Upload model

.gitignore

@@ -0,0 +1 @@
+*

arguments.json

@@ -0,0 +1,53 @@
+{
+    "model_name_or_path": "models/alpaca-7b-reproduced",
+    "max_length": 1024,
+    "trust_remote_code": true,
+    "train_datasets": [
+        [
+            "alpaca",
+            {
+                "proportion": 1.0
+            }
+        ]
+    ],
+    "eval_datasets": null,
+    "epochs": 3,
+    "per_device_train_batch_size": 8,
+    "per_device_eval_batch_size": 8,
+    "gradient_accumulation_steps": 8,
+    "gradient_checkpointing": true,
+    "lr": 2e-05,
+    "lr_scheduler_type": "cosine",
+    "lr_warmup_ratio": 0.03,
+    "weight_decay": 0.0,
+    "seed": 42,
+    "fp16": false,
+    "bf16": true,
+    "tf32": true,
+    "eval_strategy": "epoch",
+    "eval_interval": 1000000,
+    "need_eval": false,
+    "eval_split_ratio": null,
+    "output_dir": "/home/juntao/Projects/roo/models/alpaca-7b-sft",
+    "log_type": "wandb",
+    "log_dir": "/home/juntao/Projects/roo/models/alpaca-7b-sft",
+    "log_project": "SFT-alpaca",
+    "log_group": null,
+    "log_run_name": "sft-2024-09-04-12-52-34",
+    "save_16bit": false,
+    "save_interval": 1000000,
+    "local_rank": 0,
+    "zero_stage": 3,
+    "offload": "none",
+    "deepspeed": false,
+    "deepspeed_config": null,
+    "deepscale": false,
+    "deepscale_config": null,
+    "global_rank": 0,
+    "device": {
+        "type": "torch.device",
+        "repr": "device(type='cuda', index=0)"
+    },
+    "num_update_steps_per_epoch": 102,
+    "total_training_steps": 306
+}

arguments.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:028bdcd52461fcb6f24d65acdb4ba31a8f67b61a7328cfd75cefb41a300b1203
+size 1030

config.json

@@ -0,0 +1,30 @@
+{
+  "_name_or_path": "models/alpaca-7b-reproduced",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 2048,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "pad_token_id": 32000,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.41.2",
+  "use_cache": true,
+  "vocab_size": 32001
+}

