRow:
"""Returns the number of reviewed not deleted messages in the message tree."""
r = self.db.execute(text(self._sql_get_tree_size), {"message_tree_id": message_tree_id})
return ActiveTreeSizeRow.from_orm(r.one())
qry = (
self.db.query(
MessageTreeState.message_tree_id.label("message_tree_id"),
MessageTreeState.goal_tree_size.label("goal_tree_size"),
func.count(Message.id).label("tree_size"),
)
.select_from(MessageTreeState)
.outerjoin(Message, MessageTreeState.message_tree_id == Message.message_tree_id)
.filter(
MessageTreeState.active,
not_(Message.deleted),
Message.review_result,
MessageTreeState.message_tree_id == message_tree_id,
)
.group_by(MessageTreeState.message_tree_id, MessageTreeState.goal_tree_size)
)
return ActiveTreeSizeRow.from_orm(qry.one())
def query_misssing_tree_states(self) -> list[UUID]:
"""Find all initial prompt messages that have no associated message tree state"""
@@ -702,7 +777,7 @@ GROUP BY mts.message_tree_id, mts.goal_tree_size
return [m.id for m in qry_missing_tree_states.all()]
_sql_find_tree_ranking_results = """
-- get all ranking results of completed tasks for all parents with >=2 children
-- get all ranking results of completed tasks for all parents with >= 2 children
SELECT p.parent_id, mr.* FROM
(
-- find parents with > 1 children
@@ -712,7 +787,8 @@ SELECT p.parent_id, mr.* FROM
WHERE m.review_result -- must be reviewed
AND NOT m.deleted -- not deleted
AND m.parent_id IS NOT NULL -- ignore initial prompts
AND mts.message_tree_id = :message_tree_id
AND (:role IS NULL OR m.role = :role) -- children with matching role
AND mts.message_tree_id = :message_tree_id
GROUP BY m.parent_id, m.message_tree_id
HAVING COUNT(m.id) > 1
) as p
@@ -720,11 +796,21 @@ LEFT JOIN task t ON p.parent_id = t.parent_message_id AND t.done AND (t.payload_
LEFT JOIN message_reaction mr ON mr.task_id = t.id AND mr.payload_type = 'RankingReactionPayload'
"""
def query_tree_ranking_results(self, message_tree_id: UUID) -> dict[UUID, list[MessageReaction]]:
def query_tree_ranking_results(
self,
message_tree_id: UUID,
role_filter: str = "assistant",
) -> dict[UUID, list[MessageReaction]]:
"""Finds all completed ranking restuls for a message_tree"""
assert role_filter in (None, "assistant", "prompter")
r = self.db.execute(
text(self._sql_find_tree_ranking_results),
{"message_tree_id": message_tree_id},
{
"message_tree_id": message_tree_id,
"role": role_filter,
},
)
rankings_by_message = {}
@@ -754,15 +840,13 @@ LEFT JOIN message_reaction mr ON mr.task_id = t.id AND mr.payload_type = 'Rankin
return query.scalar()
def query_reviews_for_message(self, message_id: UUID) -> list[TextLabels]:
sql_qry = """
SELECT tl.*
FROM task t
INNER JOIN text_labels tl ON tl.id = t.id
WHERE t.done = TRUE
AND tl.message_id = :message_id
"""
r = self.db.execute(text(sql_qry), {"message_id": message_id})
return [TextLabels.from_orm(x) for x in r.all()]
qry = (
self.db.query(TextLabels)
.select_from(Task)
.join(TextLabels, Task.id == TextLabels.id)
.filter(Task.done, TextLabels.message_id == message_id)
)
return qry.all()
@managed_tx_method(CommitMode.FLUSH)
def _insert_tree_state(
@@ -803,12 +887,12 @@ WHERE t.done = TRUE
if __name__ == "__main__":
from oasst_backend.api.deps import get_dummy_api_client
from oasst_backend.api.deps import api_auth
from oasst_backend.database import engine
from oasst_backend.prompt_repository import PromptRepository
with Session(engine) as db:
api_client = get_dummy_api_client(db)
api_client = api_auth(settings.OFFICIAL_WEB_API_KEY, db=db)
dummy_user = protocol_schema.User(id="__dummy_user__", display_name="Dummy User", auth_method="local")
pr = PromptRepository(db=db, api_client=api_client, client_user=dummy_user)
@@ -817,15 +901,21 @@ if __name__ == "__main__":
tm = TreeManager(db, pr, cfg)
tm.ensure_tree_states()
print("query_num_active_trees", tm.query_num_active_trees())
print("query_incomplete_rankings", tm.query_incomplete_rankings())
print("query_incomplete_reply_reviews", tm.query_replies_need_review())
print("query_incomplete_initial_prompt_reviews", tm.query_prompts_need_review())
print("query_extendible_trees", tm.query_extendible_trees())
print("query_extendible_parents", tm.query_extendible_parents())
print("next_task:", tm.next_task())
# print("query_num_active_trees", tm.query_num_active_trees())
# print("query_incomplete_rankings", tm.query_incomplete_rankings())
# print("query_replies_need_review", tm.query_replies_need_review())
# print("query_incomplete_initial_prompt_reviews", tm.query_prompts_need_review())
# print("query_extendible_trees", tm.query_extendible_trees())
# print("query_extendible_parents", tm.query_extendible_parents())
# print("query_tree_size", tm.query_tree_size(message_tree_id=UUID("bdf434cf-4df5-4b74-949c-a5a157bc3292")))
print(
".query_tree_ranking_results", tm.query_tree_ranking_results(UUID("2ac20d38-6650-43aa-8bb3-f61080c0d921"))
"query_reviews_for_message",
tm.query_reviews_for_message(message_id=UUID("6a444493-0d48-4316-a9f1-7e263f5a2473")),
)
# print("next_task:", tm.next_task())
# print(
# "query_tree_ranking_results", tm.query_tree_ranking_results(UUID("6036f58f-41b5-48c4-bdd9-b16f34ab1312"))
# )
backend/oasst_backend/utils/database_utils.py
+55 -11
View File
@@ -1,13 +1,14 @@
from enum import IntEnum
from functools import wraps
from http import HTTPStatus
from typing import Callable
from loguru import logger
from oasst_backend.config import settings
from oasst_backend.database import engine
from oasst_shared.exceptions import OasstError, OasstErrorCode
from sqlalchemy.exc import OperationalError
from sqlmodel import SQLModel
MAX_DB_RETRY_COUNT = 3
from sqlmodel import Session, SQLModel
class CommitMode(IntEnum):
@@ -28,7 +29,7 @@ class CommitMode(IntEnum):
"""
def managed_tx_method(auto_commit: CommitMode = CommitMode.COMMIT, num_retries=MAX_DB_RETRY_COUNT):
def managed_tx_method(auto_commit: CommitMode = CommitMode.COMMIT, num_retries=settings.DATABASE_MAX_TX_RETRY_COUNT):
def decorator(f):
@wraps(f)
def wrapped_f(self, *args, **kwargs):
@@ -44,16 +45,15 @@ def managed_tx_method(auto_commit: CommitMode = CommitMode.COMMIT, num_retries=M
self.db.refresh(result)
return result
except OperationalError:
logger.info(f"Retrying count: {i+1} after possible db concurrent update conflict")
logger.info(f"Retry {i+1}/{num_retries} after possible DB concurrent update conflict.")
