{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"ExecuteTime": {
"end_time": "2024-05-14T00:48:09.417798Z",
"start_time": "2024-05-14T00:48:09.416342Z"
}
},
"outputs": [],
"source": [
"import os\n",
"\n",
"os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"2\""
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"ExecuteTime": {
"end_time": "2024-05-14T00:48:29.972815Z",
"start_time": "2024-05-14T00:48:09.419319Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"WILL USE FLASH ATTN: True\n",
"model loaded: /app/suno/checkpoints/2024-03-18_01-18-00/last_ckpt_infer.pt\n",
"model loaded: 5869.4M params\n",
"Compiled model\n"
]
}
],
"source": [
"import numpy as np\n",
"import torch.nn.functional as F\n",
"from suno_utils.audio import Audio\n",
"from suno_utils.gpt.chirp_v2_5 import (\n",
" GenerationConfig,\n",
" decode_stream,\n",
" preload_models,\n",
" prep_gconf,\n",
")\n",
"from suno_utils.gpt.generation_engine import (\n",
" align_codes,\n",
" make_request,\n",
")\n",
"from suno_utils.gpt.engine import Engine\n",
"import torch\n",
"from tqdm import tqdm\n",
"\n",
"preload_models(load_gpt=False, load_semantic=False, load_codec_device=\"cuda\")\n",
"\n",
"gpt_ckpt_path = \"/app/suno/checkpoints/2024-03-18_01-18-00/last_ckpt_infer.pt\"\n",
"engine = Engine(\n",
" gpt_ckpt_path,\n",
" \"/home/victor/data/models/chirp_v2/tokenizer_60k.json\",\n",
" max_sequences=16,\n",
" compile=True,\n",
")\n",
"# engine.start()"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {
"ExecuteTime": {
"end_time": "2024-05-14T00:48:29.978751Z",
"start_time": "2024-05-14T00:48:29.975794Z"
}
},
"outputs": [],
"source": [
"general_config = dict(\n",
" cfg_coef=1.0, # no text cfg for dpo stream\n",
" min_eos_p=0.1,\n",
" eos_pad_duration_s=0,\n",
" # cfg_coef_tags=0.0,\n",
" # cfg_coef_tags_max_steps=None, # collect the data for now\n",
" cfg_coef_tags=2,\n",
" cfg_coef_neg_tags=-2,\n",
" text_neg_tags=\"repetitive, loop\",\n",
" n_repeat_tags=3,\n",
" use_whisper=False,\n",
" text_start_control_tags=\"{start}\",\n",
" # text_end_control_tags=\"{end}\",\n",
" random_seed=42,\n",
" n_batch=1,\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"ExecuteTime": {
"end_time": "2024-05-14T00:48:30.044690Z",
"start_time": "2024-05-14T00:48:29.980125Z"
}
},
"outputs": [],
"source": [
"def generate_audio_with_engine(text, text_tags):\n",
" cfg = GenerationConfig(\n",
" text=text, text_tags=text_tags, max_gen_duration_s=120, **general_config\n",
" )\n",
" preped_cfg = prep_gconf(cfg)\n",
" model_conf = engine.model.config\n",
" requests = []\n",
" for i in range(1, 5):\n",
" request = make_request(\n",
" f\"{i}\", preped_cfg, engine.model.config, engine.tokenizer\n",
" )\n",
" requests.append(request)\n",
"\n",
" jobs = engine.run_request(requests)\n",
"\n",
" for job in jobs:\n",
" stream = engine.token_generator(job)\n",
" audios = []\n",
" for audio in decode_stream(align_codes(tqdm(stream), model_conf)):\n",
" audios.append(audio[0])\n",
" audio_continued = Audio.concatenate(audios)\n",
" audio_continued.play()\n",
" return"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"ExecuteTime": {
"end_time": "2024-05-14T00:51:03.149458Z",
"start_time": "2024-05-14T00:48:30.046811Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Adding 4 requests\n",
"Adding request 1 with 4 streams\n",
"add_request releasing lock for request 1\n",
"Adding request 2 with 4 streams\n",
"add_request releasing lock for request 2\n",
"Adding request 3 with 4 streams\n",
"add_request releasing lock for request 3\n",
"Adding request 4 with 4 streams\n",
"add_request releasing lock for request 4\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"[W514 00:48:44.438526042 schema_info.cpp:332] Warning: alias::a appears twice in same argument list which will make aliasing checks more conservative. (function operator())\n",
"skipping cudagraphs due to mutated inputs (1 instances)\n",
"skipping cudagraphs due to mutated inputs (1 instances)\n",
"skipping cudagraphs due to mutated inputs (1 instances)\n",
"skipping cudagraphs due to mutated inputs (1 instances)\n",
"W0514 00:48:55.707000 140234961654400 torch/_inductor/utils.py:1123] [6/0] DeviceCopy in input program\n",