environ.txt

@@ -0,0 +1,119 @@
+BAT_THEME=Monokai Extended
+BROWSER=/home/juntao/.vscode-server/cli/servers/Stable-fee1edb8d6d72a0ddff41e5f71a671c23ed924b9/server/bin/helpers/browser.sh
+CC=/usr/bin/gcc
+COLORTERM=truecolor
+CONDA_DEFAULT_ENV=roo
+CONDA_EXE=/home/juntao/Miniconda3/bin/conda
+CONDA_PREFIX=/home/juntao/Miniconda3/envs/roo
+CONDA_PROMPT_MODIFIER=(roo) 
+CONDA_PYTHONBREAKPOINT=
+CONDA_PYTHON_EXE=/home/juntao/Miniconda3/bin/python
+CONDA_ROOT=/home/juntao/Miniconda3
+CONDA_SHLVL=1
+CPLUS_INCLUDE_PATH=/usr/local/cuda/include:/usr/local/cuda/extras/CUPTI/include
+CROSS_RANK=0
+CROSS_SIZE=1
+CUDA_HOME=/usr/local/cuda
+CUDA_MODULE_LOADING=LAZY
+CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
+CXX=/usr/bin/g++
+C_INCLUDE_PATH=/usr/local/cuda/include:/usr/local/cuda/extras/CUPTI/include
+DBUS_SESSION_BUS_ADDRESS=unix:path=/run/user/1001/bus
+FC=/usr/bin/gfortran
+FPATH=/home/juntao/.oh-my-zsh/plugins/vscode:/home/juntao/.oh-my-zsh/plugins/brew:/home/juntao/.oh-my-zsh/plugins/tmux:/home/juntao/.oh-my-zsh/plugins/docker:/home/juntao/.oh-my-zsh/plugins/pylint:/home/juntao/.oh-my-zsh/plugins/pip:/home/juntao/.oh-my-zsh/plugins/python:/home/juntao/.oh-my-zsh/plugins/git-auto-fetch:/home/juntao/.oh-my-zsh/plugins/git:/home/juntao/.oh-my-zsh/plugins/alias-finder:/home/juntao/.oh-my-zsh/plugins/rsync:/home/juntao/.oh-my-zsh/plugins/cp:/home/juntao/.oh-my-zsh/plugins/copypath:/home/juntao/.oh-my-zsh/plugins/copyfile:/home/juntao/.oh-my-zsh/plugins/fzf:/home/juntao/.oh-my-zsh/plugins/colored-man-pages:/home/juntao/.oh-my-zsh/plugins/colorize:/home/juntao/.oh-my-zsh/custom/plugins/conda-zsh-completion:/home/juntao/.oh-my-zsh/custom/plugins/zsh-completions:/home/juntao/.oh-my-zsh/custom/plugins/zsh-autosuggestions:/home/juntao/.oh-my-zsh/custom/plugins/zsh-syntax-highlighting:/home/juntao/.oh-my-zsh/plugins/ubuntu:/home/juntao/.oh-my-zsh/functions:/home/juntao/.oh-my-zsh/completions:/home/juntao/.oh-my-zsh/custom/functions:/home/juntao/.oh-my-zsh/custom/completions:/home/linuxbrew/.linuxbrew/share/zsh/site-functions:/home/juntao/.oh-my-zsh/plugins/vscode:/home/juntao/.oh-my-zsh/plugins/brew:/home/juntao/.oh-my-zsh/plugins/tmux:/home/juntao/.oh-my-zsh/plugins/docker:/home/juntao/.oh-my-zsh/plugins/pylint:/home/juntao/.oh-my-zsh/plugins/pip:/home/juntao/.oh-my-zsh/plugins/python:/home/juntao/.oh-my-zsh/plugins/git-auto-fetch:/home/juntao/.oh-my-zsh/plugins/git:/home/juntao/.oh-my-zsh/plugins/alias-finder:/home/juntao/.oh-my-zsh/plugins/rsync:/home/juntao/.oh-my-zsh/plugins/cp:/home/juntao/.oh-my-zsh/plugins/copypath:/home/juntao/.oh-my-zsh/plugins/copyfile:/home/juntao/.oh-my-zsh/plugins/fzf:/home/juntao/.oh-my-zsh/plugins/colored-man-pages:/home/juntao/.oh-my-zsh/plugins/colorize:/home/juntao/.oh-my-zsh/custom/plugins/conda-zsh-completion:/home/juntao/.oh-my-zsh/custom/plugins/zsh-completions:/home/juntao/.oh-my-zsh/custom/plugins/zsh-autosuggestions:/home/juntao/.oh-my-zsh/custom/plugins/zsh-syntax-highlighting:/home/juntao/.oh-my-zsh/plugins/ubuntu:/home/juntao/.oh-my-zsh/functions:/home/juntao/.oh-my-zsh/completions:/home/juntao/.oh-my-zsh/custom/functions:/home/juntao/.oh-my-zsh/custom/completions:/home/juntao/.oh-my-zsh/cache/completions:/usr/local/share/zsh/site-functions:/usr/share/zsh/vendor-functions:/usr/share/zsh/vendor-completions:/usr/share/zsh/functions/Calendar:/usr/share/zsh/functions/Chpwd:/usr/share/zsh/functions/Completion:/usr/share/zsh/functions/Completion/AIX:/usr/share/zsh/functions/Completion/BSD:/usr/share/zsh/functions/Completion/Base:/usr/share/zsh/functions/Completion/Cygwin:/usr/share/zsh/functions/Completion/Darwin:/usr/share/zsh/functions/Completion/Debian:/usr/share/zsh/functions/Completion/Linux:/usr/share/zsh/functions/Completion/Mandriva:/usr/share/zsh/functions/Completion/Redhat:/usr/share/zsh/functions/Completion/Solaris:/usr/share/zsh/functions/Completion/Unix:/usr/share/zsh/functions/Completion/X:/usr/share/zsh/functions/Completion/Zsh:/usr/share/zsh/functions/Completion/openSUSE:/usr/share/zsh/functions/Exceptions:/usr/share/zsh/functions/MIME:/usr/share/zsh/functions/Math:/usr/share/zsh/functions/Misc:/usr/share/zsh/functions/Newuser:/usr/share/zsh/functions/Prompts:/usr/share/zsh/functions/TCP:/usr/share/zsh/functions/VCS_Info:/usr/share/zsh/functions/VCS_Info/Backends:/usr/share/zsh/functions/Zftp:/usr/share/zsh/functions/Zle:/home/linuxbrew/.linuxbrew/share/zsh/functions:/home/juntao/.oh-my-zsh/custom/plugins/zsh-completions/src:/home/juntao/.oh-my-zsh/custom/plugins/zsh-completions/src:/home/juntao/.oh-my-zsh/custom/plugins/conda-zsh-completion:/home/linuxbrew/.linuxbrew/share/zsh/site-functions
+FZF_CTRL_T_COMMAND=fdfind --type file --follow --hidden --no-ignore-vcs --exclude '.git' --exclude '[Mm]iniconda3' --exclude '[Aa]naconda3' --color=always
+FZF_DEFAULT_COMMAND=fdfind --type file --follow --hidden --no-ignore-vcs --exclude '.git' --exclude '[Mm]iniconda3' --exclude '[Aa]naconda3' --color=always
+FZF_DEFAULT_OPTS=--height=40% --layout=reverse --ansi --preview='(batcat --color=always {} || highlight -O ansi {} || cat {}) 2>/dev/null | head -100'
+GIT_ASKPASS=/home/juntao/.vscode-server/cli/servers/Stable-fee1edb8d6d72a0ddff41e5f71a671c23ed924b9/server/extensions/git/dist/askpass.sh
+GSETTINGS_SCHEMA_DIR=/home/juntao/Miniconda3/envs/roo/share/glib-2.0/schemas
+GSETTINGS_SCHEMA_DIR_CONDA_BACKUP=
+HOME=/home/juntao
+HOMEBREW_API_DOMAIN=https://mirrors.tuna.tsinghua.edu.cn/homebrew-bottles/api
+HOMEBREW_BAT=true
+HOMEBREW_BOTTLE_DOMAIN=https://mirrors.tuna.tsinghua.edu.cn/homebrew-bottles
+HOMEBREW_BREW_GIT_REMOTE=https://mirrors.tuna.tsinghua.edu.cn/git/homebrew/brew.git
+HOMEBREW_CELLAR=/home/linuxbrew/.linuxbrew/Cellar
+HOMEBREW_CORE_GIT_REMOTE=https://mirrors.tuna.tsinghua.edu.cn/git/homebrew/homebrew-core.git
+HOMEBREW_EDITOR=vim
+HOMEBREW_PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple
+HOMEBREW_PREFIX=/home/linuxbrew/.linuxbrew
+HOMEBREW_REPOSITORY=/home/linuxbrew/.linuxbrew/Homebrew
+INFOPATH=/home/linuxbrew/.linuxbrew/share/info:/home/linuxbrew/.linuxbrew/share/info:/home/linuxbrew/.linuxbrew/share/info:
+LANG=en_US.UTF-8
+LC_ALL=en_US.utf8
+LD_LIBRARY_PATH=/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64
+LESS=-R -M -i -j5
+LIBRARY_PATH=/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64
+LOCAL_RANK=0
+LOCAL_SIZE=8
+LOGLEVEL=WARNING
+LOGNAME=juntao
+LSCOLORS=Gxfxcxdxbxegedabagacad
+LS_COLORS=rs=0:di=01;34:ln=01;36:mh=00:pi=40;33:so=01;35:do=01;35:bd=40;33;01:cd=40;33;01:or=40;31;01:mi=00:su=37;41:sg=30;43:ca=30;41:tw=30;42:ow=34;42:st=37;44:ex=01;32:*.tar=01;31:*.tgz=01;31:*.arc=01;31:*.arj=01;31:*.taz=01;31:*.lha=01;31:*.lz4=01;31:*.lzh=01;31:*.lzma=01;31:*.tlz=01;31:*.txz=01;31:*.tzo=01;31:*.t7z=01;31:*.zip=01;31:*.z=01;31:*.dz=01;31:*.gz=01;31:*.lrz=01;31:*.lz=01;31:*.lzo=01;31:*.xz=01;31:*.zst=01;31:*.tzst=01;31:*.bz2=01;31:*.bz=01;31:*.tbz=01;31:*.tbz2=01;31:*.tz=01;31:*.deb=01;31:*.rpm=01;31:*.jar=01;31:*.war=01;31:*.ear=01;31:*.sar=01;31:*.rar=01;31:*.alz=01;31:*.ace=01;31:*.zoo=01;31:*.cpio=01;31:*.7z=01;31:*.rz=01;31:*.cab=01;31