self.db.rollback()
pass
raise OasstError(
"DATABASE_MAX_RETIRES_EXHAUSTED",
error_code=OasstErrorCode.DATABASE_MAX_RETRIES_EXHAUSTED,
http_status_code=HTTPStatus.SERVICE_UNAVAILABLE,
)
except Exception as e:
logger.error("Db Rollback Failure")
logger.error("DB Rollback Failure")
raise e
return wrapped_f
@@ -61,7 +61,9 @@ def managed_tx_method(auto_commit: CommitMode = CommitMode.COMMIT, num_retries=M
return decorator
def async_managed_tx_method(auto_commit: CommitMode = CommitMode.COMMIT, num_retries=MAX_DB_RETRY_COUNT):
def async_managed_tx_method(
auto_commit: CommitMode = CommitMode.COMMIT, num_retries=settings.DATABASE_MAX_TX_RETRY_COUNT
):
def decorator(f):
@wraps(f)
async def wrapped_f(self, *args, **kwargs):
@@ -77,16 +79,58 @@ def async_managed_tx_method(auto_commit: CommitMode = CommitMode.COMMIT, num_ret
self.db.refresh(result)
return result
except OperationalError:
logger.info(f"Retrying count: {i+1} after possible db concurrent update conflict")
logger.info(f"Retry {i+1}/{num_retries} after possible DB concurrent update conflict.")
self.db.rollback()
pass
raise OasstError(
"DATABASE_MAX_RETIRES_EXHAUSTED",
error_code=OasstErrorCode.DATABASE_MAX_RETRIES_EXHAUSTED,
http_status_code=HTTPStatus.SERVICE_UNAVAILABLE,
)
except Exception as e:
logger.error("Db Rollback Failure")
logger.exception("DB Rollback Failure")
raise e
return wrapped_f
return decorator
def default_session_factor() -> Session:
return Session(engine)
def managed_tx_function(
auto_commit: CommitMode = CommitMode.COMMIT,
num_retries=settings.DATABASE_MAX_TX_RETRY_COUNT,
session_factory: Callable[..., Session] = default_session_factor,
):
"""Passes Session object as first argument to wrapped function."""
def decorator(f):
@wraps(f)
def wrapped_f(*args, **kwargs):
try:
for i in range(num_retries):
with session_factory() as session:
try:
result = f(session, *args, **kwargs)
if auto_commit == CommitMode.COMMIT:
session.commit()
elif auto_commit == CommitMode.FLUSH:
session.flush()
if isinstance(result, SQLModel):
session.refresh(result)
return result
except OperationalError:
logger.info(f"Retry {i+1}/{num_retries} after possible DB concurrent update conflict.")
session.rollback()
raise OasstError(
"DATABASE_MAX_RETIRES_EXHAUSTED",
error_code=OasstErrorCode.DATABASE_MAX_RETRIES_EXHAUSTED,
http_status_code=HTTPStatus.SERVICE_UNAVAILABLE,
)
except Exception as e:
logger.error("DB Rollback Failure")
raise e
return wrapped_f
deploy/prod-node/nginx/nginx.conf
+16 -1
View File
@@ -16,6 +16,19 @@ http {
}
}
server {
listen 443 ssl http2;
server_name www.open-assistant.io;
ssl_certificate /etc/nginx/ssl/live/www.open-assistant.io/fullchain.pem;
ssl_certificate_key /etc/nginx/ssl/live/www.open-assistant.io/privkey.pem;
location / {
return 301 https://open-assistant.io$request_uri;
}
}
server {
listen 443 ssl http2;
@@ -25,7 +38,9 @@ http {
ssl_certificate_key /etc/nginx/ssl/live/open-assistant.io/privkey.pem;
location / {
return 301 https://web.prod.open-assistant.io$request_uri;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_pass http://127.0.0.1:3200;
}
}
docs/docs/research/general.md
+124 -8
View File
@@ -8,15 +8,29 @@ This page lists research papers that are relevant to the project.
- Generating Text From Language Models
- Automatically Generating Instruction Data for Training
- Uncertainty Estimation of Language Model Outputs
- Evidence-Guided Text Generation
- Reward Model Optimization
- Dialogue-Oriented RLHF
- Reduce Harms in Language Models
## Reinforcement Learning from Human Feedback <a name="reinforcement-learning-from-human-feedback"></a>
## Reinforcement Learning from Human Feedback
Reinforcement Learning from Human Feedback (RLHF) is a method for fine-tuning a
generative language models based on a reward model that is learned from human
preference data. This method facilitates the learning of instruction-tuned
models, among other things.
### Learning to summarize from human feedback [[ArXiv](https://arxiv.org/pdf/2009.01325.pdf)], [[Github](https://github.com/openai/summarize-from-feedback)]
### Fine-Tuning Language Models from Human Preferences [[ArXiv](https://arxiv.org/abs/1909.08593)], [[GitHub](https://github.com/openai/lm-human-preferences)]
> In this paper, we build on advances in generative pretraining of language
> models to apply reward learning to four natural language tasks: continuing
> text with positive sentiment or physically descriptive language, and
> summarization tasks on the TL;DR and CNN/Daily Mail datasets. For stylistic
> continuation we achieve good results with only 5,000 comparisons evaluated by
> humans. For summarization, models trained with 60,000 comparisons copy whole
> sentences from the input but skip irrelevant preamble.
### Learning to summarize from human feedback [[ArXiv](https://arxiv.org/abs/2009.01325)], [[GitHub](https://github.com/openai/summarize-from-feedback)]
> In this work, we show that it is possible to significantly improve summary
> quality by training a model to optimize for human preferences. We collect a
@@ -24,7 +38,18 @@ models, among other things.
> model to predict the human-preferred summary, and use that model as a reward
> function to fine-tune a summarization policy using reinforcement learning.
### Training language models to follow instructions with human feedback [[ArXiv](https://arxiv.org/pdf/2203.02155.pdf)]
### Recursively Summarizing Books with Human Feedback [[ArXiv](https://arxiv.org/abs/2109.10862)]
> Our method combines learning from human feedback with recursive task
> decomposition: we use models trained on smaller parts of the task to assist
> humans in giving feedback on the broader task. We collect a large volume of
> demonstrations and comparisons from human labelers. Our resulting model
> generates sensible summaries of entire books, even matching the quality of
> human-written summaries in a few cases (∼5% of books). We achieve
> state-of-the-art results on the recent BookSum dataset for book-length
> summarization. We release datasets of samples from our model.
### Training language models to follow instructions with human feedback [[ArXiv](https://arxiv.org/abs/2203.02155)]
> Starting with a set of labeler-written prompts and prompts submitted through
> the OpenAI API, we collect a dataset of labeler demonstrations of the desired
@@ -33,7 +58,7 @@ models, among other things.
> fine-tune this supervised model using reinforcement learning from human
> feedback.
### Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback [[ArXiv](https://arxiv.org/pdf/2204.05862.pdf)]
### Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback [[ArXiv](https://arxiv.org/abs/2204.05862)]
> We apply preference modeling and reinforcement learning from human feedback
> (RLHF) to finetune language models to act as helpful and harmless assistants.
@@ -41,6 +66,31 @@ models, among other things.
> evaluations, and is fully compatible with training for specialized skills such
> as python coding and summarization.
### Self-critiquing models for assisting human evaluators [[ArXiv](https://arxiv.org/abs/2206.05802)]
> We fine-tune large language models to write natural language critiques
> (natural language critical comments) using behavioral cloning. On a
> topic-based summarization task, critiques written by our models help humans
> find flaws in summaries that they would have otherwise missed. We study
> scaling properties of critiquing with both topic-based summarization and
> synthetic tasks. Finally, we motivate and introduce a framework for comparing
> critiquing ability to generation and discrimination ability. These results are
> a proof of concept for using AI-assisted human feedback to scale the
> supervision of machine learning systems to tasks that are difficult for humans
> to evaluate directly. We release our training datasets.
### Is Reinforcement Learning (Not) for Natural Language Processing?: Benchmarks, Baselines, and Building Blocks for Natural Language Policy Optimization [[ArXiv](https://arxiv.org/abs/2210.01241)]
> We tackle the problem of aligning pre-trained large language models (LMs) with
> human preferences. We present the GRUE (General Reinforced-language
> Understanding Evaluation) benchmark, a set of 6 language generation tasks
> which are supervised by reward functions which capture automated measures of
> human preference. Finally, we introduce an easy-to-use, performant RL
> algorithm, NLPO (Natural Language Policy Optimization) that learns to
> effectively reduce the combinatorial action space in language generation. We
> show that RL techniques are generally better than supervised methods at
> aligning LMs to human preferences.
