diff --git "a/actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/output/experiment/rwkv-x-exp/v5-headsize2x/v5-L6-D4096-E1e-1-ctx4k-part1.ipynb" "b/actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/output/experiment/rwkv-x-exp/v5-headsize2x/v5-L6-D4096-E1e-1-ctx4k-part1.ipynb" new file mode 100644--- /dev/null +++ "b/actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/output/experiment/rwkv-x-exp/v5-headsize2x/v5-L6-D4096-E1e-1-ctx4k-part1.ipynb" @@ -0,0 +1,116224 @@ +{ + "cells": [ + { + "attachments": {}, + "cell_type": "markdown", + "id": "3cb7ba82", + "metadata": { + "papermill": { + "duration": 0.003253, + "end_time": "2023-08-28T18:45:25.049803", + "exception": false, + "start_time": "2023-08-28T18:45:25.046550", + "status": "completed" + }, + "tags": [] + }, + "source": [ + "# RWKV v5-headsize2x / embedding init-range 1e-01 / 4k\n", + "\n", + "- 6 layers\n", + "- 4096 embedding size\n", + "\n", + "Going through the modified memory training for v5 models, across various initial embedding model weights\n", + "\n", + "**Note:** This project assumes you have the rwkv-infctx conda env setup" + ] + }, + { + "attachments": {}, + "cell_type": "markdown", + "id": "4aba7083", + "metadata": { + "papermill": { + "duration": 0.002594, + "end_time": "2023-08-28T18:45:25.054857", + "exception": false, + "start_time": "2023-08-28T18:45:25.052263", + "status": "completed" + }, + "tags": [] + }, + "source": [ + "# Basic Setup" + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "id": "b74e122f", + "metadata": { + "execution": { + "iopub.execute_input": "2023-08-28T18:45:25.060868Z", + "iopub.status.busy": "2023-08-28T18:45:25.060669Z", + "iopub.status.idle": "2023-08-28T18:45:25.776404Z", + "shell.execute_reply": "2023-08-28T18:45:25.775244Z" + }, + "papermill": { + "duration": 0.720741, + "end_time": "2023-08-28T18:45:25.777978", + "exception": false, + "start_time": "2023-08-28T18:45:25.057237", + "status": "completed" + }, + "tags": [] + }, + "outputs": [], + "source": [ + "# First lets setup the various directories, and init the model\n", + "!mkdir -p ../../../../model/\n", + "!mkdir -p ../../../../datapath/\n", + "!mkdir -p ../../../../checkpoint/" + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "id": "a8b3d9f6", + "metadata": { + "execution": { + "iopub.execute_input": "2023-08-28T18:45:25.784847Z", + "iopub.status.busy": "2023-08-28T18:45:25.784172Z", + "iopub.status.idle": "2023-08-28T18:45:28.697467Z", + "shell.execute_reply": "2023-08-28T18:45:28.696485Z" + }, + "papermill": { + "duration": 2.918657, + "end_time": "2023-08-28T18:45:28.699319", + "exception": false, + "start_time": "2023-08-28T18:45:25.780662", + "status": "completed" + }, + "tags": [] + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\r\n", + "\u001b[0m" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "\r\n", + "\u001b[1m[\u001b[0m\u001b[34;49mnotice\u001b[0m\u001b[1;39;49m]\u001b[0m\u001b[39;49m A new release of pip is available: \u001b[0m\u001b[31;49m23.1.2\u001b[0m\u001b[39;49m -> \u001b[0m\u001b[32;49m23.2.1\u001b[0m\r\n", + "\u001b[1m[\u001b[0m\u001b[34;49mnotice\u001b[0m\u001b[1;39;49m]\u001b[0m\u001b[39;49m To update, run: \u001b[0m\u001b[32;49mpython3.11 -m pip install --upgrade pip\u001b[0m\r\n" + ] + } + ], + "source": [ + "# Additional dependencies for eval stuff\n", + "!pip install -q aiocsv aiofiles" + ] + }, + { + "cell_type": "code", + "execution_count": 3, + "id": "68bfff15", + "metadata": { + "execution": { + "iopub.execute_input": "2023-08-28T18:45:28.706431Z", + "iopub.status.busy": "2023-08-28T18:45:28.706239Z", + "iopub.status.idle": "2023-08-28T18:45:28.712644Z", + "shell.execute_reply": "2023-08-28T18:45:28.711968Z" + }, + "papermill": { + "duration": 0.011435, + "end_time": "2023-08-28T18:45:28.713756", + "exception": false, + "start_time": "2023-08-28T18:45:28.702321", + "status": "completed" + }, + "tags": [] + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "DEEPSPEED_STRAT: deepspeed_stage_1\n", + "ENABLE_WANDB: True\n", + "GPU_DEVICES: auto\n", + "NOTEBOOK_DIR: /actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/notebook/experiment/rwkv-x-exp/v5-headsize2x\n", + "INFERENCE_DIR: /actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/RWKV-v5headsize2x\n", + "TRAINER_DIR: /actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/RWKV-v5headsize2x\n", + "PROJECT_DIR: /actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer\n" + ] + } + ], + "source": [ + "DEEPSPEED_STRAT=\"deepspeed_stage_1\"\n", + "GPU_DEVICES=\"auto\"\n", + "ENABLE_WANDB=True\n", + "\n", + "RWKV_WAVENET_LAYERS=1\n", + "\n", + "EMBED_SCALE=0.1\n", + "EMBED_SCALE_LABEL=str(EMBED_SCALE).replace(\".