:*.wim=01;31:*.swm=01;31:*.dwm=01;31:*.esd=01;31:*.jpg=01;35:*.jpeg=01;35:*.mjpg=01;35:*.mjpeg=01;35:*.gif=01;35:*.bmp=01;35:*.pbm=01;35:*.pgm=01;35:*.ppm=01;35:*.tga=01;35:*.xbm=01;35:*.xpm=01;35:*.tif=01;35:*.tiff=01;35:*.png=01;35:*.svg=01;35:*.svgz=01;35:*.mng=01;35:*.pcx=01;35:*.mov=01;35:*.mpg=01;35:*.mpeg=01;35:*.m2v=01;35:*.mkv=01;35:*.webm=01;35:*.ogm=01;35:*.mp4=01;35:*.m4v=01;35:*.mp4v=01;35:*.vob=01;35:*.qt=01;35:*.nuv=01;35:*.wmv=01;35:*.asf=01;35:*.rm=01;35:*.rmvb=01;35:*.flc=01;35:*.avi=01;35:*.fli=01;35:*.flv=01;35:*.gl=01;35:*.dl=01;35:*.xcf=01;35:*.xwd=01;35:*.yuv=01;35:*.cgm=01;35:*.emf=01;35:*.ogv=01;35:*.ogx=01;35:*.aac=00;36:*.au=00;36:*.flac=00;36:*.m4a=00;36:*.mid=00;36:*.midi=00;36:*.mka=00;36:*.mp3=00;36:*.mpc=00;36:*.ogg=00;36:*.ra=00;36:*.wav=00;36:*.oga=00;36:*.opus=00;36:*.spx=00;36:*.xspf=00;36:
+MASTER_ADDR=127.0.0.1
+MASTER_PORT=15892
+MOTD_SHOWN=pam
+MPICH_CC=/usr/bin/gcc
+MPICH_CXX=/usr/bin/g++
+MPICH_FC=/usr/bin/gfortran
+OLDPWD=/home/juntao/Projects/roo
+OMPI_CC=/usr/bin/gcc
+OMPI_CXX=/usr/bin/g++
+OMPI_FC=/usr/bin/gfortran
+P10K_LEAN_STYLE=true
+P9K_SSH=1
+P9K_TTY=old
+PAGER=less
+PATH=/home/juntao/Miniconda3/envs/roo/bin:/home/juntao/.gem/ruby/2.7.0/bin:/var/lib/gems/2.7.0/bin:/usr/local/cuda/bin:/home/linuxbrew/.linuxbrew/bin:/home/linuxbrew/.linuxbrew/sbin:/home/juntao/.vscode-server/cli/servers/Stable-fee1edb8d6d72a0ddff41e5f71a671c23ed924b9/server/bin/remote-cli:/home/juntao/.perl/bin:/home/juntao/Miniconda3/condabin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin:/home/juntao/.fzf/bin
+PERL5LIB=/home/juntao/.perl/lib/perl5
+PERL_LOCAL_LIB_ROOT=/home/juntao/.perl
+PERL_MB_OPT=--install_base "/home/juntao/.perl"
+PERL_MM_OPT=INSTALL_BASE=/home/juntao/.perl
+PWD=/home/juntao/Projects/roo
+PYTHONBREAKPOINT=ipdb.set_trace
+PYTHONHASHSEED=42
+PYTHONPATH=/home/juntao/Projects/roo
+RANK=0
+RUBYOPT=-W0
+SHELL=/usr/bin/zsh
+SHLVL=3
+SSH_CLIENT=218.68.227.248 2030 24100
+SSH_CONNECTION=218.68.227.248 2030 10.10.40.131 24100
+SSL_CERT_DIR=/usr/lib/ssl/certs
+SSL_CERT_FILE=/usr/lib/ssl/certs/ca-certificates.crt
+TERM=xterm-256color
+TERM_PROGRAM=vscode
+TERM_PROGRAM_VERSION=1.92.2
+TMUX=/tmp/tmux-1001/default,35347,0
+TMUX_CONF=/home/juntao/.tmux.conf
+TMUX_CONF_LOCAL=/home/juntao/.tmux.conf.local
+TMUX_PANE=%0
+TMUX_PROGRAM=/usr/bin/tmux
+TMUX_SOCKET=/tmp/tmux-1001/default
+USER=juntao
+USER_ZDOTDIR=/home/juntao
+VSCODE_GIT_ASKPASS_EXTRA_ARGS=
+VSCODE_GIT_ASKPASS_MAIN=/home/juntao/.vscode-server/cli/servers/Stable-fee1edb8d6d72a0ddff41e5f71a671c23ed924b9/server/extensions/git/dist/askpass-main.js
+VSCODE_GIT_ASKPASS_NODE=/home/juntao/.vscode-server/cli/servers/Stable-fee1edb8d6d72a0ddff41e5f71a671c23ed924b9/server/node
+VSCODE_GIT_IPC_HANDLE=/run/user/1001/vscode-git-39160f57ba.sock
+VSCODE_INJECTION=1
+VSCODE_IPC_HOOK_CLI=/run/user/1001/vscode-ipc-95a68742-c3c1-4fc3-926f-daa969efe302.sock
+VSCODE_NONCE=b605a6a9-7032-4c22-bc1d-5f29901de2f4
+WANDB_API_KEY=84ca867e702f8a930beae406c06c21ec7f8acfe3
+WANDB_SERVICE=2-2131706-tcp-localhost-43909
+WORLD_SIZE=8
+XDG_DATA_DIRS=/usr/local/share:/usr/share:/var/lib/snapd/desktop
+XDG_RUNTIME_DIR=/run/user/1001
+XDG_SESSION_CLASS=user
+XDG_SESSION_ID=3
+XDG_SESSION_TYPE=tty
+ZDOTDIR=/home/juntao
+ZSH=/home/juntao/.oh-my-zsh
+ZSH_TMUX_CONFIG=/home/juntao/.tmux.conf
+ZSH_TMUX_TERM=screen-256color
+_=/home/juntao/Miniconda3/envs/roo/bin/deepspeed
+_CE_CONDA=
+_CE_M=
+_P9K_SSH_TTY=/dev/pts/9
+_P9K_TTY=/dev/pts/9
+_ZSH_TMUX_FIXED_CONFIG=/home/juntao/.oh-my-zsh/plugins/tmux/tmux.extra.conf

