Training LLMs at Scale

Sam Foreman

2024-08-09

@ ATPESC 2024

🔗 Links

• 🏡 samforeman.me:

  • 🦜 Talks:
    • Training LLMs at Scale [slides]
  • 📦 Repos:
    • 🍋 saforem2/ezpz
      Train your model across any number of arbitrary devices, ezpz.
    • 💬 saforem2/wordplay
      Playing with words.
    • 🏎️ argonne-lcf/Megatron-DeepSpeed
      For only the largest of large language models.

🧑🏻‍💻 About Me

• Computational Scientist at Argonne National Laboratory (ALCF)
• Interested in {AI, HPC} for science
  • working on scaling large (language, vision, multi-modal) models

As a member of the AI / ML Group at ALCF, I work on:

• 🤖 🧪 AI + Science
• 🎲 Building better sampling methods for Lattice QCD
• 🧬 Genome-Scale Language Models
  • GenSLM
  • 🥇 ACM Gordon Bell Special Prize

• 🌍 Foundation models for long term climate forecasting
• 🏃‍♂️ Scaling Large Language Models
• 🏎️ Distributed training across thousands of GPUs

🚀 Scaling: Overview

• ✅ Goal:
  • 📈 Maximize: Performance
  • 📉 Minimize: Cost¹
    • or, equivalently, 📈 maximize data throughput²

Note: See 🤗 Performance and Scalability: How To Fit a Bigger Model and Train It Faster for more details

1. Typically, the amount of time (💸) spent training

2. Typically want to utilize as much of GPU as possible

AI 🤝 Compute [Historical]

• First Era:
  • [1960 – 2012]
  • 2 year doubling (Moore’s law)
    • \simeq 7\times increase

 

• Modern Era:
  • [2012 – present]
  • 3.4 month doubling
    • \simeq \mathbf{300,000}\times increase

Source.

AI 🤝 Compute [Modern]

• First Era:
  • [1960 – 2012]
  • 2 year doubling (Moore’s law)
    • \simeq 7\times increase

 

• Modern Era:
  • [2012 – present]
  • 3.4 month doubling
    • \simeq \mathbf{300,000}\times increase

Source.

Parallelism Concepts

• Single GPU
• Data Parallel (DP)
• Tensor Parallel (TP)
• Pipeline Parallel (PP)
• ZeRO
• FSDP

Figure 1: SLOW !! model size limited by GPU memory

• The simplest and most common parallelism technique

• Workers maintain identical copies of the complete model and work on a subset of the data

  • Multiple copies of the same setup
    • each copy gets fed unique data
    • all copies compute gradients w.r.t local model
    • everyone syncs up before updating weights

• The processing is done in parallel and all setups are synchronized at the end of each training step.

Figure 2: Data Parallel Training

• Each tensor is split up into multiple chunks
• So, instead of having the whole tensor reside on a single GPU, each shard of the tensor resides on its designated GPU
  • During processing each shard gets processed separately and in parallel on different GPUs and the results are synced at the end of the step.
  • This is what one may call horizontal parallelism, as the splitting happens on horizontal level.

Figure 3: Tensor Parallel Training

• Model is split up vertically (layer-level) across multiple GPUs, so that only one or several layers of the model are places on a single GPU
  • Each GPU processes in parallel different stages of the pipeline and working on a small chunk of the batch.

Figure 4: Pipeline Parallelism

Figure 5: DeepSpeed + ZeRO

• Shards tensors (~ similar to TP), except:

  • whole tensor gets reconstructed as needed

  • Doesn’t require model modifications !!

• Depending on the ZeRO stage (1, 2, 3), we can offload:

  1. Stage 1: optimizer states

  2. Stage 2: gradients + opt. states

  3. Stage 3: model params + grads + opt. states

  with increasing ZeRO stage, we are able to free up increasing amounts of GPU memory

• ZeRO Data Parallel

  • ZeRO powered data parallelism is shown below

• It also supports various offloading techniques to compensate for limited GPU memory.

