{ "cells": [ { "cell_type": "code", "execution_count": 1, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:58:21.040680Z", "start_time": "2024-05-16T13:58:19.777010Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:29:31.729492Z", "iopub.status.busy": "2025-02-25T01:29:31.729317Z", "iopub.status.idle": "2025-02-25T01:29:33.800756Z", "shell.execute_reply": "2025-02-25T01:29:33.800288Z", "shell.execute_reply.started": "2025-02-25T01:29:31.729478Z" } }, "outputs": [], "source": [ "# setup autoload\n", "%load_ext autoreload\n", "%autoreload 2\n", "\n", "import ast\n", "import os\n", "import shutil\n", "import sys\n", "from collections import defaultdict\n", "\n", "import numpy as np\n", "import pandas as pd\n", "from preference_data_preparation_4min_30b_task import *\n", "from sklearn.model_selection import train_test_split\n", "from suno_utils.utils.s3 import download_s3_files\n", "from suno_utils.utils.text import read_json, read_jsonl, write_json, write_jsonl\n", "from tqdm import tqdm\n", "import matplotlib.pyplot as plt\n", "\n", "pd.set_option(\"display.max_rows\", 500)\n", "pd.set_option(\"display.max_columns\", 500)\n", "pd.set_option(\"display.width\", 1000)" ] }, { "cell_type": "code", "execution_count": 2, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:58:21.082172Z", "start_time": "2024-05-16T13:58:21.041926Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:29:33.801399Z", "iopub.status.busy": "2025-02-25T01:29:33.801208Z", "iopub.status.idle": "2025-02-25T01:29:33.837683Z", "shell.execute_reply": "2025-02-25T01:29:33.837283Z", "shell.execute_reply.started": "2025-02-25T01:29:33.801387Z" } }, "outputs": [], "source": [ "OUT_DATA_DIR = \"/app/suno/data/dpo/13b_s32_v22/\"\n", "os.makedirs(OUT_DATA_DIR, exist_ok=True)\n", "shutil.copyfile(\n", " \"/app/suno/data/dpo/7v_v20_full/tokenizer_60k.json\",\n", " os.path.join(OUT_DATA_DIR, \"tokenizer_60k.json\"),\n", ")\n", "NPZ_DIR = \"/app/suno/data/dpo/concat_cycle_npz\"" ] }, { "cell_type": "code", "execution_count": 3, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:58:53.962528Z", "start_time": "2024-05-16T13:58:21.105919Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:29:33.838343Z", "iopub.status.busy": "2025-02-25T01:29:33.838163Z", "iopub.status.idle": "2025-02-25T01:29:36.899982Z", "shell.execute_reply": "2025-02-25T01:29:36.899471Z", "shell.execute_reply.started": "2025-02-25T01:29:33.838326Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Preference data shape (216002, 152)\n" ] } ], "source": [ "# df = pd.read_csv(\n", "# \"/home/tony/Data/Preference/13b_v0/interesting_clips_v3p5_s_8_20240813.csv\"\n", "# ) # , engine='python')\n", "df = pd.read_pickle(\n", " \"/home/tony/Data/Preference/13b_v32/interesting_clips_v4_h_s_32_20250222_full_long_final.pkl\"\n", ") # , engine='python')\n", "print(\"Preference data shape\", df.shape)" ] }, { "cell_type": "code", "execution_count": 4, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:58:56.199480Z", "start_time": "2024-05-16T13:58:53.963687Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:29:36.901275Z", "iopub.status.busy": "2025-02-25T01:29:36.901038Z", "iopub.status.idle": "2025-02-25T01:29:38.269202Z", "shell.execute_reply": "2025-02-25T01:29:38.268708Z", "shell.execute_reply.started": "2025-02-25T01:29:36.901262Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "457458\n", "457458\n", "pre-downloaded df (216002, 152)\n", "downloaded df (214457, 152)\n" ] } ], "source": [ "converted_paths = os.listdir(NPZ_DIR)\n", "print(len(converted_paths))\n", "\n", "converted_paths = set([f.replace(\".npz\", \"\") for f in converted_paths])\n", "print(len(converted_paths))\n", "\n", "if \"cycle\" in NPZ_DIR:\n", " # hack in the cycle label\n", " df[\"s3_id\"] += \"_gen_cycle\"\n", "\n", "print(\"pre-downloaded df\", df.shape)\n", "df[df[\"s3_id\"].isin(converted_paths)].shape\n", "df = df[df[\"s3_id\"].isin(converted_paths)].copy()\n", "print(\"downloaded df\", df.shape)" ] }, { "cell_type": "code", "execution_count": 5, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:58:56.467253Z", "start_time": "2024-05-16T13:58:56.207647Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:29:38.269918Z", "iopub.status.busy": "2025-02-25T01:29:38.269709Z", "iopub.status.idle": "2025-02-25T01:29:38.318780Z", "shell.execute_reply": "2025-02-25T01:29:38.318403Z", "shell.execute_reply.started": "2025-02-25T01:29:38.269905Z" } }, "outputs": [ { "data": { "text/plain": [ "is_13b\n", "True 214457\n", "Name: count, dtype: int64" ] }, "execution_count": 5, "metadata": {}, "output_type": "execute_result" } ], "source": [ "df[\"is_13b\"] = df[\"model_name\"].str.contains(\"-s-\")\n", "df[\"is_13b\"].value_counts()" ] }, { "cell_type": "code", "execution_count": 6, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T01:29:38.319409Z", "iopub.status.busy": "2025-02-25T01:29:38.319232Z", "iopub.status.idle": "2025-02-25T01:29:38.336028Z", "shell.execute_reply": "2025-02-25T01:29:38.335665Z", "shell.execute_reply.started": "2025-02-25T01:29:38.319396Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "is_public\n", "False 186370\n", "True 28087\n", "Name: count, dtype: int64\n" ] } ], "source": [ "print(df[\"is_public\"].value_counts())\n", "# remove public for now cause fucking users\n", "# df = df[~df[\"is_public\"]]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# LET's do the data prep" ] }, { "cell_type": "code", "execution_count": 7, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:58:56.592883Z", "start_time": "2024-05-16T13:58:56.470781Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:29:38.336737Z", "iopub.status.busy": "2025-02-25T01:29:38.336461Z", "iopub.status.idle": "2025-02-25T01:29:38.529272Z", "shell.execute_reply": "2025-02-25T01:29:38.528773Z", "shell.execute_reply.started": "2025-02-25T01:29:38.336724Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "preference model_name \n", "False chirp-v4-h-s-32 106617\n", "True chirp-v4-h-s-32 107840\n", "Name: count, dtype: int64\n", "(214457, 152)\n", "(214457, 152)\n" ] } ], "source": [ "## for 13b this is easy for now\n", "print(df.groupby([\"preference\"])[\"model_name\"].value_counts())\n", "print(df.shape)\n", "df = df[df[\"model_name\"].isin([\"chirp-v4-h-s-32\"])]\n", "print(df.shape)" ] }, { "cell_type": "code", "execution_count": 8, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:58:56.909539Z", "start_time": "2024-05-16T13:58:56.595736Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:29:38.529942Z", "iopub.status.busy": "2025-02-25T01:29:38.529804Z", "iopub.status.idle": "2025-02-25T01:29:38.830217Z", "shell.execute_reply": "2025-02-25T01:29:38.829717Z", "shell.execute_reply.started": "2025-02-25T01:29:38.529929Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "(214457, 