\", \"_\")\n", + "\n", + "LAYER_COUNT=6\n", + "EMBED_DIM=4096\n", + "\n", + "WANDB_PREFIX=f\"v5-hs2x-L{LAYER_COUNT}-D{EMBED_DIM}-E{EMBED_SCALE}\"\n", + "FILENAME_PREFIX=f\"v5-hs2x-L{LAYER_COUNT}-D{EMBED_DIM}-E{EMBED_SCALE_LABEL}\"\n", + "\n", + "print(\"DEEPSPEED_STRAT:\", DEEPSPEED_STRAT)\n", + "print(\"ENABLE_WANDB:\", ENABLE_WANDB)\n", + "print(\"GPU_DEVICES:\", GPU_DEVICES)\n", + "\n", + "if ENABLE_WANDB:\n", + " WANDB_MODE=\"online\"\n", + "else:\n", + " WANDB_MODE=\"disabled\"\n", + "\n", + "# Computing the notebook, and various paths\n", + "import os\n", + "NOTEBOOK_DIR=os.path.dirname(os.path.abspath(\"__file__\"))\n", + "PROJECT_DIR=os.path.abspath(os.path.join(NOTEBOOK_DIR, \"../../../../\"))\n", + "TRAINER_DIR=os.path.abspath(os.path.join(PROJECT_DIR, \"./RWKV-v5headsize2x/\"))\n", + "INFERENCE_DIR=os.path.abspath(os.path.join(PROJECT_DIR, \"./RWKV-v5headsize2x/\"))\n", + "\n", + "print(\"NOTEBOOK_DIR:\", NOTEBOOK_DIR)\n", + "print(\"INFERENCE_DIR:\", INFERENCE_DIR)\n", + "print(\"TRAINER_DIR:\", TRAINER_DIR)\n", + "print(\"PROJECT_DIR:\", PROJECT_DIR)" + ] + }, + { + "cell_type": "code", + "execution_count": 4, + "id": "d6d231c3", + "metadata": { + "execution": { + "iopub.execute_input": "2023-08-28T18:45:28.719844Z", + "iopub.status.busy": "2023-08-28T18:45:28.719677Z", + "iopub.status.idle": "2023-08-28T18:46:57.146232Z", + "shell.execute_reply": "2023-08-28T18:46:57.145151Z" + }, + "papermill": { + "duration": 88.432008, + "end_time": "2023-08-28T18:46:57.148448", + "exception": false, + "start_time": "2023-08-28T18:45:28.716440", + "status": "completed" + }, + "tags": [] + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Setting ds_accelerator to cuda (auto detect)\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[RWKV.model] Running RWKV model using 'torch-jit' with torch '2.0.1+cu118'\r\n", + "---- Initializing model ----\r\n", + "No of layers: 6\r\n", + "Embedding size: 4096\r\n", + "Output model path: ../model/L6-D4096-E0_1-neox-v5base-init.pth\r\n", + "Vocab size: 50277\r\n", + "Emb scale: 0.1\r\n", + "Note: this process takes a significant time (and ram) for large models\r\n", + "---- ----- ----\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "50277 4096 -0.1 emb.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.0.att.receptance.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.0.att.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.0.att.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.0.att.output.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "16384 4096 1.0 blocks.0.ffn.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.0.ffn.receptance.weight\r\n", + "4096 16384 0 blocks.0.ffn.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.1.att.receptance.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.1.att.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.1.att.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.1.att.output.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "16384 4096 1.0 blocks.1.ffn.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.1.ffn.receptance.weight\r\n", + "4096 16384 0 blocks.1.ffn.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.2.att.receptance.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.2.att.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.2.att.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.2.att.output.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "16384 4096 1.0 blocks.2.ffn.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.2.ffn.receptance.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 16384 0 blocks.2.ffn.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.3.att.receptance.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.3.att.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.3.att.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.3.att.output.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "16384 4096 1.0 blocks.3.ffn.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.3.ffn.receptance.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 16384 0 blocks.3.ffn.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.4.att.receptance.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.4.att.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.4.att.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.4.att.output.weight\r\n", + "16384 4096 1.0 blocks.4.ffn.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.4.ffn.receptance.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 16384 0 blocks.4.ffn.