## Generating Text From Language Models
A language model generates output text token by token, autoregressively. The
@@ -48,7 +98,7 @@ large search space of this task requires some method of narrowing down the set
of tokens to be considered in each step. This method, in turn, has a big impact
on the quality of the resulting text.
### RANKGEN: Improving Text Generation with Large Ranking Models [[ArXiv](https://arxiv.org/pdf/2205.09726.pdf)], [[Github](https://github.com/martiansideofthemoon/rankgen)]
### RANKGEN: Improving Text Generation with Large Ranking Models [[ArXiv](https://arxiv.org/abs/2205.09726)], [[GitHub](https://github.com/martiansideofthemoon/rankgen)]
> Given an input sequence (or prefix), modern language models often assign high
> probabilities to output sequences that are repetitive, incoherent, or
@@ -65,7 +115,7 @@ annotated data for the purpose of training
[instruction-aligned](https://openai.com/blog/instruction-following/) language
models.
### SELF-INSTRUCT: Aligning Language Model with Self Generated Instructions [[ArXiv](https://arxiv.org/pdf/2212.10560.pdf)], [[Github](https://github.com/yizhongw/self-instruct)].
### SELF-INSTRUCT: Aligning Language Model with Self Generated Instructions [[ArXiv](https://arxiv.org/abs/2212.10560)], [[GitHub](https://github.com/yizhongw/self-instruct)].
> We introduce SELF-INSTRUCT, a framework for improving the
> instruction-following capabilities of pretrained language models by
@@ -76,7 +126,7 @@ models.
> SuperNaturalInstructions, on par with the performance of InstructGPT-0011,
> which is trained with private user data and human annotations.
### Tuning Language Models with (Almost) No Human Labor. [[ArXiv](https://arxiv.org/pdf/2212.09689.pdf)], [[Github](https://github.com/orhonovich/unnatural-instructions)].
### Tuning Language Models with (Almost) No Human Labor. [[ArXiv](https://arxiv.org/abs/2212.09689)], [[GitHub](https://github.com/orhonovich/unnatural-instructions)].
> In this work, we introduce Unnatural Instructions: a large dataset of creative
> and diverse instructions, collected with virtually no human labor. We collect
@@ -91,7 +141,7 @@ models.
## Uncertainty Estimation of Language Model Outputs
### Teaching models to express their uncertainty in words [[Arxiv](https://arxiv.org/pdf/2205.14334.pdf)]
### Teaching models to express their uncertainty in words [[ArXiv](https://arxiv.org/abs/2205.14334)]
> We show that a GPT-3 model can learn to express uncertainty about its own
> answers in natural language -- without use of model logits. When given a
@@ -100,3 +150,69 @@ models.
> are well calibrated. The model also remains moderately calibrated under
> distribution shift, and is sensitive to uncertainty in its own answers, rather
> than imitating human examples.
## Evidence-Guided Text Generation
### WebGPT: Browser-assisted question-answering with human feedback [[ArXiv](https://arxiv.org/abs/2112.09332)]
> We fine-tune GPT-3 to answer long-form questions using a text-based
> web-browsing environment, which allows the model to search and navigate the
> web. We are able to train models on the task using imitation learning, and
> then optimize answer quality with human feedback. Models must collect
> references while browsing in support of their answers. Our best model is
> obtained by fine-tuning GPT-3 using behavior cloning, and then performing
> rejection sampling against a reward model.
### Teaching language models to support answers with verified quotes [[ArXiv](https://arxiv.org/abs/2203.11147)]
> In this work we use RLHF to train "open-book" QA models that generate answers
> whilst also citing specific evidence for their claims, which aids in the
> appraisal of correctness. Supporting evidence is drawn from multiple documents
> found via a search engine, or from a single user-provided document. However,
> analysis on the adversarial TruthfulQA dataset shows why citation is only one
> part of an overall strategy for safety and trustworthiness: not all claims
> supported by evidence are true.
## Reward Model Optimization
### Scaling Laws for Reward Model Overoptimization [[ArXiv](https://arxiv.org/abs/2210.10760)], [[Preceding Blogpost](https://openai.com/blog/measuring-goodharts-law/)]
> In this work, we use a synthetic setup in which a fixed "gold-standard" reward
> model plays the role of humans, providing labels used to train a proxy reward
> model. We study how the gold reward model score changes as we optimize against
> the proxy reward model using either reinforcement learning or best-of-n
> sampling. We study the effect on this relationship of the size of the reward
> model dataset. We explore the implications of these empirical results for
> theoretical considerations in AI alignment.
## Dialogue-Oriented RLHF
### Dynamic Planning in Open-Ended Dialogue using Reinforcement Learning [[ArXiv](https://arxiv.org/abs/2208.02294)]
> Building automated agents that can carry on rich open-ended conversations with
> humans "in the wild" remains a formidable challenge. In this work we develop a
> real-time, open-ended dialogue system that uses reinforcement learning (RL) to
> power a bot's conversational skill at scale. Trained using crowd-sourced data,
> our novel system is able to substantially exceeds several metrics of interest
> in a live experiment with real users of the Google Assistant.
### Improving alignment of dialogue agents via targeted human judgements [[ArXiv](https://arxiv.org/abs/2209.14375)]
> We present Sparrow, an information-seeking dialogue agent trained to be more
> helpful, correct, and harmless compared to prompted language model baselines
> First, to make our agent more helpful and harmless, we break down the
> requirements for good dialogue into natural language rules the agent should
> followy. Second, our agent provides evidence from sources supporting factual
> claims when collecting preference judgements over model statements.Finally, we
> conduct extensive analyses showing that though our model learns to follow our
> rules it can exhibit distributional biases.
## Reduce Harms in Language Models
### Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned [[ArXiv](https://arxiv.org/abs/2209.07858)]
> We investigate scaling behaviors for red teaming. We find that the RLHF models
> are increasingly difficult to red team as they scale, and we find a flat trend
> with scale for the other model types. We exhaustively describe our
> instructions, processes, statistical methodologies, and uncertainty about red
> teaming.