latest

@@ -0,0 +1 @@
+global_step304

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a4f065aa42dce062ab13ef1e6ebfcdcca533fd8dbe9c14ce30759c4a76c8b2b
+size 26953792682

script.sh

@@ -0,0 +1,132 @@
+#!/usr/bin/env bash
+#
+# Copyright 2023-2024 PKU-Alignment Team. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+if [ -z "${BASH_VERSION}" ]; then
+	echo "Please use bash to run this script." >&2
+	exit 1
+fi
+
+set -x
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" &>/dev/null && pwd)"
+ROOT_DIR="$(dirname "${SCRIPT_DIR}")"
+export PYTHONPATH="${ROOT_DIR}${PYTHONPATH:+:${PYTHONPATH}}"
+export LOGLEVEL="${LOGLEVEL:-WARNING}"
+
+MODEL_NAME_OR_PATH="models/alpaca-7b-reproduced"
+OUTPUT_DIR="models/alpaca-7b-sft"
+unset HOSTFILE
+ZERO_STAGE=3
+OFFLOAD="none"
+while [[ "$#" -gt 0 ]]; do
+	arg="$1"
+	shift
+	case "${arg}" in
+		--model_name_or_path)
+			MODEL_NAME_OR_PATH="$1"
+			shift
+			;;
+		--model_name_or_path=*)
+			MODEL_NAME_OR_PATH="${arg#*=}"
+			;;
+		--output_dir)
+			OUTPUT_DIR="$1"
+			shift
+			;;
+		--output_dir=*)
+			OUTPUT_DIR="${arg#*=}"
+			;;
+		--hostfile)
+			HOSTFILE="$1"
+			shift
+			;;
+		--hostfile=*)
+			HOSTFILE="${arg#*=}"
+			;;
+		--zero_stage)
+			ZERO_STAGE="$1"
+			shift
+			;;
+		--zero_stage=*)
+			ZERO_STAGE="${arg#*=}"
+			;;
+		--offload)
+			OFFLOAD="$1"
+			shift
+			;;
+		--offload=*)
+			OFFLOAD="${arg#*=}"
+			;;
+		*)
+			echo "Unknown parameter passed: '${arg}'" >&2
+			exit 1
+			;;
+	esac
+done
+
+mkdir -p "${OUTPUT_DIR}"
+OUTPUT_DIR="$(cd "${OUTPUT_DIR}" &>/dev/null && pwd)"
+if [[ ! -f "${OUTPUT_DIR}/.gitignore" ]]; then
+	echo '*' >"${OUTPUT_DIR}/.gitignore"
+fi
+
+cp -f "$0" "${OUTPUT_DIR}/script.sh"
+
+if [[ -z "${WANDB_API_KEY}" ]]; then
+	export WANDB_MODE="offline"
+fi
+
+MASTER_PORT_START=10000
+MASTER_PORT_END=65535
+MASTER_PORT="$(
+	comm -23 \
+		<(seq "${MASTER_PORT_START}" "${MASTER_PORT_END}" | sort) \
+		<(ss -Htan | awk '{ print $4 }' | awk -F ':' '{ print $NF }' | sort -u) |
+		shuf | head -n 1
+)"
+
+DEEPSPEED_ARGS=()
+if [[ -n "${HOSTFILE+x}" ]]; then
+	DEEPSPEED_ARGS+=("--hostfile" "${HOSTFILE}")
+fi
+DEEPSPEED_ARGS+=("--master_port" "${MASTER_PORT}")
+
+exec 1> >(tee "${OUTPUT_DIR}/stdout.log" >&1) 2> >(tee "${OUTPUT_DIR}/stderr.log" >&2)
+
+deepspeed "${DEEPSPEED_ARGS[@]}" \
+	--module safe_rlhf.finetune \
+	--train_datasets alpaca \
+	--model_name_or_path "${MODEL_NAME_OR_PATH}" \
+	--max_length 1024 \
+	--trust_remote_code True \
+	--epochs 3 \
+	--per_device_train_batch_size 8 \
+	--per_device_eval_batch_size 8 \
+	--gradient_accumulation_steps 8 \
+	--gradient_checkpointing \
+	--learning_rate 2e-5 \
+	--lr_scheduler_type cosine \
+	--lr_warmup_ratio 0.03 \
+	--weight_decay 0.0 \
+	--seed 42 \
+	--output_dir "${OUTPUT_DIR}" \
+	--log_type wandb \
+	--log_project SFT-alpaca \
+	--zero_stage "${ZERO_STAGE}" \
+	--offload "${OFFLOAD}" \
+	--bf16 True \
+	--tf32 True

