wassname/Open-Assistant
1
0
Fork 0
mirror of https://github.com/wassname/Open-Assistant.git synced 2026-07-18 12:10:34 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'main' of github.com:LAION-AI/Open-Chat-GPT

Browse Source
This commit is contained in:
Yannic Kilcher
2023-01-14 20:06:32 +01:00
parent 39cf2cf6db 5cd9c630fc
commit 493947fddd
4 changed files with 133 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/alembic/versions/2023_01_14_1509-619255ae9076_add_rank_to_message_table.py
+90
View File
@@ -0,0 +1,90 @@
"""add rank to message table
Revision ID: 619255ae9076
Revises: bcc2fe18d214
Create Date: 2023-01-14 15:09:03.462482
"""
import sqlalchemy as sa
import sqlmodel
from alembic import op
from sqlalchemy.dialects import postgresql
# revision identifiers, used by Alembic.
revision = "619255ae9076"
down_revision = "bcc2fe18d214"
branch_labels = None
depends_on = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
op.add_column("message", sa.Column("rank", sa.Integer(), nullable=True))
op.add_column("message_toxicity", sa.Column("score", sa.Float(), nullable=True))
op.add_column("message_toxicity", sa.Column("label", sqlmodel.sql.sqltypes.AutoString(length=256), nullable=False))
op.drop_column("message_toxicity", "toxicity")
op.add_column("user_stats", sa.Column("time_frame", sqlmodel.sql.sqltypes.AutoString(), nullable=False))
op.add_column("user_stats", sa.Column("prompts", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("replies_assistant", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("replies_prompter", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("labels_simple", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("labels_full", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("rankings_total", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("rankings_good", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("accepted_prompts", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("accepted_replies_assistant", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("accepted_replies_prompter", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("reply_assistant_ranked_1", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("reply_assistant_ranked_2", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("reply_assistant_ranked_3", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("reply_prompter_ranked_1", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("reply_prompter_ranked_2", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("reply_prompter_ranked_3", sa.Integer(), nullable=False))
op.add_column("user_stats", sa.Column("streak_last_day_date", sa.DateTime(), nullable=True))
op.add_column("user_stats", sa.Column("streak_days", sa.Integer(), nullable=True))
op.drop_column("user_stats", "messages")
op.drop_column("user_stats", "upvotes")
op.drop_column("user_stats", "task_reward")
op.drop_column("user_stats", "compare_wins")
op.drop_column("user_stats", "compare_losses")
op.drop_column("user_stats", "downvotes")
op.drop_column("user_stats", "reactions")
# ### end Alembic commands ###
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
op.add_column("user_stats", sa.Column("reactions", sa.INTEGER(), autoincrement=False, nullable=False))
op.add_column("user_stats", sa.Column("downvotes", sa.INTEGER(), autoincrement=False, nullable=False))
op.add_column("user_stats", sa.Column("compare_losses", sa.INTEGER(), autoincrement=False, nullable=False))
op.add_column("user_stats", sa.Column("compare_wins", sa.INTEGER(), autoincrement=False, nullable=False))
op.add_column("user_stats", sa.Column("task_reward", sa.INTEGER(), autoincrement=False, nullable=False))
op.add_column("user_stats", sa.Column("upvotes", sa.INTEGER(), autoincrement=False, nullable=False))
op.add_column("user_stats", sa.Column("messages", sa.INTEGER(), autoincrement=False, nullable=False))
op.drop_column("user_stats", "streak_days")
op.drop_column("user_stats", "streak_last_day_date")
op.drop_column("user_stats", "reply_prompter_ranked_3")
op.drop_column("user_stats", "reply_prompter_ranked_2")
op.drop_column("user_stats", "reply_prompter_ranked_1")
op.drop_column("user_stats", "reply_assistant_ranked_3")
op.drop_column("user_stats", "reply_assistant_ranked_2")
op.drop_column("user_stats", "reply_assistant_ranked_1")
op.drop_column("user_stats", "accepted_replies_prompter")
op.drop_column("user_stats", "accepted_replies_assistant")
op.drop_column("user_stats", "accepted_prompts")
op.drop_column("user_stats", "rankings_good")
op.drop_column("user_stats", "rankings_total")
op.drop_column("user_stats", "labels_full")
op.drop_column("user_stats", "labels_simple")
op.drop_column("user_stats", "replies_prompter")
op.drop_column("user_stats", "replies_assistant")
op.drop_column("user_stats", "prompts")
op.drop_column("user_stats", "time_frame")
op.add_column(
"message_toxicity",
sa.Column("toxicity", postgresql.DOUBLE_PRECISION(precision=53), autoincrement=False, nullable=True),
)
op.drop_column("message_toxicity", "label")
op.drop_column("message_toxicity", "score")
op.drop_column("message", "rank")
# ### end Alembic commands ###
backend/oasst_backend/config.py
+8 -8
View File
@@ -20,28 +20,28 @@ class TreeManagerConfiguration(BaseModel):
"""Maximum number of reply messages per tree node."""