oasst-shared/oasst_shared/exceptions/oasst_api_error.py
+1
View File
@@ -32,6 +32,7 @@ class OasstErrorCode(IntEnum):
TASK_INTERACTION_REQUEST_FAILED = 1004
TASK_GENERATION_FAILED = 1005
TASK_REQUESTED_TYPE_NOT_AVAILABLE = 1006
TASK_AVAILABILITY_QUERY_FAILED = 1007
# 2000-3000: prompt_repository
INVALID_FRONTEND_MESSAGE_ID = 2000
website/package-lock.json
Generated
+26 -95
View File
@@ -29,7 +29,6 @@
"eslint-config-next": "13.0.6",
"eslint-plugin-simple-import-sort": "^8.0.0",
"focus-visible": "^5.2.0",
"formik": "^2.2.9",
"framer-motion": "^6.5.1",
"install": "^0.13.0",
"next": "13.0.6",
@@ -40,6 +39,7 @@
"react": "18.2.0",
"react-dom": "18.2.0",
"react-feature-flags": "^1.0.0",
"react-hook-form": "^7.42.1",
"react-icons": "^4.7.1",
"react-table": "^7.8.0",
"sharp": "^0.31.3",
@@ -20304,47 +20304,6 @@
"node": ">= 6"
}
},
"node_modules/formik": {
"version": "2.2.9",
"resolved": "https://registry.npmjs.org/formik/-/formik-2.2.9.tgz",
"integrity": "sha512-LQLcISMmf1r5at4/gyJigGn0gOwFbeEAlji+N9InZF6LIMXnFNkO42sCI8Jt84YZggpD4cPWObAZaxpEFtSzNA==",
"funding": [
{
"type": "individual",
"url": "https://opencollective.com/formik"
}
],
"dependencies": {
"deepmerge": "^2.1.1",
"hoist-non-react-statics": "^3.3.0",
"lodash": "^4.17.21",
"lodash-es": "^4.17.21",
"react-fast-compare": "^2.0.1",
"tiny-warning": "^1.0.2",
"tslib": "^1.10.0"
},
"peerDependencies": {
"react": ">=16.8.0"
}
},
"node_modules/formik/node_modules/deepmerge": {
"version": "2.2.1",
"resolved": "https://registry.npmjs.org/deepmerge/-/deepmerge-2.2.1.tgz",
"integrity": "sha512-R9hc1Xa/NOBi9WRVUWg19rl1UB7Tt4kuPd+thNJgFZoxXsTz7ncaPaeIm+40oSGuP33DfMb4sZt1QIGiJzC4EA==",
"engines": {
"node": ">=0.10.0"
}
},
"node_modules/formik/node_modules/react-fast-compare": {
"version": "2.0.4",
"resolved": "https://registry.npmjs.org/react-fast-compare/-/react-fast-compare-2.0.4.tgz",
"integrity": "sha512-suNP+J1VU1MWFKcyt7RtjiSWUjvidmQSlqu+eHslq+342xCbGTYmC0mEhPCOHxlW0CywylOC1u2DFAT+bv4dBw=="
},
"node_modules/formik/node_modules/tslib": {
"version": "1.14.1",
"resolved": "https://registry.npmjs.org/tslib/-/tslib-1.14.1.tgz",
"integrity": "sha512-Xni35NKzjgMrwevysHTCArtLDpPvye8zV/0E4EyYn43P7/7qvQwPh9BGkHewbMulVntbigmcT7rdX3BNo9wRJg=="
},
"node_modules/forwarded": {
"version": "0.2.0",
"resolved": "https://registry.npmjs.org/forwarded/-/forwarded-0.2.0.tgz",
@@ -26443,12 +26402,8 @@
"node_modules/lodash": {
"version": "4.17.21",
"resolved": "https://registry.npmjs.org/lodash/-/lodash-4.17.21.tgz",
"integrity": "sha512-v2kDEe57lecTulaDIuNTPy3Ry4gLGJ6Z1O3vE1krgXZNrsQ+LFTGHVxVjcXPs17LhbZVGedAJv8XZ1tvj5FvSg=="
},
"node_modules/lodash-es": {
"version": "4.17.21",
"resolved": "https://registry.npmjs.org/lodash-es/-/lodash-es-4.17.21.tgz",
"integrity": "sha512-mKnC+QJ9pWVzv+C4/U3rRsHapFfHvQFoFB92e52xeyGMcX6/OlIl78je1u8vePzYZSkkogMPJ2yjxxsb89cxyw=="
"integrity": "sha512-v2kDEe57lecTulaDIuNTPy3Ry4gLGJ6Z1O3vE1krgXZNrsQ+LFTGHVxVjcXPs17LhbZVGedAJv8XZ1tvj5FvSg==",
"dev": true
},
"node_modules/lodash.debounce": {
"version": "4.0.8",
@@ -32527,6 +32482,21 @@
"node": ">=10"
}
},
"node_modules/react-hook-form": {
"version": "7.42.1",
"resolved": "https://registry.npmjs.org/react-hook-form/-/react-hook-form-7.42.1.tgz",
"integrity": "sha512-2UIGqwMZksd5HS55crTT1ATLTr0rAI4jS7yVuqTaoRVDhY2Qc4IyjskCmpnmdYqUNOYFy04vW253tb2JRVh+IQ==",
"engines": {
"node": ">=12.22.0"
},
"funding": {
"type": "opencollective",
"url": "https://opencollective.com/react-hook-form"
},
"peerDependencies": {
"react": "^16.8.0 || ^17 || ^18"
}
},
"node_modules/react-icons": {
"version": "4.7.1",
"resolved": "https://registry.npmjs.org/react-icons/-/react-icons-4.7.1.tgz",
@@ -35486,11 +35456,6 @@
"resolved": "https://registry.npmjs.org/tiny-invariant/-/tiny-invariant-1.3.1.tgz",
"integrity": "sha512-AD5ih2NlSssTCwsMznbvwMZpJ1cbhkGd2uueNxzv2jDlEeZdU04JQfRnggJQ8DrcVBGjAsCKwFBbDlVNtEMlzw=="
},
"node_modules/tiny-warning": {
"version": "1.0.3",
"resolved": "https://registry.npmjs.org/tiny-warning/-/tiny-warning-1.0.3.tgz",
"integrity": "sha512-lBN9zLN/oAf68o3zNXYrdCt1kP8WsiGW8Oo2ka41b2IM5JL/S1CTyX1rW0mb/zSuJun0ZUrDxx4sqvYS2FWzPA=="
},
"node_modules/tmp": {
"version": "0.2.1",
"resolved": "https://registry.npmjs.org/tmp/-/tmp-0.2.1.tgz",
@@ -52929,37 +52894,6 @@
"mime-types": "^2.1.12"
}
},
"formik": {
"version": "2.2.9",
"resolved": "https://registry.npmjs.org/formik/-/formik-2.2.9.tgz",
"integrity": "sha512-LQLcISMmf1r5at4/gyJigGn0gOwFbeEAlji+N9InZF6LIMXnFNkO42sCI8Jt84YZggpD4cPWObAZaxpEFtSzNA==",
"requires": {
"deepmerge": "^2.1.1",
"hoist-non-react-statics": "^3.3.0",
"lodash": "^4.17.21",
"lodash-es": "^4.17.21",
"react-fast-compare": "^2.0.1",
"tiny-warning": "^1.0.2",
"tslib": "^1.10.0"
},
"dependencies": {
"deepmerge": {
"version": "2.2.1",
"resolved": "https://registry.npmjs.org/deepmerge/-/deepmerge-2.2.1.tgz",
"integrity": "sha512-R9hc1Xa/NOBi9WRVUWg19rl1UB7Tt4kuPd+thNJgFZoxXsTz7ncaPaeIm+40oSGuP33DfMb4sZt1QIGiJzC4EA=="
},
"react-fast-compare": {
"version": "2.0.4",
"resolved": "https://registry.npmjs.org/react-fast-compare/-/react-fast-compare-2.0.4.tgz",
"integrity": "sha512-suNP+J1VU1MWFKcyt7RtjiSWUjvidmQSlqu+eHslq+342xCbGTYmC0mEhPCOHxlW0CywylOC1u2DFAT+bv4dBw=="
},
"tslib": {
"version": "1.14.1",
"resolved": "https://registry.npmjs.org/tslib/-/tslib-1.14.1.tgz",
"integrity": "sha512-Xni35NKzjgMrwevysHTCArtLDpPvye8zV/0E4EyYn43P7/7qvQwPh9BGkHewbMulVntbigmcT7rdX3BNo9wRJg=="
}
}
},
"forwarded": {
"version": "0.2.0",
"resolved": "https://registry.npmjs.org/forwarded/-/forwarded-0.2.0.tgz",
@@ -57574,12 +57508,8 @@
"lodash": {
"version": "4.17.21",
"resolved": "https://registry.npmjs.org/lodash/-/lodash-4.17.21.tgz",
"integrity": "sha512-v2kDEe57lecTulaDIuNTPy3Ry4gLGJ6Z1O3vE1krgXZNrsQ+LFTGHVxVjcXPs17LhbZVGedAJv8XZ1tvj5FvSg=="
},
"lodash-es": {
"version": "4.17.21",
"resolved": "https://registry.npmjs.org/lodash-es/-/lodash-es-4.17.21.tgz",
"integrity": "sha512-mKnC+QJ9pWVzv+C4/U3rRsHapFfHvQFoFB92e52xeyGMcX6/OlIl78je1u8vePzYZSkkogMPJ2yjxxsb89cxyw=="
"integrity": "sha512-v2kDEe57lecTulaDIuNTPy3Ry4gLGJ6Z1O3vE1krgXZNrsQ+LFTGHVxVjcXPs17LhbZVGedAJv8XZ1tvj5FvSg==",
"dev": true
},
"lodash.debounce": {
"version": "4.0.8",
@@ -61952,6 +61882,12 @@
}
}
},
"react-hook-form": {
"version": "7.42.1",
"resolved": "https://registry.npmjs.org/react-hook-form/-/react-hook-form-7.42.1.tgz",
"integrity": "sha512-2UIGqwMZksd5HS55crTT1ATLTr0rAI4jS7yVuqTaoRVDhY2Qc4IyjskCmpnmdYqUNOYFy04vW253tb2JRVh+IQ==",
"requires": {}
},
"react-icons": {
"version": "4.7.1",
"resolved": "https://registry.npmjs.org/react-icons/-/react-icons-4.7.1.tgz",
@@ -64265,11 +64201,6 @@