152)\n", "(213062, 152)\n", "preference model_name \n", "False chirp-v4-h-s-32 106531\n", "True chirp-v4-h-s-32 106531\n", "Name: count, dtype: int64\n" ] } ], "source": [ "print(df.shape)\n", "df = df[\n", " df[\"request_id\"].isin(\n", " df[\"request_id\"].value_counts().index[df[\"request_id\"].value_counts() == 2]\n", " )\n", "]\n", "print(df.shape)\n", "print(df.groupby([\"preference\"])[\"model_name\"].value_counts())\n", "assert df.shape[0] == df[\"request_id\"].nunique() * 2" ] }, { "cell_type": "code", "execution_count": 9, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T01:29:38.830943Z", "iopub.status.busy": "2025-02-25T01:29:38.830736Z", "iopub.status.idle": "2025-02-25T01:29:38.844060Z", "shell.execute_reply": "2025-02-25T01:29:38.843702Z", "shell.execute_reply.started": "2025-02-25T01:29:38.830930Z" } }, "outputs": [], "source": [ "import json\n", "\n", "def custom_parse(x):\n", " try:\n", " return json.loads(x)\n", " except:\n", " return {}" ] }, { "cell_type": "code", "execution_count": 10, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:36.043975Z", "start_time": "2024-05-16T13:58:56.910958Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:29:38.844597Z", "iopub.status.busy": "2025-02-25T01:29:38.844488Z", "iopub.status.idle": "2025-02-25T01:30:04.197305Z", "shell.execute_reply": "2025-02-25T01:30:04.196781Z", "shell.execute_reply.started": "2025-02-25T01:29:38.844586Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "unique_requests 106531\n" ] } ], "source": [ "# Let's use the old selection for now -- for quality assurance\n", "# expand the metadata columns -- this takes forever...~ 6 mins\n", "# test_slice = df[\"metadata\"].apply(lambda x: ast.literal_eval(str(x)))\n", "test_slice = df[\"metadata\"] #.apply(lambda x: custom_parse(x))\n", "test_slice_series = test_slice.apply(pd.Series)\n", "df = pd.concat([df, test_slice_series], axis=1, join=\"inner\")\n", "print(\"unique_requests\", df[\"request_id\"].nunique())" ] }, { "cell_type": "code", "execution_count": 11, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T01:30:04.198199Z", "iopub.status.busy": "2025-02-25T01:30:04.197874Z", "iopub.status.idle": "2025-02-25T01:30:04.230172Z", "shell.execute_reply": "2025-02-25T01:30:04.229763Z", "shell.execute_reply.started": "2025-02-25T01:30:04.198185Z" } }, "outputs": [], "source": [ "# def modify_model_name(model_name, metadata):\n", "# if (\n", "# model_name.startswith(\"chirp-v3p5-engine-t\")\n", "# or model_name.startswith(\"chirp-v3p5-engine-s\")\n", "# or model_name.startswith(\"chirp-v4\")\n", "# or model_name.startswith(\"chirp-v3p5-h-s-31\")\n", "# ):\n", "# if \"param_experiment\" in metadata:\n", "# exp = metadata.get(\"param_experiment\", \"\")\n", "# if exp:\n", "# return f\"{model_name}_{exp}\"\n", "# return model_name\n", "\n", "\n", "# df[\"param_model_name\"] = df.apply(\n", "# lambda row: modify_model_name(row[\"model_name\"], row[\"metadata\"]), axis=1\n", "# )\n", "# diff_experiments = (df[\"param_model_name\"].str.contains(\"text_\"))| (df[\"param_model_name\"].str.contains(\"step_\")) | (df[\"param_model_name\"].str.contains(\"tk_\"))" ] }, { "cell_type": "code", "execution_count": 12, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T01:30:04.230818Z", "iopub.status.busy": "2025-02-25T01:30:04.230633Z", "iopub.status.idle": "2025-02-25T01:30:04.271194Z", "shell.execute_reply": "2025-02-25T01:30:04.270788Z", "shell.execute_reply.started": "2025-02-25T01:30:04.230805Z" } }, "outputs": [], "source": [ "# # try taking out diffusion experiment -- cause they could literally be noise\n", "# print(\"unique_requests\", df[\"request_id\"].nunique())\n", "# df = df[~diff_experiments].copy()\n", "# df = df[\n", "# df[\"request_id\"].isin(\n", "# df[\"request_id\"].value_counts().index[df[\"request_id\"].value_counts() == 2]\n", "# )\n", "# ]\n", "# print(\"unique_requests\", df[\"request_id\"].nunique())" ] }, { "cell_type": "code", "execution_count": 13, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:36.393047Z", "start_time": "2024-05-16T13:59:36.048831Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:30:04.273274Z", "iopub.status.busy": "2025-02-25T01:30:04.272896Z", "iopub.status.idle": "2025-02-25T01:30:04.331662Z", "shell.execute_reply": "2025-02-25T01:30:04.331170Z", "shell.execute_reply.started": "2025-02-25T01:30:04.273260Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "unique_requests 106531\n" ] } ], "source": [ "# GPT requests are also fine for now\n", "print(\"unique_requests\", df[\"request_id\"].nunique())" ] }, { "cell_type": "code", "execution_count": 14, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:40.799375Z", "start_time": "2024-05-16T13:59:36.394236Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:30:04.332456Z", "iopub.status.busy": "2025-02-25T01:30:04.332261Z", "iopub.status.idle": "2025-02-25T01:30:05.493950Z", "shell.execute_reply": "2025-02-25T01:30:05.493473Z", "shell.execute_reply.started": "2025-02-25T01:30:04.332442Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "6646\n", "good_continue_at\n", "True 213062\n", "Name: count, dtype: int64\n", "\n", " Check some basics... \n", " preference\n", "False 106531\n", "True 106531\n", "Name: count, dtype: int64 is_13b\n", "True 213062\n", "Name: count, dtype: int64 model_name\n", "chirp-v4-h-s-32 213062\n", "Name: count, dtype: int64 preference model_name \n", "False chirp-v4-h-s-32 106531\n", "True chirp-v4-h-s-32 106531\n", "Name: count, dtype: int64\n", "task\n", " 198468\n", "extend 10518\n", "upload_extend 4074\n", "cover 2\n", "Name: count, dtype: int64\n" ] } ], "source": [ "df = df.loc[:, ~df.columns.duplicated()].copy()\n", "# get the original duration of the clips, if they are concacted\n", "df[\"original_duration_s\"] = df[\"total_start_s\"] + df[\"duration\"]\n", "# classify the continue at behavoirs by the duration choice\n", "audio_prompt_id_to_continue_at = {}\n", "for _, row in df[~df[\"continued_parent\"].isna()].iterrows():\n", " audio_prompt_id = row[\"continued_parent\"]\n", " if audio_prompt_id not in audio_prompt_id_to_continue_at:\n", " audio_prompt_id_to_continue_at[audio_prompt_id] = row[\"continue_at\"]\n", " else:\n", " # pick the max\n", " audio_prompt_id = max(\n", " audio_prompt_id_to_continue_at[audio_prompt_id], row[\"continue_at\"]\n", " )\n", "print(len(audio_prompt_id_to_continue_at))\n", "df[\"has_continue_and_start_continue_at\"] = df[\"s3_id\"].apply(\n", " lambda x: audio_prompt_id_to_continue_at.get(x)\n", ")\n", "# we want continue at to be at most of the clip...