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.5.att.receptance.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.5.att.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 1.0 blocks.5.att.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.5.att.output.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "16384 4096 1.0 blocks.5.ffn.key.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 4096 0 blocks.5.ffn.receptance.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "4096 16384 0 blocks.5.ffn.value.weight\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "50277 4096 0.5 head.weight\r\n" + ] + } + ], + "source": [ + "# Init the model\n", + "!cd \"{TRAINER_DIR}\" && \\\n", + " export RWKV_WAVENET_LAYERS=\"{RWKV_WAVENET_LAYERS}\" && \\\n", + " python3 ./init_model.py \\\n", + " --n_layer \"{LAYER_COUNT}\" --n_embd \"{EMBED_DIM}\" \\\n", + " --emb-scale \"{EMBED_SCALE}\" \\\n", + " --vocab_size neox --skip-if-exists \\\n", + " \"../model/L{LAYER_COUNT}-D{EMBED_DIM}-E{EMBED_SCALE_LABEL}-neox-v5base-init.pth\"" + ] + }, + { + "cell_type": "markdown", + "id": "b95d87fb", + "metadata": { + "papermill": { + "duration": 0.004479, + "end_time": "2023-08-28T18:46:57.159346", + "exception": false, + "start_time": "2023-08-28T18:46:57.154867", + "status": "completed" + }, + "tags": [] + }, + "source": [ + "## Enwiki Stage 1 : Foundation 4k model training" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "id": "fec23852", + "metadata": { + "execution": { + "iopub.execute_input": "2023-08-28T18:46:57.169965Z", + "iopub.status.busy": "2023-08-28T18:46:57.169753Z", + "iopub.status.idle": "2023-08-28T18:47:08.337785Z", + "shell.execute_reply": "2023-08-28T18:47:08.336964Z" + }, + "papermill": { + "duration": 11.175653, + "end_time": "2023-08-28T18:47:08.339555", + "exception": false, + "start_time": "2023-08-28T18:46:57.163902", + "status": "completed" + }, + "tags": [] + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Found cached dataset parquet (/actions-runner/.cache/huggingface/datasets/teven___parquet/teven--enwiki_100k-1359e81b212c2dd6/0.0.0/14a00e99c0d15a23649d0db8944380ac81082d4b021f398733dd84f3a6c569a7)\r\n", + "\r", + " 0%| | 0/1 [00:00=12.1), as this is known to have freeze issues\r\n", + "# - The terms used in wandb / the progress bar can be confusing, see the github README.md for beter clarifications\r\n", + "# - When resuming from checkpoint, the estimated time is inaccurate\r\n", + "#\r\n", + "LOCAL_RANK: 3 - CUDA_VISIBLE_DEVICES: [0,1,2,3,4,5,6,7]\r\n", + "\r\n", + "[RWKV.model] Configuring optimizer with\r\n", + " - lr_init: 6.000e-04 (0.0006)\r\n", + " - lr_final: 4.000e-04 (0.0004)\r\n", + "\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Detected CUDA files, patching ldflags\r\n", + "Emitting ninja build file /root/.cache/torch_extensions/py311_cu118/fused_adam/build.ninja...\r\n", + "Building extension module fused_adam...\r\n", + "Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)\r\n", + "ninja: no work to do.\r\n", + "Loading extension module fused_adam...\r\n", + "Time to load fused_adam op: 0.07090473175048828 seconds\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Loading extension module fused_adam...\r\n", + "Loading extension module fused_adam...\r\n", + "Loading extension module fused_adam...\r\n", + "Loading extension module fused_adam...\r\n", + "Loading extension module fused_adam...\r\n", + "Time to load fused_adam op: 0.10126924514770508 seconds\r\n", + "Time to load fused_adam op: 0.10122466087341309 seconds\r\n", + "Loading extension module fused_adam...\r\n", + "Loading extension module fused_adam...\r\n", + "Time to load fused_adam op: 0.10141754150390625 seconds\r\n", + "Time to load fused_adam op: 0.1015462875366211 seconds\r\n", + "Time to load fused_adam op: 0.10146403312683105 seconds\r\n", + "Time to load fused_adam op: 0.10181498527526855 seconds\r\n", + "Time to load fused_adam op: 0.10191535949707031 seconds\r\n", + "Loading `train_dataloader` to estimate number of stepping batches.\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Emitting ninja build file /root/.cache/torch_extensions/py311_cu118/utils/build.ninja...\r\n", + "Building extension module utils...\r\n", + "Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "ninja: no work to do.\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.0714881420135498 seconds\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Loading extension module utils...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.10209321975708008 seconds\r\n", + "Time to load utils op: 0.10218286514282227 seconds\r\n", + "Loading extension module utils...