special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<pad>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

stdout.log

@@ -0,0 +1,64 @@
+[2024-09-04 12:51:56,498] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
+[2024-09-04 12:51:58,109] [WARNING] [runner.py:212:fetch_hostfile] Unable to find hostfile, will proceed with training with local resources only.
+[2024-09-04 12:51:58,109] [INFO] [runner.py:585:main] cmd = /home/juntao/Miniconda3/envs/roo/bin/python -u -m deepspeed.launcher.launch --world_info=eyJsb2NhbGhvc3QiOiBbMCwgMSwgMiwgMywgNCwgNSwgNiwgN119 --master_addr=127.0.0.1 --master_port=15892 --module --enable_each_rank_log=None safe_rlhf.finetune --train_datasets alpaca --model_name_or_path models/alpaca-7b-reproduced --max_length 1024 --trust_remote_code True --epochs 3 --per_device_train_batch_size 8 --per_device_eval_batch_size 8 --gradient_accumulation_steps 8 --gradient_checkpointing --learning_rate 2e-5 --lr_scheduler_type cosine --lr_warmup_ratio 0.03 --weight_decay 0.0 --seed 42 --output_dir /home/juntao/Projects/roo/models/alpaca-7b-sft --log_type wandb --log_project SFT-alpaca --zero_stage 3 --offload none --bf16 True --tf32 True
+[2024-09-04 12:51:59,771] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
+[2024-09-04 12:52:01,856] [INFO] [launch.py:146:main] WORLD INFO DICT: {'localhost': [0, 1, 2, 3, 4, 5, 6, 7]}
+[2024-09-04 12:52:01,856] [INFO] [launch.py:152:main] nnodes=1, num_local_procs=8, node_rank=0
+[2024-09-04 12:52:01,856] [INFO] [launch.py:163:main] global_rank_mapping=defaultdict(<class 'list'>, {'localhost': [0, 1, 2, 3, 4, 5, 6, 7]})
+[2024-09-04 12:52:01,856] [INFO] [launch.py:164:main] dist_world_size=8
+[2024-09-04 12:52:01,856] [INFO] [launch.py:168:main] Setting CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
+[2024-09-04 12:52:01,858] [INFO] [launch.py:256:main] process 2131706 spawned with command: ['/home/juntao/Miniconda3/envs/roo/bin/python', '-u', '-m', 'safe_rlhf.finetune', '--local_rank=0', '--train_datasets', 'alpaca', '--model_name_or_path', 'models/alpaca-7b-reproduced', '--max_length', '1024', '--trust_remote_code', 'True', '--epochs', '3', '--per_device_train_batch_size', '8', '--per_device_eval_batch_size', '8', '--gradient_accumulation_steps', '8', '--gradient_checkpointing', '--learning_rate', '2e-5', '--lr_scheduler_type', 'cosine', '--lr_warmup_ratio', '0.03', '--weight_decay', '0.0', '--seed', '42', '--output_dir', '/home/juntao/Projects/roo/models/alpaca-7b-sft', '--log_type', 'wandb', '--log_project', 'SFT-alpaca', '--zero_stage', '3', '--offload', 'none', '--bf16', 'True', '--tf32', 'True']
+[2024-09-04 12:52:01,859] [INFO] [launch.py:256:main] process 2131707 spawned with command: ['/home/juntao/Miniconda3/envs/roo/bin/python', '-u', '-m', 'safe_rlhf.finetune', '--local_rank=1', '--train_datasets', 'alpaca', '--model_name_or_path', 'models/alpaca-7b-reproduced', '--max_length', '1024', '--trust_remote_code', 'True', '--epochs', '3', '--per_device_train_batch_size', '8', '--per_device_eval_batch_size', '8', '--gradient_accumulation_steps', '8', '--gradient_checkpointing', '--learning_rate', '2e-5', '--lr_scheduler_type', 'cosine', '--lr_warmup_ratio', '0.03', '--weight_decay', '0.0', '--seed', '42', '--output_dir', '/home/juntao/Projects/roo/models/alpaca-7b-sft', '--log_type', 'wandb', '--log_project', 'SFT-alpaca', '--zero_stage', '3', '--offload', 'none', '--bf16', 'True', '--tf32', 'True']
+[2024-09-04 12:52:01,860] [INFO] [launch.py:256:main] process 2131708 spawned with command: ['/home/juntao/Miniconda3/envs/roo/bin/python', '-u', '-m', 'safe_rlhf.finetune', '--local_rank=2', '--train_datasets', 'alpaca', '--model_name_or_path', 'models/alpaca-7b-reproduced', '--max_length', '1024', '--trust_remote_code', 'True', '--epochs', '3', '--per_device_train_batch_size', '8', '--per_device_eval_batch_size', '8', '--gradient_accumulation_steps', '8', '--gradient_checkpointing', '--learning_rate', '2e-5', '--lr_scheduler_type', 'cosine', '--lr_warmup_ratio', '0.03', '--weight_decay', '0.0', '--seed', '42', '--output_dir', '/home/juntao/Projects/roo/models/alpaca-7b-sft', '--log_type', 'wandb', '--log_project', 'SFT-alpaca', '--zero_stage', '3', '--offload', 'none', '--bf16', 'True', '--tf32', 'True']