"W0514 00:48:55.715000 140234961654400 torch/_inductor/utils.py:1123] [6/0] DeviceCopy in input program\n",
"skipping cudagraphs due to skipping cudagraphs due to cpu device (copy)\n",
"/home/tony/anaconda3/envs/suno_env/lib/python3.10/site-packages/torch/cuda/graphs.py:82: UserWarning: The CUDA Graph is empty. This usually means that the graph was attempted to be captured on wrong device or stream. (Triggered internally at ../aten/src/ATen/cuda/CUDAGraph.cpp:235.)\n",
" super().capture_end()\n",
"skipping cudagraphs due to mutated inputs (1 instances)\n",
"skipping cudagraphs due to mutated inputs (1 instances)\n",
"skipping cudagraphs due to mutated inputs (1 instances)\n",
"skipping cudagraphs due to mutated inputs (1 instances)\n",
"W0514 00:49:27.517000 140234961654400 torch/_inductor/utils.py:1123] [6/1] DeviceCopy in input program\n",
"W0514 00:49:27.522000 140234961654400 torch/_inductor/utils.py:1123] [6/1] DeviceCopy in input program\n",
"skipping cudagraphs due to skipping cudagraphs due to cpu device (copy)\n",
"/home/tony/anaconda3/envs/suno_env/lib/python3.10/site-packages/torch/cuda/graphs.py:82: UserWarning: The CUDA Graph is empty. This usually means that the graph was attempted to be captured on wrong device or stream. (Triggered internally at ../aten/src/ATen/cuda/CUDAGraph.cpp:235.)\n",
" super().capture_end()\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Job 2 found EOS at step 1166\n",
"Job 3 found EOS at step 1183\n",
"Job 2 completed at step 1271\n",
"Job 2 completed\n",
"Job 3 completed at step 1288\n",
"Job 3 completed\n",
"Job 1 found EOS at step 3000\n",
"Job 4 found EOS at step 3000\n",
"Job 1 completed\n",
"Job 4 completed\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"3105it [00:02, 1331.17it/s]\n"
]
},
{
"data": {
"text/html": [
"\n",
"\n"
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"1271it [00:00, 1453.87it/s]\n"
]
},
{
"data": {
"text/html": [
"\n",
"\n"
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"1288it [00:00, 1469.86it/s]\n"
]
},
{
"data": {
"text/html": [
"\n",
"\n"
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"3105it [00:02, 1381.48it/s]\n"
]
},
{
"data": {
"text/html": [
"\n",
"\n"
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"text = \"\"\"\n",
"[Verse 1]\n",
"Almost Heaven, West Virginia\n",
"Blue Ridge Mountains, Shenandoah River\n",
"Life is old there, older than the trees\n",
"Younger than the mountains, growing like a breeze\n",
"\n",
"[Chorus]\n",
"Country roads, take me home\n",
"To the place I belong\n",
"West Virginia, mountain mama\n",
"Take me home, country roads\n",
"\"\"\"\n",
"text_tags = \"bluegrass, female\"\n",
"\n",
"generate_audio_with_engine(text, text_tags)"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {
"ExecuteTime": {
"end_time": "2024-05-14T00:52:42.991975Z",
"start_time": "2024-05-14T00:51:03.151731Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Adding 4 requests\n",
"Adding request 1 with 4 streams\n",
"add_request releasing lock for request 1\n",
"Adding request 2 with 4 streams\n",
"add_request releasing lock for request 2\n",
"Adding request 3 with 4 streams\n",
"add_request releasing lock for request 3\n",
"Adding request 4 with 4 streams\n",
"add_request releasing lock for request 4\n",
"Job 2 found EOS at step 1680\n",
"Job 2 completed at step 1785\n",
"Job 2 completed\n",
"Job 1 found EOS at step 2098\n",
"Job 1 completed at step 2203\n",
"Job 1 completed\n",
"Job 3 found EOS at step 2533\n",
"Job 3 completed at step 2638\n",
"Job 3 completed\n",
"Job 4 found EOS at step 2861\n",
"Job 4 completed at step 2966\n",
"Job 4 completed\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"2203it [00:01, 1394.27it/s]\n"
]
},
{
"data": {
"text/html": [
"\n",
"\n"
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"1785it [00:01, 1416.66it/s]\n"
]
},
{
"data": {
"text/html": [
"\n",
"\n"
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"2638it [00:01, 1378.89it/s]\n"
]
},
{
"data": {
"text/html": [
"\n",
"\n"
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"2966it [00:02, 1377.40it/s]\n"
]
},
{
"data": {
"text/html": [
"\n",
"