\r\n", + "Loading extension module utils...\r\n", + "Loading extension module utils...\r\n", + "Loading extension module utils...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.1021718978881836 seconds\r\n", + "Time to load utils op: 0.10290384292602539 seconds\r\n", + "Time to load utils op: 0.10239219665527344 seconds\r\n", + "Time to load utils op: 0.10289859771728516 seconds\r\n", + "Time to load utils op: 0.10258913040161133 seconds\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 2 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 1 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 7 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 3 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n", + "Rank: 5 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 4 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 6 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 0 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.0005891323089599609 seconds\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.00074005126953125 seconds\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Time to load utils op: 0.0006580352783203125 seconds\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.0008976459503173828 seconds\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.0006477832794189453 seconds\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.0007245540618896484 seconds\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.0014486312866210938 seconds\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.0009706020355224609 seconds\r\n", + "\r\n", + " | Name | Type | Params\r\n", + "--------------------------------------\r\n", + "0 | emb | Embedding | 205 M \r\n", + "1 | blocks | ModuleList | 1.3 B \r\n", + "2 | ln_out | LayerNorm | 8.2 K \r\n", + "3 | head | Linear | 205 M \r\n", + "--------------------------------------\r\n", + "1.7 B Trainable params\r\n", + "0 Non-trainable params\r\n", + "1.7 B Total params\r\n", + "6,883.117 Total estimated model params size (MB)\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "\r", + "Training: 0it [00:00, ?it/s]\r", + "Training: 0%| | 0/10186 [00:00\r\n", + " asyncio.run(main_function())\r\n", + " File \"/usr/lib/python3.11/asyncio/runners.py\", line 190, in run\r\n", + " return runner.run(main)\r\n", + " ^^^^^^^^^^^^^^^^\r\n", + " File \"/usr/lib/python3.11/asyncio/runners.py\", line 118, in run\r\n", + " return self._loop.run_until_complete(task)\r\n", + " ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n", + " File \"/usr/lib/python3.11/asyncio/base_events.py\", line 653, in run_until_complete\r\n", + " return future.result()\r\n", + " ^^^^^^^^^^^^^^^\r\n", + " File \"/actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/notebook/experiment/rwkv-x-exp/v5-headsize2x/../memory_script/eval_v5_memory_guided.py\", line 58, in main_function\r\n", + " model = SimpleRWKV(model_path, device=\"cuda\")\r\n", + " ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n", + " File \"/actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/RWKV-v5/src/model.py\", line 1378, in __init__\r\n", + " self.model = RWKV(**model_config)\r\n", + " ^^^^^^^^^^^^^^^^^^^^\r\n", + " File \"/actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/RWKV-v5/src/model.py\", line 650, in __init__\r\n", + " self.load_state_dict(model_weights)\r\n", + " File \"/usr/local/lib/python3.11/dist-packages/torch/nn/modules/module.py\", line 2041, in load_state_dict\r\n", + " raise RuntimeError('Error(s) in loading state_dict for {}:\\n\\t{}'.format(\r\n", + "RuntimeError: Error(s) in loading state_dict for RWKV:\r\n", + "\tsize mismatch for blocks.0.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.0.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.1.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.1.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.2.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.2.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.3.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.3.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.4.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.4.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.5.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.5.