+[2024-09-04 12:52:01,861] [INFO] [launch.py:256:main] process 2131709 spawned with command: ['/home/juntao/Miniconda3/envs/roo/bin/python', '-u', '-m', 'safe_rlhf.finetune', '--local_rank=3', '--train_datasets', 'alpaca', '--model_name_or_path', 'models/alpaca-7b-reproduced', '--max_length', '1024', '--trust_remote_code', 'True', '--epochs', '3', '--per_device_train_batch_size', '8', '--per_device_eval_batch_size', '8', '--gradient_accumulation_steps', '8', '--gradient_checkpointing', '--learning_rate', '2e-5', '--lr_scheduler_type', 'cosine', '--lr_warmup_ratio', '0.03', '--weight_decay', '0.0', '--seed', '42', '--output_dir', '/home/juntao/Projects/roo/models/alpaca-7b-sft', '--log_type', 'wandb', '--log_project', 'SFT-alpaca', '--zero_stage', '3', '--offload', 'none', '--bf16', 'True', '--tf32', 'True']
+[2024-09-04 12:52:01,862] [INFO] [launch.py:256:main] process 2131710 spawned with command: ['/home/juntao/Miniconda3/envs/roo/bin/python', '-u', '-m', 'safe_rlhf.finetune', '--local_rank=4', '--train_datasets', 'alpaca', '--model_name_or_path', 'models/alpaca-7b-reproduced', '--max_length', '1024', '--trust_remote_code', 'True', '--epochs', '3', '--per_device_train_batch_size', '8', '--per_device_eval_batch_size', '8', '--gradient_accumulation_steps', '8', '--gradient_checkpointing', '--learning_rate', '2e-5', '--lr_scheduler_type', 'cosine', '--lr_warmup_ratio', '0.03', '--weight_decay', '0.0', '--seed', '42', '--output_dir', '/home/juntao/Projects/roo/models/alpaca-7b-sft', '--log_type', 'wandb', '--log_project', 'SFT-alpaca', '--zero_stage', '3', '--offload', 'none', '--bf16', 'True', '--tf32', 'True']
+[2024-09-04 12:52:01,863] [INFO] [launch.py:256:main] process 2131711 spawned with command: ['/home/juntao/Miniconda3/envs/roo/bin/python', '-u', '-m', 'safe_rlhf.finetune', '--local_rank=5', '--train_datasets', 'alpaca', '--model_name_or_path', 'models/alpaca-7b-reproduced', '--max_length', '1024', '--trust_remote_code', 'True', '--epochs', '3', '--per_device_train_batch_size', '8', '--per_device_eval_batch_size', '8', '--gradient_accumulation_steps', '8', '--gradient_checkpointing', '--learning_rate', '2e-5', '--lr_scheduler_type', 'cosine', '--lr_warmup_ratio', '0.03', '--weight_decay', '0.0', '--seed', '42', '--output_dir', '/home/juntao/Projects/roo/models/alpaca-7b-sft', '--log_type', 'wandb', '--log_project', 'SFT-alpaca', '--zero_stage', '3', '--offload', 'none', '--bf16', 'True', '--tf32', 'True']
+[2024-09-04 12:52:01,864] [INFO] [launch.py:256:main] process 2131712 spawned with command: ['/home/juntao/Miniconda3/envs/roo/bin/python', '-u', '-m', 'safe_rlhf.finetune', '--local_rank=6', '--train_datasets', 'alpaca', '--model_name_or_path', 'models/alpaca-7b-reproduced', '--max_length', '1024', '--trust_remote_code', 'True', '--epochs', '3', '--per_device_train_batch_size', '8', '--per_device_eval_batch_size', '8', '--gradient_accumulation_steps', '8', '--gradient_checkpointing', '--learning_rate', '2e-5', '--lr_scheduler_type', 'cosine', '--lr_warmup_ratio', '0.03', '--weight_decay', '0.0', '--seed', '42', '--output_dir', '/home/juntao/Projects/roo/models/alpaca-7b-sft', '--log_type', 'wandb', '--log_project', 'SFT-alpaca', '--zero_stage', '3', '--offload', 'none', '--bf16', 'True', '--tf32', 'True']
+[2024-09-04 12:52:01,865] [INFO] [launch.py:256:main] process 2131713 spawned with command: ['/home/juntao/Miniconda3/envs/roo/bin/python', '-u', '-m', 'safe_rlhf.finetune', '--local_rank=7', '--train_datasets', 'alpaca', '--model_name_or_path', 'models/alpaca-7b-reproduced', '--max_length', '1024', '--trust_remote_code', 'True', '--epochs', '3', '--per_device_train_batch_size', '8', '--per_device_eval_batch_size', '8', '--gradient_accumulation_steps', '8', '--gradient_checkpointing', '--learning_rate', '2e-5', '--lr_scheduler_type', 'cosine', '--lr_warmup_ratio', '0.03', '--weight_decay', '0.0', '--seed', '42', '--output_dir', '/home/juntao/Projects/roo/models/alpaca-7b-sft', '--log_type', 'wandb', '--log_project', 'SFT-alpaca', '--zero_stage', '3', '--offload', 'none', '--bf16', 'True', '--tf32', 'True']