"resolved": "https://registry.npmjs.org/tiny-invariant/-/tiny-invariant-1.3.1.tgz",
"integrity": "sha512-AD5ih2NlSssTCwsMznbvwMZpJ1cbhkGd2uueNxzv2jDlEeZdU04JQfRnggJQ8DrcVBGjAsCKwFBbDlVNtEMlzw=="
},
"tiny-warning": {
"version": "1.0.3",
"resolved": "https://registry.npmjs.org/tiny-warning/-/tiny-warning-1.0.3.tgz",
"integrity": "sha512-lBN9zLN/oAf68o3zNXYrdCt1kP8WsiGW8Oo2ka41b2IM5JL/S1CTyX1rW0mb/zSuJun0ZUrDxx4sqvYS2FWzPA=="
},
"tmp": {
"version": "0.2.1",
"resolved": "https://registry.npmjs.org/tmp/-/tmp-0.2.1.tgz",
website/package.json
+1 -1
View File
@@ -46,7 +46,6 @@
"eslint-config-next": "13.0.6",
"eslint-plugin-simple-import-sort": "^8.0.0",
"focus-visible": "^5.2.0",
"formik": "^2.2.9",
"framer-motion": "^6.5.1",
"install": "^0.13.0",
"next": "13.0.6",
@@ -57,6 +56,7 @@
"react": "18.2.0",
"react-dom": "18.2.0",
"react-feature-flags": "^1.0.0",
"react-hook-form": "^7.42.1",
"react-icons": "^4.7.1",
"react-table": "^7.8.0",
"sharp": "^0.31.3",
website/src/components/CollapsableText.tsx
+5 -2
View File
@@ -11,6 +11,8 @@ import {
} from "@chakra-ui/react";
import React, { ReactNode } from "react";
const killEvent = (e) => e.stopPropagation();
export const CollapsableText = ({
text,
maxLength = 220,
@@ -44,8 +46,9 @@ export const CollapsableText = ({
</Button>
</span>
<Modal isOpen={isOpen} onClose={onClose} size="xl" scrollBehavior={"inside"}>
<ModalOverlay style={{ width: "100%", height: "100%" }}>
<ModalContent maxH="400">
{/* we kill the event here to disable drag and drop, since it is in the same container */}
<ModalOverlay onMouseDown={killEvent}>
<ModalContent alignItems="center">
<ModalHeader>Full Text</ModalHeader>
<ModalCloseButton />
<ModalBody>{text}</ModalBody>
website/src/components/Dashboard/LeaderboardTable.tsx
+5 -29
View File
@@ -1,17 +1,16 @@
import { Box, Link, Stack, StackDivider, Text, useColorModeValue } from "@chakra-ui/react";
import { Box, Link, Text, useColorModeValue } from "@chakra-ui/react";
import NextLink from "next/link";
import { get } from "src/lib/api";
import useSWR from "swr";
import { LeaderboardGridCell } from "src/components/LeaderboardGridCell";
import { LeaderboardTimeFrame } from "src/types/Leaderboard";
export function LeaderboardTable() {
const backgroundColor = useColorModeValue("white", "gray.700");
const accentColor = useColorModeValue("gray.200", "gray.900");
const { data: leaderboardEntries } = useSWR("/api/leaderboard", get);
return (
<main className="h-fit col-span-3">
<div className="flex flex-col gap-4">
<div className="flex items-end justify-between">
<Text className="text-2xl font-bold">Top 5 Contributors</Text>
<Text className="text-2xl font-bold">Top 5 Contributors Today</Text>
<Link as={NextLink} href="/leaderboard" _hover={{ textDecoration: "none" }}>
<Text color="blue.400" className="text-sm font-bold">
View All -&gt;
@@ -25,30 +24,7 @@ export function LeaderboardTable() {
borderRadius="xl"
className="p-6 shadow-sm"
>
<Stack divider={<StackDivider />} spacing="4">
<div className="grid grid-cols-4 items-center font-bold">
<p>Name</p>
<div className="col-start-4 flex justify-center">
<p>Score</p>
</div>
</div>
{leaderboardEntries?.map(({ display_name, score }, idx) => (
<div key={idx} className="grid grid-cols-4 items-center">
<div className="flex items-center gap-3">
{/*
<Image alt="Profile Picture" src={item.image} boxSize="7" borderRadius="full"></Image>
*/}
<p>{display_name}</p>
{/*
<Badge colorScheme="purple">{item.streakCount}</Badge>
*/}
</div>
<Box bg={backgroundColor} className="col-start-4 flex justify-center">
<p>{score}</p>
</Box>
</div>
))}
</Stack>
<LeaderboardGridCell timeFrame={LeaderboardTimeFrame.day} />
</Box>
</div>
</main>
website/src/components/Dashboard/SlimFooter.tsx
+45
View File
@@ -0,0 +1,45 @@
import { Box, Divider } from "@chakra-ui/react";
import Image from "next/image";
import Link from "next/link";
import { useMemo } from "react";
export function SlimFooter() {
return (
<footer>
<Box>
<Divider />
<Box display="flex" gap="4" flexDir="column" alignItems="center" my="8">
<Box display="flex" alignItems="center">
<Link href="/" aria-label="Home" className="flex items-center gap-1">
<Image src="/images/logos/logo.svg" className="mx-auto object-fill" width="48" height="48" alt="logo" />
</Link>
</Box>
<nav>
<Box display="flex" gap="5" fontSize="xs" color="blue.500">
<FooterLink href="/privacy-policy" label="Privacy Policy" />
<FooterLink href="/terms-of-service" label="Terms of Service" />
<FooterLink href="https://github.com/LAION-AI/Open-Assistant" label="Github" />
<FooterLink href="https://ykilcher.com/open-assistant-discord" label="Discord" />
</Box>
</nav>
</Box>
</Box>
</footer>
);
}
const FooterLink = ({ href, label }: { href: string; label: string }) =>
useMemo(
() => (
<Link
href={href}
rel="noopener noreferrer nofollow"
target="_blank"
aria-label={label}
className="hover:underline underline-offset-2"
>
{label}
</Link>
),
[href, label]
);
website/src/components/Footer.tsx
+57 -35
View File
@@ -1,40 +1,66 @@
import { useColorMode } from "@chakra-ui/react";
import { Box, Divider, Flex, Text, useColorMode } from "@chakra-ui/react";
import Image from "next/image";
import Link from "next/link";
import { useMemo } from "react";
export function Footer() {
const { colorMode } = useColorMode();
const bgColorClass = colorMode === "light" ? "bg-transparent" : "bg-gray-800";
const borderClass = colorMode === "light" ? "border-slate-200" : "border-transparent";
const backgroundColor = colorMode === "light" ? "white" : "gray.800";
const textColor = colorMode === "light" ? "black" : "gray.300";
return (
<footer className={bgColorClass}>
<div className={`flex mx-auto max-w-7xl justify-between border-t p-10 ${borderClass}`}>
<div className="flex items-center pr-8">
<Link href="/" aria-label="Home" className="flex items-center">
<Image src="/images/logos/logo.svg" className="mx-auto object-fill" width="52" height="52" alt="logo" />
</Link>
<footer>
<Box bg={backgroundColor}>
<Divider />
<Box
display="flex"
flexDirection={["column", "row"]}
justifyContent="space-between"
alignItems="center"
gap="6"
p="8"
pb={["14", "8"]}
w="full"
mx="auto"
maxWidth="7xl"
>
<Flex alignItems="center">
<Box pr="2">
<Link href="/" aria-label="Home">
<Image src="/images/logos/logo.svg" width="52" height="52" alt="logo" />
</Link>
</Box>
<div className="ml-2">
<p className="text-base font-bold">Open Assistant</p>
<p className="text-sm">Conversational AI for everyone.</p>
</div>
</div>
<Box>
<Text fontSize="md" fontWeight="bold">
Open Assistant
</Text>
<Text fontSize="sm" color="gray.500">
Conversational AI for everyone.