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n" + ] + } + ], + "source": [ + "# Lets do a quick memory test\n", + "!export RWKV_WAVENET_LAYERS=\"{RWKV_WAVENET_LAYERS}\" && \\\n", + " python3 ../memory_script/eval_v5_memory_guided.py \"{PROJECT_DIR}/model/{FILENAME_PREFIX}-enwiki-4k.pth\"" + ] + }, + { + "attachments": {}, + "cell_type": "markdown", + "id": "1e197c3b", + "metadata": { + "papermill": { + "duration": 0.584635, + "end_time": "2023-08-29T01:43:14.328059", + "exception": false, + "start_time": "2023-08-29T01:43:13.743424", + "status": "completed" + }, + "tags": [] + }, + "source": [ + "# Enwiki Stage 2 : Basic Instruct Tuning" + ] + }, + { + "cell_type": "code", + "execution_count": 10, + "id": "9f645ad6", + "metadata": { + "execution": { + "iopub.execute_input": "2023-08-29T01:43:15.622056Z", + "iopub.status.busy": "2023-08-29T01:43:15.621812Z", + "iopub.status.idle": "2023-08-29T01:43:22.600320Z", + "shell.execute_reply": "2023-08-29T01:43:22.599545Z" + }, + "papermill": { + "duration": 7.694583, + "end_time": "2023-08-29T01:43:22.602460", + "exception": false, + "start_time": "2023-08-29T01:43:14.907877", + "status": "completed" + }, + "tags": [] + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Found cached dataset parquet (/actions-runner/.cache/huggingface/datasets/c-s-ale___parquet/c-s-ale--dolly-15k-instruction-alpaca-format-9dfbb23260d63d9d/0.0.0/14a00e99c0d15a23649d0db8944380ac81082d4b021f398733dd84f3a6c569a7)\r\n", + "\r", + " 0%| | 0/1 [00:00=12.1), as this is known to have freeze issues\r\n", + "# - The terms used in wandb / the progress bar can be confusing, see the github README.md for beter clarifications\r\n", + "# - When resuming from checkpoint, the estimated time is inaccurate\r\n", + "#\r\n", + "\r\n", + "[RWKV.model] Configuring optimizer with\r\n", + " - lr_init: 4.000e-04 (0.0004)\r\n", + " - lr_final: 3.000e-04 (0.0003)\r\n", + "\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Detected CUDA files, patching ldflags\r\n", + "Emitting ninja build file /root/.cache/torch_extensions/py311_cu118/fused_adam/build.ninja...\r\n", + "Building extension module fused_adam...\r\n", + "Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)\r\n", + "ninja: no work to do.\r\n", + "Loading extension module fused_adam...\r\n", + "Time to load fused_adam op: 0.07155156135559082 seconds\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Loading extension module fused_adam...\r\n", + "Loading extension module fused_adam...\r\n", + "Loading extension module fused_adam...\r\n", + "Loading extension module fused_adam...\r\n", + "Loading extension module fused_adam...\r\n", + "Loading extension module fused_adam...\r\n", + "Loading extension module fused_adam...\r\n", + "Time to load fused_adam op: 0.10176849365234375 seconds\r\n", + "Time to load fused_adam op: 0.10148835182189941 seconds\r\n", + "Time to load fused_adam op: 0.1015634536743164 seconds\r\n", + "Time to load fused_adam op: 0.10209345817565918 seconds\r\n", + "Time to load fused_adam op: 0.10170292854309082 seconds\r\n", + "Time to load fused_adam op: 0.10150504112243652 seconds\r\n", + "Time to load fused_adam op: 0.10160398483276367 seconds\r\n", + "Loading `train_dataloader` to estimate number of stepping batches.\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Emitting ninja build file /root/.cache/torch_extensions/py311_cu118/utils/build.ninja...\r\n", + "Building extension module utils...\r\n", + "Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)\r\n", + "ninja: no work to do.\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Loading extension module utils...\r\n", + "Time to load utils op: 0.07130551338195801 seconds\r\n", + "Loading extension module utils...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.10222482681274414 seconds\r\n", + "Time to load utils op: 0.10239815711975098 seconds\r\n", + "Loading extension module utils...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.10255551338195801 seconds\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.1025240421295166 seconds\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.10241293907165527 seconds\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.10340571403503418 seconds\r\n", + "Time to load utils op: 0.1028285026550293 seconds\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 6 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 4 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 1 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 3 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 0 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 7 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 5 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Rank: 2 partition count [8, 8] and sizes[(215097344, False), (48, False)] \r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.0006301403045654297 seconds\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Time to load utils op: 0.0006008148193359375 seconds\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Time to load utils op: 0.0006322860717773438 seconds\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "Time to load utils op: 0.0006577968597412109 seconds\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.0010302066802978516 seconds\r\n", + "Time to load utils op: 0.0007059574127197266 seconds\r\n", + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.001132965087890625 seconds\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Using /root/.cache/torch_extensions/py311_cu118 as PyTorch extensions root...