• 🔗 See also:

  • ZeRO — DeepSpeed 0.14.5 documentation
  • ZeRO-Offload
  • ZeRO & DeepSpeed: New system optimizations enable training models with over 100 billion parameters - Microsoft Research

• Instead of maintaining per-GPU copy of {params, grads, opt_states}, FSDP shards (distributes) these across data-parallel workers
  • can optionally offload the sharded model params to CPU

FSDP Workflow. Source

• 🔗 See also:
  • Introducing PyTorch Fully Sharded Data Parallel (FSDP) API | PyTorch

See: 🤗 Model Parallelism for additional details

Data Parallelism

Data Parallel Training

• Relatively simple to get up and running (minor modifications to code)
• saforem2/ezpz
• PyTorch – DDP
• DeepSpeed

Figure 6: Data Parallelism

Also see: 🎬 “Parallel Training Techniques”_

Deal with Data

• At each training step, we want to ensure that each worker receives unique data

• This can be done in one of two ways:

  1. Manually partition data (ahead of time) and assign different sections to different workers
     1. Each worker can only see their local portion of the data
  2. From each worker, randomly select a mini-batch
     1. Each worker can see the full dataset

  ⚠️ Warning

  Don’t forget your seed!

  When randomly selecting, it is important that each worker uses different seeds to ensure they receive unique data

Broadcast Initial State

• At the start of training (or when loading from a checkpoint), we want all of our workers to be initialized consistently

  • Broadcast the model and optimizer states from rank() == 0 worker

  flowchart TD
    0["GPU0"] --> 1["GPU 1"]
    0 --> 2["GPU 2"]
    0 -->|Model + Optimizer State| 3["GPU 3"]
    0 --> ...
    0 --> N["GPU N"]

Best Practices

🤝 Keeping things in Sync

Computation stalls during communication !!

Keeping the communication to computation ratio small is important for effective scaling.

• Use parallel IO whenever possible
  • Feed each rank from different files
  • Use MPI IO to have each rank read its own batch from a file
  • Use several ranks to read data, MPI to scatter to remaining ranks
    • Most practical in big at-scale training
• Take advantage of data storage
  • Use striping on lustre
  • Use the right optimizations for Aurora, Polaris, etc.
• Preload data when possible
  • Offloading to a GPU frees CPU cycles for loading the next batch of data
    • minimize IO latency this way

Why Distributed Training?

• Splitting data across workers \longrightarrow larger batch size¹
  • [micro_batch_size = 1] \times [N GPUs] \rightarrow [global_batch_size = N]
• Smooth loss landscape
• Improved gradient estimators
• Less iterations needed for same number of epochs
  • May need to train for more epochs if another change is not made
  • e.g. scaling learning rate
• See Large Batch Training of Convolutional Networks

1. micro_batch_size = batch_size per GPU

Recent Progress

Year	Author	Batch Size	GPU	# GPUs	TIME	ACC
2016	He	256	P100	8	29 Hour	75.30%
2019	Yamazaki	81,920	V100	2048	1.2 Min	75.08%

Deciding on a Parallelism Strategy

• Single GPU
• Single Node / Multi-GPU
• Multi-Node / Multi-GPU

• Model fits onto a single GPU:
  • Normal use
• Model DOES NOT fit on a single GPU:
  • ZeRO + Offload CPU (or, optionally, NVMe)
• Largest layer DOES NOT fit on a single GPU:
  • ZeRO + Enable Memory Centric Tiling (MCT)
    • MCT Allows running of arbitrarily large layers by automatically splitting them and executing them sequentially.

• Model fits onto a single GPU

  • DDP
  • ZeRO

• Model DOES NOT fit onto a single GPU[^connectivity]

  1. Pipeline Parallelism (PP)
  2. ZeRO
  3. Tensor Parallelism (TP)

• With sufficiently fast connectivity between nodes, these three strategies should be comparable.

  Otherwise, PP > ZeRO \simeq TP.

• When you have fast inter-node connectivity:

  • ZeRO (virtually NO modifications)
  • PP + ZeRO + TP + DP (less communication, at the cost of MAJOR modifications)

    • when you have slow inter-node connectivity and still low on GPU memory:

      DP + PP + TP + ZeRO-1

  • NOTE: TP is almost always used within a single node, e.g. TP <= GPUS_PER_NODE

Tensor (/ Model) Parallel Training: Example

\begin{align*} y &= \sum_{i} w_{i} * x_{i} \\ &= w_0 * x_0 + w_1 * x_1 + w_2 * x_2 \end{align*}

1. Compute y_{0} = w_{0} * x_{0} and send to \longrightarrow GPU1
2. Compute y_{1} = y_{0} + w_{1} * x_{1} and send to \longrightarrow GPU2
3. Compute y = y_{1} + w_{2} * x_{2} ✅

flowchart LR
  subgraph X0["GPU0"]
    direction LR
    a["w0"]
  end
  subgraph X1["GPU1"]
    direction LR
    b["w1"]
  end
  subgraph X2["GPU2"]
    direction LR
    c["w2"]
  end
  X1 & X0 <--> X2
  X0 <--> X1
  x["x0, x1, x2"] --> X0

Model Parallel Training

• Split up network over multiple workers
  • Each receives disjoint subset
  • All communication associated with subsets are distributed
• Communication whenever dataflow between two subsets
• Typically more complicated to implement than data parallel training
• Suitable when the model is too large to fit onto a single device (CPU / GPU)
• argonne-lcf/Megatron-DeepSpeed
• 🤗 huggingface/nanotron

Figure 7

Tensor (Model) Parallelism¹

• In Tensor Paralleism each GPU processes only a slice of a tensor and only aggregates the full tensor for operations that require the whole thing.