</Text>
</Box>
</Flex>
<nav className="grid grid-cols-2 gap-20 leading-5 text-sm">
<div className="flex flex-col">
<b className="pb-1">Legal</b>
<FooterLink href="/privacy-policy" label="Privacy Policy" />
<FooterLink href="/terms-of-service" label="Terms of Service" />
</div>
<div className="flex flex-col">
<b className="pb-1">Connect</b>
<FooterLink href="https://github.com/LAION-AI/Open-Assistant" label="Github" />
<FooterLink href="https://ykilcher.com/open-assistant-discord" label="Discord" />
</div>
</nav>
</div>
<nav>
<Box display="flex" flexDirection={["column", "row"]} gap={["6", "14"]} alignItems="center" fontSize="sm">
<Flex direction="column" alignItems={["center", "start"]}>
<Text fontWeight="bold" color={textColor}>
Legal
</Text>
<FooterLink href="/privacy-policy" label="Privacy Policy" />
<FooterLink href="/terms-of-service" label="Terms of Service" />
</Flex>
<Flex direction="column" alignItems={["center", "start"]}>
<Text fontWeight="bold" color={textColor}>
Connect
</Text>
<FooterLink href="https://github.com/LAION-AI/Open-Assistant" label="Github" />
<FooterLink href="https://ykilcher.com/open-assistant-discord" label="Discord" />
</Flex>
</Box>
</nav>
</Box>
</Box>
</footer>
);
}
@@ -42,14 +68,10 @@ export function Footer() {
const FooterLink = ({ href, label }: { href: string; label: string }) =>
useMemo(
() => (
<Link
href={href}
rel="noopener noreferrer nofollow"
target="_blank"
aria-label={label}
className="hover:underline underline-offset-2"
>
{label}
<Link href={href} rel="noopener noreferrer nofollow" target="_blank" aria-label={label}>
<Text color="blue.500" textUnderlineOffset={2} _hover={{ textDecoration: "underline" }}>
{label}
</Text>
</Link>
),
[href, label]
website/src/components/Layout.tsx
+10 -4
View File
@@ -1,9 +1,11 @@
// https://nextjs.org/docs/basic-features/layouts
import { Box, Grid } from "@chakra-ui/react";
import type { NextPage } from "next";
import { FiBarChart2, FiLayout, FiMessageSquare, FiUsers } from "react-icons/fi";
import { Header } from "src/components/Header";
import { SlimFooter } from "./Dashboard/SlimFooter";
import { Footer } from "./Footer";
import { SideMenuLayout } from "./SideMenuLayout";
@@ -28,7 +30,7 @@ export const getTransparentHeaderLayout = (page: React.ReactElement) => (
);
export const getDashboardLayout = (page: React.ReactElement) => (
<div className="grid grid-rows-[min-content_1fr_min-content] h-full justify-items-stretch">
<Grid templateRows="min-content 1fr" h="full">
<Header transparent={true} />
<SideMenuLayout
menuButtonOptions={[
@@ -52,10 +54,14 @@ export const getDashboardLayout = (page: React.ReactElement) => (
},
]}
>
{page}
<Grid templateRows="1fr min-content" h="full">
<Box>{page}</Box>
<Box mt="10">
<SlimFooter />
</Box>
</Grid>
</SideMenuLayout>
<Footer />
</div>
</Grid>
);
export const getAdminLayout = (page: React.ReactElement) => (
website/src/components/LeaderboardGridCell/LeaderboardGridCell.tsx
+1 -1
View File
@@ -8,7 +8,7 @@ import useSWRImmutable from "swr/immutable";
const columns = [
{
Header: "Rank",
accessor: (item: LeaderboardEntity, rowIndex: number) => "#" + item.rank,
accessor: "rank",
style: { width: "90px" },
},
{
website/src/components/Roadmap.tsx
+1 -1
View File
@@ -51,7 +51,7 @@ const Roadmap = () => {
</div>
<h4 className="font-bold text-xl text-[#858585] text-center max-w-[10rem]">Growing Up</h4>
<ul className="ml-6 md:ml-8 lg:ml-6 space-y-4 text-[#858585] list-disc">
<li>Third-Party Extentions</li>
<li>Third-Party Extensions</li>
<li>Device Control</li>
<li>Multi-Modality</li>
</ul>
website/src/components/SideMenuLayout.tsx
+3 -3
View File
@@ -12,11 +12,11 @@ export const SideMenuLayout = (props: SideMenuLayoutProps) => {
return (
<Box backgroundColor={colorMode === "light" ? colors.light.bg : colors.dark.bg} className="sm:overflow-hidden">
<Box className="sm:flex h-full lg:gap-6">
<Box className="p-3 lg:p-6 lg:pr-0">
<Box display="flex" flexDirection={["column", "row"]} h="full" gap={["0", "0", "0", "6"]}>
<Box p={["3", "3", "3", "6"]} pr={["3", "3", "3", "0"]}>
<SideMenu buttonOptions={props.menuButtonOptions} />
</Box>
<Box className="flex flex-col overflow-y-auto p-3 lg:p-6 lg:pl-0 gap-14 w-full">{props.children}</Box>
<Box className="overflow-y-auto p-3 lg:p-6 lg:pl-0 w-full">{props.children}</Box>
</Box>
</Box>
);
website/src/components/Survey/LabelRadioGroup.tsx
+49 -4
View File
@@ -1,4 +1,18 @@
import { Box, Button, Flex, useColorMode } from "@chakra-ui/react";
import {
Box,
Button,
Flex,
IconButton,
Popover,
PopoverArrow,
PopoverBody,
PopoverCloseButton,
PopoverContent,
PopoverTrigger,
Text,
useColorMode,
} from "@chakra-ui/react";
import { InformationCircleIcon } from "@heroicons/react/20/solid";
import { useId, useState } from "react";
import { colors } from "src/styles/Theme/colors";
@@ -8,6 +22,17 @@ interface LabelRadioGroupProps {
isEditable?: boolean;
}
const label_messages: { [label: string]: { description: string; explanation: string[] } } = {
spam: {
description: "The message is spam?",
explanation: [
'We consider the following unwanted content as spam: trolling, intentional undermining of our purpose, illegal material, material that violates our code of conduct, and other things that are inappropriate for our dataset. We collect these under the common heading of "spam".',
"This is not an assessment of whether this message is the best possible answer. Especially for prompts or user-replies, we very much want to retain all kinds of responses in the dataset, so that the assistant can learn to reply appropriately.",
"Please mark this text as spam only if it is clearly unsuited to be part of our dataset, as outlined above, and try not to make any subjective value-judgments beyond that.",
],
},
};
export const LabelRadioGroup = (props: LabelRadioGroupProps) => {
const [labelValues, setLabelValues] = useState<number[]>(Array.from({ length: props.labelIDs.length }).map(() => 0));
const [interactionFlag, setInteractionFlag] = useState(false);
@@ -17,7 +42,7 @@ export const LabelRadioGroup = (props: LabelRadioGroupProps) => {
{props.labelIDs.map((labelId, idx) => (
<LabelRadioItem
key={idx}
labelId={labelId}
labelText={label_messages[labelId] || { description: labelId }}
labelValue={labelValues[idx]}
clickHandler={(newValue) => {
const newState = labelValues.slice();
@@ -45,7 +70,7 @@ interface ButtonState {
}
interface LabelRadioItemProps {
labelId: string;
labelText: { description: string; explanation?: string[] };
labelValue: number;
clickHandler: (newVal: number) => unknown;
states: ButtonState[];
@@ -63,7 +88,27 @@ const LabelRadioItem = (props: LabelRadioItemProps) => {
<Box data-cy="label-group-item" data-label-type="radio">
<label className="text-sm" htmlFor={id}>
{/* TODO: display real text instead of just the id */}
<span className={labelTextClass}>{props.labelId}</span>
<span className={labelTextClass}>{props.labelText.description}</span>
{props.labelText.explanation ? (
<Popover>
<PopoverTrigger>
<IconButton
aria-label="explanation"
variant="link"
icon={<InformationCircleIcon className="h-5 w-5" />}
></IconButton>
</PopoverTrigger>
<PopoverContent>
<PopoverArrow />
<PopoverCloseButton />
<PopoverBody>
{props.labelText.explanation.map((paragraph, idx) => (
<Text key={idx}>{paragraph}</Text>
))}
</PopoverBody>
</PopoverContent>
</Popover>
) : null}
</label>
<Flex direction="row" gap={6} justify="center">
{props.states.map((item, idx) => (
website/src/components/Tasks/LabelTask/LabelTask.tsx
+1 -1
View File
@@ -69,7 +69,7 @@ export const LabelTask = ({