goal_tree_size: int = 15
"""Total number of messages to gather per tree"""
"""Total number of messages to gather per tree."""
num_reviews_initial_prompt: int = 3
"""Number of peer review checks to collect in INITIAL_PROMPT_REVIEW state."""
num_reviews_reply: int = 3
"""Number of peer review checks to collect per reply (other than initial_prompt)"""
"""Number of peer review checks to collect per reply (other than initial_prompt)."""
p_full_labeling_review_prompt: float = 0.1
"""Probability of full text-labeling (instead of mandatory only) for initial prompts"""
"""Probability of full text-labeling (instead of mandatory only) for initial prompts."""
p_full_labeling_review_reply_assistant: float = 0.1
"""Probability of full text-labeling (instead of mandatory only) for assistant replies"""
"""Probability of full text-labeling (instead of mandatory only) for assistant replies."""
p_full_labeling_review_reply_prompter: float = 0.1
"""Probability of full text-labeling (instead of mandatory only) for prompter replies"""
"""Probability of full text-labeling (instead of mandatory only) for prompter replies."""
acceptance_threshold_initial_prompt: float = 0.6
"""Threshold for accepting an initial prompt"""
"""Threshold for accepting an initial prompt."""
acceptance_threshold_reply: float = 0.6
"""Threshold for accepting a reply"""
"""Threshold for accepting a reply."""
num_required_rankings: int = 3
"""Number of rankings in which the message participated."""
@@ -50,7 +50,7 @@ class TreeManagerConfiguration(BaseModel):
"""Mandatory labels in text-labeling tasks for initial prompts."""
mandatory_labels_assistant_reply: Optional[list[protocol_schema.TextLabel]] = [protocol_schema.TextLabel.spam]
"""Mandatory labels in text-labeling tasks for assistant reylies."""
"""Mandatory labels in text-labeling tasks for assistant replies."""
mandatory_labels_prompter_reply: Optional[list[protocol_schema.TextLabel]] = [protocol_schema.TextLabel.spam]
"""Mandatory labels in text-labeling tasks for prompter replies."""
backend/oasst_backend/models/message.py
+2
View File
@@ -45,6 +45,8 @@ class Message(SQLModel, table=True):
review_result: bool = Field(sa_column=sa.Column(sa.Boolean, default=False, server_default=false(), nullable=False))
ranking_count: int = Field(sa_column=sa.Column(sa.Integer, default=0, server_default=sa.text("0"), nullable=False))
rank: Optional[int] = Field(nullable=True)
def ensure_is_message(self) -> None:
if not self.payload or not isinstance(self.payload.payload, MessagePayload):
raise OasstError("Invalid message", OasstErrorCode.INVALID_MESSAGE, HTTPStatus.INTERNAL_SERVER_ERROR)
backend/oasst_backend/models/user_stats.py
+33 -7
View File
@@ -1,4 +1,5 @@
from datetime import datetime
from enum import Enum
from typing import Optional
from uuid import UUID
@@ -7,21 +8,46 @@ import sqlalchemy.dialects.postgresql as pg
from sqlmodel import Field, SQLModel
class UserStatsTimeFrame(str, Enum):
day = "day"
week = "week"
month = "month"
total = "total"
class UserStats(SQLModel, table=True):
__tablename__ = "user_stats"
user_id: Optional[UUID] = Field(
sa_column=sa.Column(pg.UUID(as_uuid=True), sa.ForeignKey("user.id"), primary_key=True)
)
time_frame: Optional[str] = Field(nullable=False, primary_key=True)
leader_score: int = 0
modified_date: Optional[datetime] = Field(
sa_column=sa.Column(sa.DateTime(), nullable=False, server_default=sa.func.current_timestamp())
)
reactions: int = 0 # reactions sent by user
messages: int = 0 # messages sent by user
upvotes: int = 0 # received upvotes (form other users)
downvotes: int = 0 # received downvotes (from other users)
task_reward: int = 0 # reward for task completions
compare_wins: int = 0 # num times user's message won compare tasks
compare_losses: int = 0 # num times users's message lost compare tasks
prompts: int = 0
replies_assistant: int = 0
replies_prompter: int = 0
labels_simple: int = 0
labels_full: int = 0
rankings_total: int = 0
rankings_good: int = 0
accepted_prompts: int = 0
accepted_replies_assistant: int = 0
accepted_replies_prompter: int = 0
reply_assistant_ranked_1: int = 0
reply_assistant_ranked_2: int = 0
reply_assistant_ranked_3: int = 0
reply_prompter_ranked_1: int = 0
reply_prompter_ranked_2: int = 0
reply_prompter_ranked_3: int = 0
# only used for time span "total"
streak_last_day_date: Optional[datetime] = Field(nullable=True)
streak_days: Optional[int] = Field(nullable=True)