\n", "df[\"good_continue_at\"] = (\n", " (df[\"has_continue_and_start_continue_at\"] / df[\"duration\"]) > 0.9\n", ") | df[\"has_continue_and_start_continue_at\"].isna()\n", "print(df[\"good_continue_at\"].value_counts())\n", "\n", "\n", "print(\n", " \"\\n Check some basics... \\n\",\n", " df[\"preference\"].value_counts(),\n", " df[\"is_13b\"].value_counts(),\n", " df[\"model_name\"].value_counts(),\n", " df.groupby([\"preference\"])[\"model_name\"].value_counts(),\n", ")\n", "\n", "df = df.sort_values(by=[\"request_id\", \"preference\"])\n", "df[\"duration_rel_diff\"] = df[\"duration\"].diff()\n", "df[\"play_rel_diff\"] = df[\"reaction_play_count\"].diff()\n", "print(df[\"task\"].value_counts())" ] }, { "cell_type": "code", "execution_count": 15, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T01:30:05.494670Z", "iopub.status.busy": "2025-02-25T01:30:05.494459Z", "iopub.status.idle": "2025-02-25T01:30:05.778980Z", "shell.execute_reply": "2025-02-25T01:30:05.778563Z", "shell.execute_reply.started": "2025-02-25T01:30:05.494657Z" } }, "outputs": [ { "data": { "text/plain": [ "pos_diff_preference\n", "2.0 106531\n", "Name: count, dtype: int64" ] }, "execution_count": 15, "metadata": {}, "output_type": "execute_result" } ], "source": [ "df = df.sort_values(by=[\"request_id\", \"preference\", \"diff_preference\"])\n", "df[\"pos_diff_preference\"] = df[\"diff_preference\"].diff()\n", "# df[\"cer_diff_preference\"] = df[\"cer\"].diff()\n", "df[df[\"preference\"]][\"pos_diff_preference\"].value_counts()" ] }, { "cell_type": "code", "execution_count": 16, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T01:30:05.779605Z", "iopub.status.busy": "2025-02-25T01:30:05.779475Z", "iopub.status.idle": "2025-02-25T01:30:05.792404Z", "shell.execute_reply": "2025-02-25T01:30:05.792055Z", "shell.execute_reply.started": "2025-02-25T01:30:05.779592Z" } }, "outputs": [], "source": [ "# df[df[\"preference\"]][\"cer_diff_preference\"].hist(bins=50)\n", "# print(df[df[\"preference\"]][\"cer_diff_preference\"].quantile(0.95))\n", "# plt.show()\n", "# print(df[df[\"preference\"]][\"cer\"].hist(bins=50))\n", "# plt.show()" ] }, { "cell_type": "code", "execution_count": 17, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T01:30:05.792952Z", "iopub.status.busy": "2025-02-25T01:30:05.792844Z", "iopub.status.idle": "2025-02-25T01:30:05.902154Z", "shell.execute_reply": "2025-02-25T01:30:05.901645Z", "shell.execute_reply.started": "2025-02-25T01:30:05.792942Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "positive with likes (68145, 152)\n" ] } ], "source": [ "test_mask = (df[\"preference\"] == True) & (\n", " (df[\"upvote_count\"] >= 1)\n", ")\n", "print(\"positive with likes\", df[test_mask].shape)\n", "# positive with likes (285955, 149) -- 0215 data\n", "# positive with likes (517621, 151) -- 0217 data" ] }, { "cell_type": "code", "execution_count": 18, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.035167Z", "start_time": "2024-05-16T13:59:40.801098Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:30:05.902959Z", "iopub.status.busy": "2025-02-25T01:30:05.902743Z", "iopub.status.idle": "2025-02-25T01:30:06.202845Z", "shell.execute_reply": "2025-02-25T01:30:06.202350Z", "shell.execute_reply.started": "2025-02-25T01:30:05.902946Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "negative 106531 positive 106530\n", "total pair requests 106531 selected pair requests 106530 frac 1.000\n" ] } ], "source": [ "normal_pos_play_count = 10\n", "# this is lower, cause a concat is probably already ensuring that it is good\n", "concat_pos_play_count = 2\n", "# this is a filter on the concated clip\n", "concat_total_play_count = 10\n", "\n", "neg_filter_selection_mask = (\n", " (df[\"preference\"] == False) # get basics aligned\n", " & (df[\"reaction_play_count\"] >= 1) # has to be played once\n", " # & (df[\"play_count\"] <= 3) # if it is actually bad, shouldn't be listened often\n", " & (df[\"duration\"] >= 10) # can't be too short, otherwise it is obvious\n", " # & (df[\"duration\"] <= 60) # can't be badly long\n", " & (df[\"has_continue_and_start_continue_at\"].isna()) # won't have any continues\n", " & ((df[\"norm_play_frac\"] <= 2.1))\n", " # & (df[\"dislike_count\"] >= 1) # this is kinda strict\n", " # & (\n", " # (df_slice[\"is_in_playlist\"] == False)\n", " # & (df_slice[\"concat_in_playlist\"] == False)\n", " # ) # can't be part of a playlist -- otherwise there are some like signal in it?\n", ")\n", "pos_filter_selectin_mask = (\n", " (df[\"preference\"] == True) # get basics aligned\n", " & (\n", " df[\"good_continue_at\"] == True\n", " ) # if continue, needs to continue off a certain percentage\n", " & (df[\"reaction_play_count\"] >= 1)\n", " & (df[\"play_rel_diff\"] >= 0) # this is more like quality assurance\n", " & (df[\"duration\"] >= 10) # can't be too short, otherwise it is obvious\n", " # & (df[\"duration\"] <= 60) # can't be badly long\n", " & (df[\"dislike_count\"] == 0) # can't have dislikes\n", " & (df[\"flag_count\"] == 0) # can't have issues\n", " & (\n", " (\n", " (df[\"part_of_concat\"] == True)\n", " & (df[\"reaction_play_count\"] >= concat_pos_play_count)\n", " & (df[\"concat_play_counts\"] >= concat_total_play_count)\n", " )\n", " | (\n", " (df[\"part_of_concat\"] == False)\n", " & (df[\"reaction_play_count\"] >= normal_pos_play_count)\n", " )\n", " )\n", " & (df[\"user_n_clips\"] >= 200) # user needs to have genereated at least 20\n", " # & (df[\"duration_rel_diff\"] < 10) # positive isn't just longer\n", " # & ((df[\"upvote_count\"] >= 1) )\n", " & ((df[\"norm_play_frac\"] >= 5.1) | (~df[\"continued_parent\"].isna()))\n", " & (df[\"norm_play_frac\"] >= df[\"reaction_play_count\"] / 3) # play duration is not low on average\n", " & (df[\"pos_diff_preference\"] == 2)\n", " & (df[\"task\"].isin([\"upload_extend\", \"extend\", \"\"]))\n", ")\n", "print(\n", " \"negative\",\n", " sum(neg_filter_selection_mask),\n", " \"positive\",\n", " sum(pos_filter_selectin_mask),\n", ")\n", "\n", "neg_filter_requests = df[neg_filter_selection_mask][\"request_id\"].unique()\n", "pos_filter_requests = df[pos_filter_selectin_mask][\"request_id\"].unique()\n", "# looking for very strong signal here:\n", "# listen to the positive/negative more than once\n", "# disliked one of the clips\n", "unique_requests = set(pos_filter_requests).intersection(neg_filter_requests)\n", "print(\n", " \"total pair requests\",\n", " df[\"request_id\"].nunique(),\n", " \"selected pair requests\",\n", " len(unique_requests),\n", " f\"frac {len(unique_requests) / df['request_id'].nunique():.3f}\",\n", ")" ] }, { "cell_type": "code", "execution_count": 19, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.250737Z", "start_time": "2024-05-16T13:59:41.036434Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:30:06.203509Z", "iopub.status.busy": "2025-02-25T01:30:06.203387Z", "iopub.status.idle": "2025-02-25T01:30:06.724810Z", "shell.execute_reply": "2025-02-25T01:30:06.724306Z", "shell.execute_reply.started": "2025-02-25T01:30:06.203497Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ " requests 106530 clips 213060 total khrs 11.323; N gpus for 1000 iters 13.316; 8 gpus for x iters 832.266; n unique users 25050 n pro users 22553\n" ] } ], "source": [ "df_slice = df[df[\"request_id\"].isin(set(unique_requests))].copy()\n", "print(\n", " f\"{os.path.basename(OUT_DATA_DIR)} requests\",\n", " df_slice[\"request_id\"].nunique(),\n", " \"clips\",\n", " df_slice.shape[0],\n", " f\"total khrs {sum(df_slice['duration'] / 3600 / 1000):.3f};\",\n", " f\"N gpus for 1000 iters {df_slice.shape[0] / 8 / 2 / 1000:.3f};\",\n", " f\"8 gpus for x iters {df_slice.shape[0] / 8 / 8 / 4:.3f};\",\n", " f\"n unique users {df_slice['user_id'].nunique()}\",\n", " f\"n pro users {df_slice[df_slice['is_pro_user']]['user_id'].nunique()}\",\n", ")\n", "# 76171 152342 total khrs 2.880 n gpus for 1250 iters 3.809\n", "# 8 v9 requests 106299 clips 212598 total khrs 9.776; N gpus for 1000 iters 13.287; n unique users 27775\n", "# 29 v2 requests 22572 clips 45144 total khrs 2.252; N gpus for 1000 iters 2.821; 4 gpus for x iters 705.375; n unique users 18149 n pro users 10155\n", "# 29 v4 requests 32479 clips 64958 total khrs 3.248; N gpus for 1000 iters 4.060; 4 gpus for x iters 1014.969; n unique users 25192 n pro users 13547\n", "# samve for v5\n", "# 31 v9 requests 63922 clips 127844 total khrs 6.631; N gpus for 1000 iters 7.990; 4 gpus for x iters 1997.562; n unique users 25461 n pro users 24396\n", "# 32 v1 requests 42336 clips 84672 total khrs 4.311; N gpus for 1000 iters 5.292; 4 gpus for x iters 1323.000; n unique users 20440 n pro users 14949\n", "# 32 v2 requests 83218 clips 166436 total khrs 8.672; N gpus for 1000 iters 10.402; 4 gpus for x iters 2600.562; n unique users 36912 n pro users 24545\n", "# 32 v3 requests 71020 clips 142040 total khrs 7.562; N gpus for 1000 iters 8.877; 4 gpus for x iters 2219.375; n unique users 25143 n pro users 22341\n", "# 32 v5 requests 83743 clips 167486 total khrs 8.923; N gpus for 1000 iters 10.468; 4 gpus for x iters 2616.969; n unique users 28348 n pro users 24894\n", "# 32 v6 requests 63924 clips 127848 total khrs 6.725; N gpus for 1000 iters 7.990; 4 gpus for x iters 1997.625; n unique users 24675 n pro users 22071\n", "# 32 v7 requests 106565 clips 213130 total khrs 11.335; N gpus for 1000 iters 13.321; 4 gpus for x iters 3330.156; n unique users 34779 n pro users 29833\n", "# 32 v8 requests 135955 clips 271910 total khrs 14.459; N gpus for 1000 iters 16.994; 4 gpus for x iters 4248.594; n unique users 42469 n pro users 35114\n", "# 32 v9 requests 60046 clips 120092 total khrs 6.444; N gpus for 1000 iters 7.506; 4 gpus for x iters 1876.438; n unique users 15885 n pro users 14566\n", "# 32 v10 requests 79415 clips 158830 total khrs 8.497; N gpus for 1000 iters 9.927; 4 gpus for x iters 2481.719; n unique users 19892 n pro users 17860\n", "# 32 v13 requests 173814 clips 347628 total khrs 18.463; N gpus for 1000 iters 21.727; 4 gpus for x iters 5431.688; n unique users 43848 n pro users 39116\n", "# 32 v14 requests 87118 clips 174236 total khrs 8.689; N gpus for 1000 iters 10.890; 4 gpus for x iters 2722.438; n unique users 22826 n pro users 20899\n", "# 32 v15 requests 68523 clips 137046 total khrs 7.169; N gpus for 1000 iters 8.565; 4 gpus for x iters 1070.672; n unique users 26101 n pro users 23037\n", "# 32 v16 requests 88908 clips 177816 total khrs 8.876; N gpus for 1000 iters 11.114; 4 gpus for x iters 1389.188; n unique users 23225 n pro users 21243\n", "# 32 v17 requests 43677 clips 87354 total khrs 4.468; N gpus for 1000 iters 5.460; 4 gpus for x iters 682.453; n unique users 17554 n pro users 16917\n", "# 32 v18 requests 89035 clips 178070 total khrs 9.312; N gpus for 1000 iters 11.129; 4 gpus for x iters 1391.172; n unique users 26572 n pro users 23755\n", "# 32 v19 requests 117202 clips 234404 total khrs 12.520; N gpus for 1000 iters 14.650; 8 gpus for x iters 915.641; n unique users 31847 n pro users 28249\n", "# 32 v20 requests 52910 clips 105820 total khrs 5.572; N gpus for 1000 iters 6.614; 4 gpus for x iters 826.719; n unique users 16708 n pro users 15184\n", "# 32 v21 requests 108001 clips 216002 total khrs 11.480; N gpus for 1000 iters 13.500; 8 gpus for x iters 843.758; n unique users 25205 n pro users 22700\n", "# 32 v22 requests 106530 clips 213060 total khrs 11.323; N gpus for 1000 iters 13.316; 8 gpus for x iters 832.266; n unique users 25050 n pro users 22553" ] }, { "cell_type": "code", "execution_count": 20, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.277006Z", "start_time": "2024-05-16T13:59:41.252105Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:30:06.725627Z", "iopub.status.busy": "2025-02-25T01:30:06.725341Z", "iopub.status.idle": "2025-02-25T01:30:06.799577Z", "shell.execute_reply": "2025-02-25T01:30:06.799074Z", "shell.execute_reply.started": "2025-02-25T01:30:06.725613Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "positive in playlist (43307, 152)\n" ] } ], "source": [ "test_mask = (df_slice[\"preference\"] == True) & (\n", " (df_slice[\"is_in_playlist\"] == True) | (df_slice[\"concat_in_playlist\"] == True)\n", ")\n", "print(\"positive in playlist\", df_slice[test_mask].shape)" ] }, { "cell_type": "code", "execution_count": 21, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T01:30:06.800300Z", "iopub.status.busy": "2025-02-25T01:30:06.800096Z", "iopub.status.idle": "2025-02-25T01:30:07.155125Z", "shell.execute_reply": "2025-02-25T01:30:07.154624Z", "shell.execute_reply.started": "2025-02-25T01:30:06.800286Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "positive with likes (68144, 152)\n" ] } ], "source": [ "test_mask = (df_slice[\"preference\"] == True) & (\n", " (df_slice[\"upvote_count\"] >= 1)\n", ")\n", "print(\"positive with likes\", df_slice[test_mask].shape)\n", "# positive with likes (60349, 149)" ] }, { "cell_type": "code", "execution_count": 22, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T01:30:07.155967Z", "iopub.status.busy": "2025-02-25T01:30:07.155639Z", "iopub.status.idle": "2025-02-25T01:30:07.168820Z", "shell.execute_reply": "2025-02-25T01:30:07.168451Z", "shell.execute_reply.started": "2025-02-25T01:30:07.155953Z" } }, "outputs": [], "source": [ "# BREAK" ] }, { "cell_type": "code", "execution_count": 23, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.323409Z", "start_time": "2024-05-16T13:59:41.278278Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:30:07.169471Z", "iopub.status.busy": "2025-02-25T01:30:07.169358Z", "iopub.status.idle": "2025-02-25T01:30:07.213228Z", "shell.execute_reply": "2025-02-25T01:30:07.212815Z", "shell.execute_reply.started": "2025-02-25T01:30:07.169460Z" } }, "outputs": [], "source": [ "# interesting_clips_must_be_positive_mask = (\n", "# (df_slice[\"upvoted\"] == True)\n", "# | (df_slice[\"has_action\"] == True)\n", "# | (df_slice[\"part_of_concat\"] == True)\n", "# )\n", "# interesting_clips_must_be_not_negative_mask = (df_slice[\"downvoted\"] == False) # & (df_slice[\"dislike_count\"] < 1)\n", "# interesting_clips_mask = interesting_clips_must_be_positive_mask & interesting_clips_must_be_not_negative_mask\n", "# assert interesting_clips_mask.eq(df_slice[\"preference\"]).all()" ] }, { "cell_type": "code", "execution_count": 24, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.392244Z", "start_time": "2024-05-16T13:59:41.324472Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:30:07.213937Z", "iopub.status.busy": "2025-02-25T01:30:07.213685Z", "iopub.status.idle": "2025-02-25T01:30:07.270793Z", "shell.execute_reply": "2025-02-25T01:30:07.270356Z", "shell.execute_reply.started": "2025-02-25T01:30:07.213923Z" } }, "outputs": [], "source": [ "# save positive ids\n", "# positive_preference_ids = df_slice[df_slice[\"preference\"] == False][\"s3_id\"].to_json(orient='values')\n", "# with open('/home/tony/Data/Preference/7b_v2/7v_v20_full_recut_id_negative.json', 'w') as file:\n", "# file.write(positive_preference_ids)" ] }, { "cell_type": "code", "execution_count": 25, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T14:00:20.866354Z", "start_time": "2024-05-16T14:00:12.443344Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:30:07.271390Z", "iopub.status.busy": "2025-02-25T01:30:07.271277Z", "iopub.status.idle": "2025-02-25T01:30:07.499789Z", "shell.execute_reply": "2025-02-25T01:30:07.498932Z", "shell.execute_reply.started": "2025-02-25T01:30:07.271379Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "(213060, 152)\n", "task\n", " 198468\n", "extend 10518\n", "upload_extend 4074\n", "Name: count, dtype: int64\n" ] }, { "ename": "NameError", "evalue": "name 'BREAK' is not defined", "output_type": "error", "traceback": [ "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m", "\u001b[0;31mNameError\u001b[0m Traceback (most recent call last)", "Cell \u001b[0;32mIn[25], line 5\u001b[0m\n\u001b[1;32m 3\u001b[0m \u001b[38;5;28mprint\u001b[39m(df_slice\u001b[38;5;241m.\u001b[39mshape)\n\u001b[1;32m 4\u001b[0m \u001b[38;5;28mprint\u001b[39m(df_slice[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtask\u001b[39m\u001b[38;5;124m\"\u001b[39m]\u001b[38;5;241m.\u001b[39mvalue_counts())\n\u001b[0;32m----> 5\u001b[0m \u001b[43mBREAK\u001b[49m\n", "\u001b[0;31mNameError\u001b[0m: name 'BREAK' is not defined" ] } ], "source": [ "# df_slice.to_csv(\"/home/tony/Data/Preference/13b_v32/interesting_clips_v4_h_s_32_20250222_full_long_final.csv\")\n", "# df_slice.to_pickle(\"/home/tony/Data/Preference/13b_v32/interesting_clips_v4_h_s_32_20250222_full_long_final.pkl\")\n", "print(df_slice.shape)\n", "print(df_slice[\"task\"].value_counts())\n", "BREAK" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Need to kick out the ones has gpt prompt -- these are pairs with different text inputs" ] }, { "cell_type": "code", "execution_count": 26, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.932296Z", "start_time": "2024-05-16T13:59:41.932287Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:32:03.173204Z", "iopub.status.busy": "2025-02-25T01:32:03.172870Z", "iopub.status.idle": "2025-02-25T01:32:03.940712Z", "shell.execute_reply": "2025-02-25T01:32:03.940220Z", "shell.execute_reply.started": "2025-02-25T01:32:03.173188Z" } }, "outputs": [], "source": [ "# don't have continue at\n", "# df_slice[df_slice[\"continue_at\"].isna()][\"request_id\"].nunique()" ] }, { "cell_type": "code", "execution_count": 27, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.932966Z", "start_time": "2024-05-16T13:59:41.932957Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:32:03.941716Z", "iopub.status.busy": "2025-02-25T01:32:03.941508Z", "iopub.status.idle": "2025-02-25T01:32:03.998226Z", "shell.execute_reply": "2025-02-25T01:32:03.997764Z", "shell.execute_reply.started": "2025-02-25T01:32:03.941703Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "106530\n" ] } ], "source": [ "final_filtered_requests = df_slice[\"request_id\"].unique()\n", "print(len(final_filtered_requests))" ] }, { "cell_type": "code", "execution_count": 28, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.933558Z", "start_time": "2024-05-16T13:59:41.933550Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:32:03.998896Z", "iopub.status.busy": "2025-02-25T01:32:03.998710Z", "iopub.status.idle": "2025-02-25T01:32:04.016931Z", "shell.execute_reply": "2025-02-25T01:32:04.016554Z", "shell.execute_reply.started": "2025-02-25T01:32:03.998883Z" } }, "outputs": [], "source": [ "# df_slice.to_csv(\"/home/tony/Data/Preference/7b_v2/7b_before_recode_20240412\", index=False)" ] }, { "cell_type": "code", "execution_count": 29, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.934277Z", "start_time": "2024-05-16T13:59:41.934268Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:32:04.017463Z", "iopub.status.busy": "2025-02-25T01:32:04.017351Z", "iopub.status.idle": "2025-02-25T01:32:04.556122Z", "shell.execute_reply": "2025-02-25T01:32:04.555633Z", "shell.execute_reply.started": "2025-02-25T01:32:04.017452Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "105464 1066\n", "(210928, 152) (2132, 152)\n" ] } ], "source": [ "train_requests, val_requests = train_test_split(\n", " sorted(list(final_filtered_requests)), test_size=0.01, random_state=42\n", ")\n", "print(len(train_requests), len(val_requests))\n", "\n", "train_df = df_slice[df_slice[\"request_id\"].isin(set(train_requests))].copy()\n", "val_df = df_slice[df_slice[\"request_id\"].isin(set(val_requests))].copy()\n", "train_df = train_df.sort_values(by=[\"request_id\", \"preference\"])\n", "train_df = train_df# .reset_index()\n", "val_df = val_df.sort_values(by=[\"request_id\", \"preference\"])\n", "val_df = val_df#.reset_index()\n", "\n", "print(train_df.shape, val_df.shape)" ] }, { "cell_type": "code", "execution_count": 30, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T01:32:04.557376Z", "iopub.status.busy": "2025-02-25T01:32:04.557155Z", "iopub.status.idle": "2025-02-25T01:32:04.570115Z", "shell.execute_reply": "2025-02-25T01:32:04.569765Z", "shell.execute_reply.started": "2025-02-25T01:32:04.557362Z" } }, "outputs": [], "source": [ "# BREAK" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Actually make" ] }, { "cell_type": "code", "execution_count": 31, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.934954Z", "start_time": "2024-05-16T13:59:41.934946Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:32:04.570828Z", "iopub.status.busy": "2025-02-25T01:32:04.570534Z", "iopub.status.idle": "2025-02-25T01:32:04.608088Z", "shell.execute_reply": "2025-02-25T01:32:04.607691Z", "shell.execute_reply.started": "2025-02-25T01:32:04.570816Z" } }, "outputs": [], "source": [ "# val_df[[\"request_id\", \"metadata\", \"updated_at\", \"user_id\", \"preference\"]].head()" ] }, { "cell_type": "code", "execution_count": 32, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.935620Z", "start_time": "2024-05-16T13:59:41.935613Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:32:04.608624Z", "iopub.status.busy": "2025-02-25T01:32:04.608515Z", "iopub.status.idle": "2025-02-25T01:32:09.035004Z", "shell.execute_reply": "2025-02-25T01:32:09.034498Z", "shell.execute_reply.started": "2025-02-25T01:32:04.608613Z" } }, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "100%|█████████████████████████████████████████████████████████████████████████████████████████████████████| 210928/210928 [00:04<00:00, 48116.31it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "11,209 hours of 210928 clips, 6.5915 nodes, 1098.5833333333333 steps\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ "\n" ] } ], "source": [ "total_duration = 0\n", "for i, row in tqdm(train_df.iterrows(), total=len(train_df)):\n", " # we need to alternate between preference: neg, pos\n", " # print(i, row)\n", " try:\n", " assert row[\"preference\"] == (i % 2 == 1)\n", " except:\n", " print(i, row)\n", " total_duration += row[\"duration\"]\n", "print(\n", " f\"{round(total_duration / 60 / 60):,} hours of {train_df.shape[0]} clips, {train_df.shape[0] / 8 / 4 / 1000} nodes, {train_df.shape[0] / 8 / 6 / 4} steps\"\n", ")" ] }, { "cell_type": "code", "execution_count": 33, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.936268Z", "start_time": "2024-05-16T13:59:41.936260Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:32:09.035749Z", "iopub.status.busy": "2025-02-25T01:32:09.035531Z", "iopub.status.idle": "2025-02-25T01:32:32.367099Z", "shell.execute_reply": "2025-02-25T01:32:32.366626Z", "shell.execute_reply.started": "2025-02-25T01:32:09.035736Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "t_data_memmap is set to: 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ "100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2132/2132 [00:23<00:00, 92.48it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Total 1990 clips, 1 different prompts\n", "55 hours of False\n", "54 hours of True\n", "gen: 108.9 hours\n", "Error extend: 100\n", "Error upload_extend: 42\n", "Done\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ "\n" ] } ], "source": [ "make_dataset(val_df, OUT_DATA_DIR, is_val=True, npz_dir=NPZ_DIR)" ] }, { "cell_type": "code", "execution_count": 34, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.936964Z", "start_time": "2024-05-16T13:59:41.936957Z" }, "execution": { "iopub.execute_input": "2025-02-25T01:32:32.367809Z", "iopub.status.busy": "2025-02-25T01:32:32.367570Z", "iopub.status.idle": "2025-02-25T02:18:13.438015Z", "shell.execute_reply": "2025-02-25T02:18:13.437521Z", "shell.execute_reply.started": "2025-02-25T01:32:32.367796Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "t_data_memmap is set to: 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ " 7%|███████▋ | 15532/210928 [02:47<37:29, 86.85it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Overflow at 162838, 178489e6-79a1-45a6-91a7-24fecf3ed3f3_gen_cycle: 10747 > 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ " 8%|███████▉ | 15989/210928 [02:53<43:53, 74.02it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Overflow at 167400, 09b9b0e2-4aa8-44e5-89a0-9d56b798c772_gen_cycle: 7247 > 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ " 26%|███████████████████████████▏ | 54628/210928 [11:33<34:55, 74.58it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Overflow at 576316, 3b1aebc5-22e9-4617-922d-cda151fcf18f_gen_cycle: 6352 > 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ " 36%|█████████████████████████████████████▌ | 75334/210928 [16:06<30:00, 75.29it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Overflow at 796724, 0dec69f5-8d21-4800-8195-971119ed0f81_gen_cycle: 10970 > 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ " 50%|████████████████████████████████████████████████████▏ | 105809/210928 [23:05<23:06, 75.81it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Overflow at 1120376, 062479a0-f1f6-4784-9b59-196fe40849cb_gen_cycle: 6959 > 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ " 55%|█████████████████████████████████████████████████████████▌ | 116753/210928 [25:29<21:19, 73.62it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Overflow at 1234950, 0a07fc08-a351-4981-9919-98165e1577c2_gen_cycle: 8840 > 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ " 63%|█████████████████████████████████████████████████████████████████ | 132056/210928 [28:46<17:11, 76.44it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Overflow at 1396734, 0b65a603-cdcd-4c75-b74c-2ad8cc740cdb_gen_cycle: 8222 > 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ " 63%|██████████████████████████████████████████████████████████████████ | 133860/210928 [29:10<16:08, 79.59it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Overflow at 1414643, 5fad386c-3007-4c1d-8600-7ad6580dbcdb_gen_cycle: 6477 > 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ " 68%|███████████████████████████████████████████████████████████████████████▏ | 144341/210928 [31:26<14:50, 74.80it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Overflow at 1524970, 4b8b56d4-ddec-4ee6-8e8d-4c9761392842_gen_cycle: 8480 > 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ " 95%|██████████████████████████████████████████████████████████████████████████████████████████████████▉ | 200556/210928 [43:26<02:19, 74.27it/s]" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Overflow at 2119082, 57311cb7-32dd-4ba9-96b6-38941a83fba4_gen_cycle: 6455 > 6016\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ "100%|████████████████████████████████████████████████████████████████████████████████████████████████████████| 210928/210928 [45:39<00:00, 77.00it/s]\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Total 196478 clips, 34 different prompts\n", "5,427 hours of False\n", "5,321 hours of True\n", "gen: 10748.0 hours\n", "Error upload_extend: 4032\n", "Error extend: 10418\n", "Done\n" ] } ], "source": [ "make_dataset(train_df, OUT_DATA_DIR, is_val=False, npz_dir=NPZ_DIR)" ] }, { "cell_type": "markdown", "metadata": { "ExecuteTime": { "end_time": "2024-01-29T19:46:47.549860Z", "start_time": "2024-01-29T19:46:47.548015Z" } }, "source": [ "# Validation" ] }, { "cell_type": "code", "execution_count": 35, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.937879Z", "start_time": "2024-05-16T13:59:41.937870Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.438982Z", "iopub.status.busy": "2025-02-25T02:18:13.438520Z", "iopub.status.idle": "2025-02-25T02:18:13.495324Z", "shell.execute_reply": "2025-02-25T02:18:13.494892Z", "shell.execute_reply.started": "2025-02-25T02:18:13.438966Z" } }, "outputs": [], "source": [ "# verify\n", "mm = np.memmap(os.path.join(OUT_DATA_DIR, f\"data_val.bin\"), dtype=np.uint16, mode=\"r\")\n", "test_metas = read_jsonl(os.path.join(OUT_DATA_DIR, f\"meta_val.jsonl\"))\n", "test_info = read_json(os.path.join(OUT_DATA_DIR, f\"info_val.json\"))\n", "mm = mm.reshape(-1, 6016, 13)\n", "assert len(mm) == len(test_metas)\n", "assert mm[:100, :, 0].min() >= 0\n", "assert mm[:100, :, 0].max() <= 4000\n", "assert mm[:100, :, 1:].min() >= 0\n", "assert mm[:100, :, 1:].max() <= 2048" ] }, { "cell_type": "code", "execution_count": 36, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.938629Z", "start_time": "2024-05-16T13:59:41.938621Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.495992Z", "iopub.status.busy": "2025-02-25T02:18:13.495799Z", "iopub.status.idle": "2025-02-25T02:18:13.510826Z", "shell.execute_reply": "2025-02-25T02:18:13.510473Z", "shell.execute_reply.started": "2025-02-25T02:18:13.495979Z" } }, "outputs": [], "source": [ "# # randomly listen to some stuff\n", "# from suno_utils.tasks.dac_2c_12cb import preload_models as preload_codec_models\n", "# from suno_utils.tasks.dac_2c_12cb import (\n", "# encode as codec_encode,\n", "# decode_stream_to_full_audio as codec_decode,\n", "# EMBEDDING_RATE as CODEC_EMBEDDING_RATE,\n", "# decode as decode\n", "# )\n", "# os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"0\"\n", "# _ = preload_codec_models(\"/app/suno/data/dpo/models/dac_2c_25x12.pt\", device=\"cuda\")\n", "# assert len(test_metas) == len(mm)\n", "# idx_list = list(range(len(test_metas)))\n", "# # random.shuffle(idx_list)\n", "# # idx_list = [idx for idx in idx_list if \"text\" in test_metas[idx]]\n", "# print(len(mm))" ] }, { "cell_type": "code", "execution_count": 37, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.939205Z", "start_time": "2024-05-16T13:59:41.939198Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.511557Z", "iopub.status.busy": "2025-02-25T02:18:13.511260Z", "iopub.status.idle": "2025-02-25T02:18:13.548437Z", "shell.execute_reply": "2025-02-25T02:18:13.548048Z", "shell.execute_reply.started": "2025-02-25T02:18:13.511544Z" } }, "outputs": [], "source": [ "# import random\n", "# idx = random.choice(test_info[\"perference_0\"][\"idx_list\"])\n", "# assert \"original_duration_s\" in test_metas[idx]\n", "# # positive index should be shifted by 1\n", "# pos_idx = idx + 1\n", "# print(\n", "# \"tags:\",\n", "# test_metas[idx].get(\"tags\") == test_metas[pos_idx].get(\"tags\"),\n", "# test_metas[idx].get(\"tags\"),\n", "# )\n", "# arr = mm[idx, 1:].copy().astype(np.int16)[:, 1:]\n", "# pos_arr = mm[pos_idx, 1:].copy().astype(np.int16)[:, 1:]\n", "# pad_idx_arr = np.where(arr == COARSE_PAD_TOKEN)[0]\n", "# if len(pad_idx_arr) > 0:\n", "# arr = arr[: pad_idx_arr[0], :]\n", "# pos_pad_idx_arr = np.where(pos_arr == COARSE_PAD_TOKEN)[0]\n", "# if len(pos_pad_idx_arr) > 0:\n", "# pos_arr = pos_arr[: pos_pad_idx_arr[0], :]\n", "# a = decode(arr)\n", "# print(\"\\n negative example \\n\", test_metas[idx])\n", "# a.play(compress=False)\n", "# pos_a = decode(pos_arr)\n", "# print(\"\\n positive example \\n\", test_metas[pos_idx])\n", "# pos_a.play(compress=False)\n", "# print(\n", "# \"text:\",\n", "# test_metas[idx].get(\"text\") == test_metas[pos_idx].get(\"text\"),\n", "# test_metas[idx].get(\"text\"),\n", "# )" ] }, { "cell_type": "code", "execution_count": 38, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.939977Z", "start_time": "2024-05-16T13:59:41.939969Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.549077Z", "iopub.status.busy": "2025-02-25T02:18:13.548963Z", "iopub.status.idle": "2025-02-25T02:18:13.594242Z", "shell.execute_reply": "2025-02-25T02:18:13.593841Z", "shell.execute_reply.started": "2025-02-25T02:18:13.549066Z" } }, "outputs": [], "source": [ "# val_df[val_df[\"tags\"] == 'a vibrant blend of experimental jazz fusion, drum-and-bass and swagger fuzzed-out guitars']" ] }, { "cell_type": "code", "execution_count": 39, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.940610Z", "start_time": "2024-05-16T13:59:41.940603Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.595762Z", "iopub.status.busy": "2025-02-25T02:18:13.595574Z", "iopub.status.idle": "2025-02-25T02:18:13.646438Z", "shell.execute_reply": "2025-02-25T02:18:13.646023Z", "shell.execute_reply.started": "2025-02-25T02:18:13.595750Z" } }, "outputs": [], "source": [ "# from collections import Counter\n", "# c = Counter()\n", "# for _, row in df_slice.iterrows():\n", "# # print(row[\"metadata\"])\n", "# for k in ast.literal_eval(row[\"metadata\"]).keys():\n", "# c[k] += 1" ] }, { "cell_type": "code", "execution_count": 40, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.941167Z", "start_time": "2024-05-16T13:59:41.941159Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.646973Z", "iopub.status.busy": "2025-02-25T02:18:13.646860Z", "iopub.status.idle": "2025-02-25T02:18:13.687758Z", "shell.execute_reply": "2025-02-25T02:18:13.687349Z", "shell.execute_reply.started": "2025-02-25T02:18:13.646962Z" } }, "outputs": [], "source": [ "# original_npz_path = f\"/app/suno/data/dpo/7b_npz/{test_metas[idx]['id']}.npz\"\n", "# original_npz_path = \"/app/suno/data/dpo/7b_npz/729c3011-f672-4ccd-8d82-1cbf2b52ff69.npz\"\n", "# original_arr = np.load(original_npz_path)[\"v2_raw\"]" ] }, { "cell_type": "code", "execution_count": 41, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.941801Z", "start_time": "2024-05-16T13:59:41.941793Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.688480Z", "iopub.status.busy": "2025-02-25T02:18:13.688321Z", "iopub.status.idle": "2025-02-25T02:18:13.734814Z", "shell.execute_reply": "2025-02-25T02:18:13.734408Z", "shell.execute_reply.started": "2025-02-25T02:18:13.688468Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "995 0\n" ] } ], "source": [ "def validation_on_metas(input_metas):\n", "\n", " total_bad = 0\n", " total_good = 0\n", " for idx in range(len(input_metas)):\n", " if idx % 2 == 0:\n", " pos_idx = idx + 1\n", " if input_metas[idx].get(\"tags\") != input_metas[pos_idx].get(\"tags\"):\n", " # print(test_metas[idx].get(\"text\") == test_metas[pos_idx].get(\"text\"), test_metas[idx].get(\"tags\"), test_metas[pos_idx].get(\"tags\"))\n", " total_bad += 1\n", " else:\n", " total_good += 1\n", " print(total_good, total_bad)\n", " return\n", "\n", "\n", "validation_on_metas(test_metas)" ] }, { "cell_type": "code", "execution_count": 42, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.942520Z", "start_time": "2024-05-16T13:59:41.942511Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.735420Z", "iopub.status.busy": "2025-02-25T02:18:13.735248Z", "iopub.status.idle": "2025-02-25T02:18:13.848416Z", "shell.execute_reply": "2025-02-25T02:18:13.847940Z", "shell.execute_reply.started": "2025-02-25T02:18:13.735408Z" } }, "outputs": [], "source": [ "train_info = read_json(os.path.join(OUT_DATA_DIR, f\"info_tr.json\"))" ] }, { "cell_type": "code", "execution_count": 43, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.943072Z", "start_time": "2024-05-16T13:59:41.943065Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.849201Z", "iopub.status.busy": "2025-02-25T02:18:13.848902Z", "iopub.status.idle": "2025-02-25T02:18:13.870036Z", "shell.execute_reply": "2025-02-25T02:18:13.869680Z", "shell.execute_reply.started": "2025-02-25T02:18:13.849187Z" } }, "outputs": [], "source": [ "n_neg_tr = train_info[\"perference_0\"][\"idx_list\"]\n", "n_pos_tr = train_info[\"perference_1\"][\"idx_list\"]\n", "assert len(n_pos_tr) == len(n_neg_tr)" ] }, { "cell_type": "code", "execution_count": 44, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.944246Z", "start_time": "2024-05-16T13:59:41.944237Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.870716Z", "iopub.status.busy": "2025-02-25T02:18:13.870452Z", "iopub.status.idle": "2025-02-25T02:18:13.907036Z", "shell.execute_reply": "2025-02-25T02:18:13.906650Z", "shell.execute_reply.started": "2025-02-25T02:18:13.870703Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "total samples 196478 (210928, 152)\n" ] } ], "source": [ "total_iters = len(n_neg_tr) + len(n_pos_tr)\n", "print(\"total samples\", total_iters, train_df.shape)" ] }, { "cell_type": "code", "execution_count": 45, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.945249Z", "start_time": "2024-05-16T13:59:41.945241Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.907632Z", "iopub.status.busy": "2025-02-25T02:18:13.907466Z", "iopub.status.idle": "2025-02-25T02:18:13.946093Z", "shell.execute_reply": "2025-02-25T02:18:13.945701Z", "shell.execute_reply.started": "2025-02-25T02:18:13.907620Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "1 epoch per batch 4, total 3069.96875\n" ] } ], "source": [ "print(\"1 epoch per batch 4, total\", total_iters / 8 / 4 / 2)" ] }, { "cell_type": "code", "execution_count": 46, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.945972Z", "start_time": "2024-05-16T13:59:41.945964Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:13.946674Z", "iopub.status.busy": "2025-02-25T02:18:13.946506Z", "iopub.status.idle": "2025-02-25T02:18:14.192736Z", "shell.execute_reply": "2025-02-25T02:18:14.192214Z", "shell.execute_reply.started": "2025-02-25T02:18:13.946662Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Submitted batch job 3598\n" ] } ], "source": [ "!cd /home/tony/Work/tony/slurm/13b_dpo && sbatch sbatch_ipo_13b_s32" ] }, { "cell_type": "code", "execution_count": 47, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T02:18:14.193619Z", "iopub.status.busy": "2025-02-25T02:18:14.193377Z", "iopub.status.idle": "2025-02-25T02:18:14.216946Z", "shell.execute_reply": "2025-02-25T02:18:14.216582Z", "shell.execute_reply.started": "2025-02-25T02:18:14.193605Z" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Cache kept!\n" ] } ], "source": [ "import shutil\n", "\n", "# Basic file copy\n", "shutil.copy('/home/tony/Work/tony/Preference/make_dataset_13b_v3p5data_s32_cycle.ipynb', os.path.join(OUT_DATA_DIR, \"make_dataset.ipynb\"))\n", "print(\"Cache kept!\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# some gymathtics loading prev data" ] }, { "cell_type": "code", "execution_count": 48, "metadata": { "ExecuteTime": { "end_time": "2024-05-16T13:59:41.946562Z", "start_time": "2024-05-16T13:59:41.946555Z" }, "execution": { "iopub.execute_input": "2025-02-25T02:18:14.217575Z", "iopub.status.busy": "2025-02-25T02:18:14.217409Z", "iopub.status.idle": "2025-02-25T02:18:14.249433Z", "shell.execute_reply": "2025-02-25T02:18:14.249073Z", "shell.execute_reply.started": "2025-02-25T02:18:14.217562Z" } }, "outputs": [], "source": [ "# prev_v3_data = \"/app/suno/data/dpo/7v_v20_full/\"\n", "\n", "# test_val_metas = read_jsonl(os.path.join(prev_v3_data, f\"meta_val.jsonl\"))\n", "# test_tr_metas = read_jsonl(os.path.join(prev_v3_data, f\"meta_tr.jsonl\"))\n", "\n", "# all_ids = set()\n", "# for meta in test_val_metas:\n", "# all_ids.add(meta[\"id\"])\n", "# for meta in test_tr_metas:\n", "# all_ids.add(meta[\"id\"])\n", "# print(len(all_ids), len(test_val_metas) + len(test_tr_metas))\n", "\n", "# all_ids = list(all_ids)\n", "# with open(\"/home/tony/Data/Preference/7b_v2/7v_v20_full_recut_id.json\", \"w\") as fp:\n", "# json.dump(all_ids, fp)" ] }, { "cell_type": "code", "execution_count": 49, "metadata": { "execution": { "iopub.execute_input": "2025-02-25T02:18:14.250034Z", "iopub.status.busy": "2025-02-25T02:18:14.249869Z", "iopub.status.idle": "2025-02-25T02:18:14.285192Z", "shell.execute_reply": "2025-02-25T02:18:14.284819Z", "shell.execute_reply.started": "2025-02-25T02:18:14.250022Z" } }, "outputs": [], "source": [ "# train_df[\"lang\"].value_counts()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "Python 3 (ipykernel)", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.10.15" }, "toc": { "base_numbering": 1, "nav_menu": {}, "number_sections": true, "sideBar": true, "skip_h1_title": false, "title_cell": "Table of Contents", "title_sidebar": "Contents", "toc_cell": false, "toc_position": {}, "toc_section_display": true, "toc_window_display": false } }, "nbformat": 4, "nbformat_minor": 4 }