</Box>
)}
</>
{valid_labels.length === 1 ? (
{task.mode === "simple" ? (
<LabelRadioGroup labelIDs={task.valid_labels} isEditable={isEditable} onChange={onSliderChange} />
) : (
<LabelSliderGroup labelIDs={task.valid_labels} isEditable={isEditable} onChange={onSliderChange} />
website/src/components/Tasks/Task/Task.tsx
+1 -1
View File
@@ -27,7 +27,7 @@ export const Task = ({ frontendId, task, trigger, mutate }) => {
const replyContent = useRef<TaskContent>(null);
const [showUnchangedWarning, setShowUnchangedWarning] = useState(false);
const taskType = TaskTypes.find((taskType) => taskType.type === task.type);
const taskType = TaskTypes.find((taskType) => taskType.type === task.type && taskType.mode === task.mode);
const { trigger: sendRejection } = useSWRMutation("/api/reject_task", post, {
onSuccess: async () => {
website/src/components/Tasks/TaskTypes.tsx
+37 -2
View File
@@ -11,6 +11,7 @@ export interface TaskInfo {
category: TaskCategory;
pathname: string;
type: string;
mode?: string;
overview?: string;
instruction?: string;
update_type: string;
@@ -90,7 +91,7 @@ export const TaskTypes: TaskInfo[] = [
unchanged_title: "Order Unchanged",
unchanged_message: "You have not changed the order of the prompts. Are you sure you would like to continue?",
},
// label
// label (full)
{
label: "Label Initial Prompt",
desc: "Provide labels for a prompt.",
@@ -98,6 +99,7 @@ export const TaskTypes: TaskInfo[] = [
pathname: "/label/label_initial_prompt",
overview: "Provide labels for the following prompt",
type: "label_initial_prompt",
mode: "full",
update_type: "text_labels",
},
{
@@ -105,8 +107,9 @@ export const TaskTypes: TaskInfo[] = [
desc: "Provide labels for a prompt.",
category: TaskCategory.Label,
pathname: "/label/label_prompter_reply",
overview: "Given the following discussion, provide labels for the final promp",
overview: "Given the following discussion, provide labels for the final prompt",
type: "label_prompter_reply",
mode: "full",
update_type: "text_labels",
},
{
@@ -116,6 +119,38 @@ export const TaskTypes: TaskInfo[] = [
pathname: "/label/label_assistant_reply",
overview: "Given the following discussion, provide labels for the final prompt.",
type: "label_assistant_reply",
mode: "full",
update_type: "text_labels",
},
// label (simple)
{
label: "Classify Initial Prompt",
desc: "Provide labels for a prompt.",
category: TaskCategory.Label,
pathname: "/label/label_initial_prompt",
overview: "Read the following prompt and then answer the question about it.",
type: "label_initial_prompt",
mode: "simple",
update_type: "text_labels",
},
{
label: "Classify Prompter Reply",
desc: "Provide labels for a prompt.",
category: TaskCategory.Label,
pathname: "/label/label_prompter_reply",
overview: "Read the following conversation and then answer the question about the last prompt in the discussion.",
type: "label_prompter_reply",
mode: "simple",
update_type: "text_labels",
},
{
label: "Classify Assistant Reply",
desc: "Provide labels for a prompt.",
category: TaskCategory.Label,
pathname: "/label/label_assistant_reply",
overview: "Read the following conversation and then answer the question about the last prompt in the discussion.",
type: "label_assistant_reply",
mode: "simple",
update_type: "text_labels",
},
];
website/src/lib/oasst_api_client.ts
+1 -1
View File
@@ -176,7 +176,7 @@ export class OasstApiClient {
const params = new URLSearchParams();
params.append("max_count", max_count.toString());
// The backend API uses different query paramters depending on the
// The backend API uses different query parameters depending on the
// pagination direction but they both take the same cursor value.
// Depending on direction, pick the right query param.
if (cursor !== "") {
website/src/pages/404.tsx
-3
View File
@@ -6,9 +6,6 @@ import { PageEmptyState } from "src/components/EmptyState";
import { getTransparentHeaderLayout } from "src/components/Layout";
function Error() {
const router = useRouter();
const backgroundColor = useColorModeValue("white", "gray.800");
return (
<>
<Head>
website/src/pages/admin/manage_user/[id].tsx
+44 -45
View File
@@ -1,18 +1,28 @@
import { Button, Container, FormControl, FormLabel, Input, Stack, useToast } from "@chakra-ui/react";
import { Field, Form, Formik } from "formik";
import { Button, Card, CardBody, Container, FormControl, FormLabel, Input, Stack, useToast } from "@chakra-ui/react";
import { InferGetServerSidePropsType } from "next";
import Head from "next/head";
import { useRouter } from "next/router";
import { useSession } from "next-auth/react";
import { useEffect } from "react";
import { useForm } from "react-hook-form";
import { getAdminLayout } from "src/components/Layout";
import { RoleSelect } from "src/components/RoleSelect";
import { Role, RoleSelect } from "src/components/RoleSelect";
import { UserMessagesCell } from "src/components/UserMessagesCell";
import { post } from "src/lib/api";
import { oasstApiClient } from "src/lib/oasst_api_client";
import prisma from "src/lib/prismadb";
import useSWRMutation from "swr/mutation";
const ManageUser = ({ user }) => {
interface UserForm {
user_id: string;
id: string;
auth_method: string;
display_name: string;
role: Role;
notes: string;
}
const ManageUser = ({ user }: InferGetServerSidePropsType<typeof getServerSideProps>) => {
const toast = useToast();
const router = useRouter();
const { data: session, status } = useSession();
@@ -52,6 +62,10 @@ const ManageUser = ({ user }) => {
},
});
const { register, handleSubmit } = useForm<UserForm>({
defaultValues: user,
});
return (
<>
<Head>
@@ -62,46 +76,31 @@ const ManageUser = ({ user }) => {
/>
</Head>
<Stack gap="4">
<Container className="oa-basic-theme">
<Formik
initialValues={user}
onSubmit={(values) => {
trigger(values);
}}
>
<Form>
<Field name="user_id" type="hidden" />
<Field name="id" type="hidden" />
<Field name="auth_method" type="hidden" />
<Field name="display_name">
{({ field }) => (
<FormControl>
<FormLabel>Display Name</FormLabel>
<Input {...field} isDisabled />
</FormControl>
)}
</Field>
<Field name="role">
{({ field }) => (
<FormControl>
<FormLabel>Role</FormLabel>
<RoleSelect {...field}></RoleSelect>
</FormControl>
)}
</Field>
<Field name="notes">
{({ field }) => (
<FormControl>
<FormLabel>Notes</FormLabel>
<Input {...field} />
</FormControl>
)}
</Field>
<Button mt={4} type="submit">
Update
</Button>
</Form>
</Formik>
<Container>
<Card>
<CardBody>
<form onSubmit={handleSubmit((data) => trigger(data))}>
<input type="hidden" {...register("user_id")}></input>
<input type="hidden" {...register("id")}></input>
<input type="hidden" {...register("auth_method")}></input>
<FormControl>
<FormLabel>Display Name</FormLabel>
<Input {...register("display_name")} isDisabled />
</FormControl>
<FormControl>
<FormLabel>Role</FormLabel>
<RoleSelect {...register("role")}></RoleSelect>
</FormControl>
<FormControl>
<FormLabel>Notes</FormLabel>
<Input {...register("notes")} />
</FormControl>
<Button mt={4} type="submit">
Update
</Button>
</form>
</CardBody>
</Card>
</Container>
<UserMessagesCell path={`/api/admin/user_messages?user=${user.user_id}`} />
</Stack>
@@ -122,7 +121,7 @@ export async function getServerSideProps({ query }) {
});
const user = {
...backend_user,
role: local_user?.role || "general",
role: (local_user?.role || "general") as Role,
};
return {
props: {
website/src/pages/api/leaderboard.ts
+1 -1
View File
@@ -6,7 +6,7 @@ import { LeaderboardTimeFrame } from "src/types/Leaderboard";
* Returns the set of valid labels that can be applied to messages.