+[2024-09-04 12:52:04,266] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
+[2024-09-04 12:52:05,590] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
+[2024-09-04 12:52:05,973] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
+[2024-09-04 12:52:06,006] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
+[2024-09-04 12:52:06,071] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
+[2024-09-04 12:52:06,085] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
+[2024-09-04 12:52:06,108] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
+[2024-09-04 12:52:06,170] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
+[2024-09-04 12:52:07,786] [INFO] [comm.py:652:init_distributed] cdb=None
+[2024-09-04 12:52:08,898] [INFO] [comm.py:652:init_distributed] cdb=None
+[2024-09-04 12:52:09,351] [INFO] [comm.py:652:init_distributed] cdb=None
+[2024-09-04 12:52:09,394] [INFO] [comm.py:652:init_distributed] cdb=None
+[2024-09-04 12:52:09,425] [INFO] [comm.py:652:init_distributed] cdb=None
+[2024-09-04 12:52:09,488] [INFO] [comm.py:652:init_distributed] cdb=None
+[2024-09-04 12:52:09,525] [INFO] [comm.py:652:init_distributed] cdb=None
+[2024-09-04 12:52:09,589] [INFO] [comm.py:652:init_distributed] cdb=None
+[2024-09-04 12:52:09,589] [INFO] [comm.py:683:init_distributed] Initializing TorchBackend in DeepSpeed with backend nccl
+Set logger level to WARNING.
+ninja: no work to do.
+Time to load fused_adam op: 0.13483238220214844 seconds
+Time to load fused_adam op: 0.20363140106201172 seconds
+Time to load fused_adam op: 0.2036271095275879 secondsTime to load fused_adam op: 0.20359563827514648 seconds
+
+Time to load fused_adam op: 0.20377779006958008 seconds
+Time to load fused_adam op: 0.2039332389831543 seconds
+Time to load fused_adam op: 0.20380067825317383 seconds
+Time to load fused_adam op: 0.20460724830627441 seconds
+Parameter Offload: Total persistent parameters: 266240 in 65 params
+***** Running training *****
+Saving model to "/home/juntao/Projects/roo/models/alpaca-7b-sft" ...
+Saving DeepSpeed Checkpoints...
+Converting DeepSpeed Checkpoints to Hugging Face format...
+[2024-09-04 13:41:40,053] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
+Processing zero checkpoint './global_step304'
+Detected checkpoint of type zero stage ZeroStageEnum.weights, world_size: 8
+Parsing checkpoint created by deepspeed==0.15.0
+Reconstructed Trainable fp32 state dict with 291 params 6738423808 elements
+Saving fp32 state dict to pytorch_model.bin
+Model saved!
+[2024-09-04 13:42:46,338] [INFO] [launch.py:351:main] Process 2131710 exits successfully.
+[2024-09-04 13:42:46,338] [INFO] [launch.py:351:main] Process 2131712 exits successfully.
+[2024-09-04 13:42:46,338] [INFO] [launch.py:351:main] Process 2131707 exits successfully.
+[2024-09-04 13:42:46,338] [INFO] [launch.py:351:main] Process 2131713 exits successfully.
+[2024-09-04 13:42:46,339] [INFO] [launch.py:351:main] Process 2131709 exits successfully.
+[2024-09-04 13:42:46,339] [INFO] [launch.py:351:main] Process 2131711 exits successfully.
+[2024-09-04 13:42:47,339] [INFO] [launch.py:351:main] Process 2131708 exits successfully.
+[2024-09-04 13:42:56,341] [INFO] [launch.py:351:main] Process 2131706 exits successfully.

tokenizer_config.json

@@ -0,0 +1,50 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32000": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 1024,
+  "pad_token": "<pad>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

zero_to_fp32.py

@@ -0,0 +1,604 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None, exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+                                               args.output_file,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)