\r\n", + "No modifications detected for re-loaded extension module utils, skipping build step...\r\n", + "Loading extension module utils...\r\n", + "Time to load utils op: 0.0008440017700195312 seconds\r\n", + "\r\n", + " | Name | Type | Params\r\n", + "--------------------------------------\r\n", + "0 | emb | Embedding | 205 M \r\n", + "1 | blocks | ModuleList | 1.3 B \r\n", + "2 | ln_out | LayerNorm | 8.2 K \r\n", + "3 | head | Linear | 205 M \r\n", + "--------------------------------------\r\n", + "1.7 B Trainable params\r\n", + "0 Non-trainable params\r\n", + "1.7 B Total params\r\n", + "6,883.117 Total estimated model params size (MB)\r\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "\r", + "Training: 0it [00:00, ?it/s]\r", + "Training: 0%| | 0/1867 [00:00\r\n", + " asyncio.run(main_function())\r\n", + " File \"/usr/lib/python3.11/asyncio/runners.py\", line 190, in run\r\n", + " return runner.run(main)\r\n", + " ^^^^^^^^^^^^^^^^\r\n", + " File \"/usr/lib/python3.11/asyncio/runners.py\", line 118, in run\r\n", + " return self._loop.run_until_complete(task)\r\n", + " ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n", + " File \"/usr/lib/python3.11/asyncio/base_events.py\", line 653, in run_until_complete\r\n", + " return future.result()\r\n", + " ^^^^^^^^^^^^^^^\r\n", + " File \"/actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/notebook/experiment/rwkv-x-exp/v5-headsize2x/../memory_script/eval_v5_memory_guided.py\", line 58, in main_function\r\n", + " model = SimpleRWKV(model_path, device=\"cuda\")\r\n", + " ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n", + " File \"/actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/RWKV-v5/src/model.py\", line 1378, in __init__\r\n", + " self.model = RWKV(**model_config)\r\n", + " ^^^^^^^^^^^^^^^^^^^^\r\n", + " File \"/actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/RWKV-v5/src/model.py\", line 650, in __init__\r\n", + " self.load_state_dict(model_weights)\r\n", + " File \"/usr/local/lib/python3.11/dist-packages/torch/nn/modules/module.py\", line 2041, in load_state_dict\r\n", + " raise RuntimeError('Error(s) in loading state_dict for {}:\\n\\t{}'.format(\r\n", + "RuntimeError: Error(s) in loading state_dict for RWKV:\r\n", + "\tsize mismatch for blocks.0.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.0.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.1.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.1.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.2.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.2.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.3.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.3.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.4.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.4.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.5.att.time_decay: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n", + "\tsize mismatch for blocks.5.att.time_faaaa: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([64]).\r\n" + ] + } + ], + "source": [ + "# Lets do a quick memory test\n", + "!export RWKV_WAVENET_LAYERS=\"{RWKV_WAVENET_LAYERS}\" && \\\n", + " python3 ../memory_script/eval_v5_memory_guided.py \"{PROJECT_DIR}/model/{FILENAME_PREFIX}-enwiki-instruct.pth\"" + ] + } + ], + "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": "3.11.4" + }, + "papermill": { + "default_parameters": {}, + "duration": 28184.999752, + "end_time": "2023-08-29T02:35:09.072726", + "environment_variables": {}, + "exception": null, + "input_path": "/actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/notebook/experiment/rwkv-x-exp/v5-headsize2x/v5-L6-D4096-E1e-1-ctx4k-part1.ipynb", + "output_path": "/actions-runner/_work/RWKV-infctx-trainer/RWKV-infctx-trainer/output/experiment/rwkv-x-exp/v5-headsize2x/v5-L6-D4096-E1e-1-ctx4k-part1.ipynb", + "parameters": {}, + "start_time": "2023-08-28T18:45:24.072974", + "version": "2.4.0" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +} \ No newline at end of file