*/
const handler = withoutRole("banned", async (req, res) => {
const time_frame = req.query.time_frame as LeaderboardTimeFrame;
const time_frame = (req.query.time_frame as LeaderboardTimeFrame) || LeaderboardTimeFrame.day;
const { leaderboard } = await oasstApiClient.fetch_leaderboard(time_frame);
res.status(200).json(leaderboard);
});
website/src/pages/auth/signin.tsx
+1 -1
View File
@@ -42,7 +42,7 @@ const errorMessages: Record<SignInErrorTypes, string> = {
interface SigninProps {
providers: Awaited<ReturnType<typeof getProviders>>;
}
// eslint-disable-next-line @typescript-eslint/no-unused-vars
function Signin({ providers }: SigninProps) {
const router = useRouter();
const { discord, email, github, credentials } = providers;
website/src/pages/create/initial_prompt.tsx
+2 -2
View File
@@ -19,8 +19,8 @@ const InitialPrompt = () => {
return (
<>
<Head>
<title>Reply as Assistant</title>
<meta name="description" content="Reply as Assistant." />
<title>Initial Prompt</title>
<meta name="description" content="Add an initial Prompt." />
</Head>
<Task key={tasks[0].task.id} frontendId={tasks[0].id} task={tasks[0].task} trigger={trigger} mutate={reset} />
</>
website/src/pages/create/user_reply.tsx
+2 -2
View File
@@ -19,8 +19,8 @@ const UserReply = () => {
return (
<>
<Head>
<title>Reply as Assistant</title>
<meta name="description" content="Reply as Assistant." />
<title>Reply as User</title>
<meta name="description" content="Reply as User." />
</Head>
<Task key={tasks[0].task.id} frontendId={tasks[0].id} task={tasks[0].task} trigger={trigger} mutate={reset} />
</>
website/src/pages/dashboard.tsx
+5 -2
View File
@@ -1,3 +1,4 @@
import { Flex } from "@chakra-ui/react";
import Head from "next/head";
import { useSession } from "next-auth/react";
import { LeaderboardTable, TaskOption } from "src/components/Dashboard";
@@ -16,8 +17,10 @@ const Dashboard = () => {
<title>Dashboard - Open Assistant</title>
<meta name="description" content="Chat with Open Assistant and provide feedback." />
</Head>
<TaskOption displayTaskCategories={[TaskCategory.Tasks]} />
<LeaderboardTable />
<Flex direction="column" gap="10">
<TaskOption displayTaskCategories={[TaskCategory.Tasks]} />
<LeaderboardTable />
</Flex>
</>
);
};
website/src/pages/leaderboard.tsx
+1 -1
View File
@@ -1,4 +1,4 @@
import { Box, Heading, Tabs, TabList, TabPanels, Tab, TabPanel } from "@chakra-ui/react";
import { Box, Heading, Tab, TabList, TabPanel, TabPanels, Tabs } from "@chakra-ui/react";
import Head from "next/head";
import { getDashboardLayout } from "src/components/Layout";
import { LeaderboardGridCell } from "src/components/LeaderboardGridCell";
website/src/pages/privacy-policy.tsx
+1 -1
View File
@@ -88,7 +88,7 @@ const PrivacyPolicy = () => {
{
number: "4",
title: "Inquiries",
desc: "When you contact us via e-mail, telephone or telefax, your inquiry, including all personal data arising thereof will be stored by us for the purpose of processing your request. We will not pass on these data without your consent. The processing of these data is based on Article 6 (1) (1) (b) GDPR, if your inquiry is related to the fulfilment of a contract concluded with us or required for the implementation of pre-contractual measures. Furthermore, the processing is based on Article 6 (1) (1) (f) GDPR, because we have a legitimate interest in the effective handling of requests sent to us. In addition, according to Article 6 (1) (1) (c) GDPR we are also entitled to the processing of the above-mentioned data, because we are legally bound to enable fast electronic contact and immediate communication. Of course, your data will only be used strictly according to purpose and only for processing and responding to your request. After final processing, your data will immediately be anonymized or deleted, unless we are bound by a legally prescribed storage period.",
desc: "When you contact us via e-mail, telephone or telefax, your inquiry, including all personal data arising thereof will be stored by us for the purpose of processing your request. We will not pass on these data without your consent. The processing of these data is based on Article 6 (1) (1) (b) GDPR, if your inquiry is related to the fulfillment of a contract concluded with us or required for the implementation of pre-contractual measures. Furthermore, the processing is based on Article 6 (1) (1) (f) GDPR, because we have a legitimate interest in the effective handling of requests sent to us. In addition, according to Article 6 (1) (1) (c) GDPR we are also entitled to the processing of the above-mentioned data, because we are legally bound to enable fast electronic contact and immediate communication. Of course, your data will only be used strictly according to purpose and only for processing and responding to your request. After final processing, your data will immediately be anonymized or deleted, unless we are bound by a legally prescribed storage period.",
sections: [],
},
{
website/src/pages/terms-of-service.tsx
+1 -1
View File
@@ -14,7 +14,7 @@ const TermsOfService = () => {
{
number: "1.1",
title: "",
desc: `LAION (association in formation), Marie-Henning-Weg 143, 21035 Hamburg (hereinafter referred to as: "LAION") operates an online portal for the producing a machine learning model called Open Assistant using crowdsourced data.`,
desc: `LAION (association in formation), Marie-Henning-Weg 143, 21035 Hamburg (hereinafter referred to as: "LAION") operates an online portal for the producing a machine learning model called Open Assistant using crowd-sourced data.`,
},
{
number: "1.2",
website/src/styles/Home.module.css
View File
website/src/styles/Theme/components/Card.ts
+27
View File
@@ -0,0 +1,27 @@
import { cardAnatomy } from "@chakra-ui/anatomy";
import { createMultiStyleConfigHelpers } from "@chakra-ui/react";
const { definePartsStyle, defineMultiStyleConfig } = createMultiStyleConfigHelpers(cardAnatomy.keys);
export const cardTheme = defineMultiStyleConfig({
baseStyle: definePartsStyle(({ colorMode }) => {
const isLightMode = colorMode === "light";
return {
container: {
backgroundColor: isLightMode ? "white" : "gray.700",
},
header: {},
body: {
padding: 6,
},
footer: {},
};
}),
variants: {
elevated: definePartsStyle({
container: {
borderRadius: "xl",
},
}),
},
});
website/src/styles/Theme/index.ts
+2
View File
@@ -2,6 +2,7 @@ import { type ThemeConfig, extendTheme } from "@chakra-ui/react";
import { Styles } from "@chakra-ui/theme-tools";
import { colors } from "./colors";
import { cardTheme } from "./components/Card";
import { containerTheme } from "./components/Container";
const config: ThemeConfig = {
@@ -12,6 +13,7 @@ const config: ThemeConfig = {
const components = {
Container: containerTheme,
Card: cardTheme,
};
const breakpoints = {