  • The main building block of any transformer is a fully connected nn.Linear followed by a nonlinear activation GeLU.

    • Y = GeLU(XA), where X and Y are the input and output vectors, and A is the weight matrix.

  • If we look at the computation in matrix form, it’s easy to see how the matrix multiplication can be split between multiple GPUs:

1. Efficient Large-Scale Language Model Training on GPU Clusters

Tensor Parallelism

Figure 8: Tensor Parallel GEMM. This information is based on (the much more in-depth) TP Overview by @anton-l

3D Parallelism

• DP + TP + PP (3D) Parallelism

Figure 9: Figure taken from 3D parallelism: Scaling to trillion-parameter models

Collective Communication

Typically, we assign 1 rank to each GPU (or accelerator), i.e. rank \in [0, 1, ..., WORLD_SIZE-1].

• AllReduce
• Reduce
• Broadcast
• AllGather
• Scatter

• Perform reductions on data (e.g. sum, min, max) across ranks, send result back to everyone

Figure 10: All-Reduce operation: each rank receives the reduction of input values across ranks.

• Perform a reduction on data across ranks, send to individual

Figure 11: Reduce operation: one rank receives the reduction of input values across ranks

• broadcast (send) a tensor x from one rank to all ranks

Figure 12

• Gathers tensors from the whole group in a list.

Figure 13

• Scatters a list of tensors to the whole group

Figure 14

Collective Operations

⌛ Timeouts

• Collective operations have to be called for each rank to form a complete collective operation.
  • Failure to do so will result in other ranks waiting indefinitely

Large Language Models

Figure 15: Large Language Models have (LLM)s have taken the NLP community world by storm¹.

1. Hannibal046/Awesome-LLM

Emergent Abilities

Figure 16: Emergent abilities of Large Language Models Yao et al. (2023)

Training LLMs

Modern parallelism techniques¹ enable the training of large language models

Figure 17: It’s hungry! Wei et al. (2022)

Figure 18: Visualization from Yang et al. (2023)

1. See my slides on Parallel Training Techniques for additional details

Life-Cycle of the LLM

1. Data collection + preprocessing

2. Pre-training

   • Architecture decisions:
     {model_size, hyperparameters,
parallelism, lr_schedule, ...}

3. Supervised Fine-Tuning

   • Instruction Tuning
   • Alignment

4. Deploy (+ monitor, re-evaluate, etc.)

Figure 19: Pre-training: Virtually all of the compute used during pre-training

Figure from The Illustrated Transformer

Forward Pass

Video

Figure 20: Language Model trained for causal language modeling. Video from: 🤗 Generation with LLMs

Generating Text

Video

Figure 21: Language Model trained for causal language modeling.

Video from: 🤗 Generation with LLMs

Life-Cycle of the LLM: Pre-training

Pre-training: Virtually all of the compute used during pretraining phase

Life-Cycle of the LLM: Fine-Tuning

Figure 22: Fine-tuning¹: Fine-tuning actually updates the model’s weights to make the model better at a certain task.

1. Figure from The Illustrated Transformer

Assistant Models

Figure 23

🍋 ezpz

Clone Repo(s)

#[⭐][07:33:08 AM][foremans@x3101c0s13b0n0][~/tmp]
$ mkdir ~/tmp/polaris-talk

#[⭐][07:33:21 AM][foremans@x3101c0s13b0n0][~/tmp]
$ cd ~/tmp/polaris-talk

#[⭐][07:33:25 AM][foremans@x3101c0s13b0n0][~/tmp/polaris-talk]
$ NOW=$(tstamp) && mkdir "${NOW}" && cd "${NOW}" # && mkdir "core-dumps-${NOW}" && mv -v **core\.** "core-dumps-${NOW}" && mv "core-dumps-${NOW}" core-dumps

#[⭐][07:33:27 AM][foremans@x3101c0s13b0n0][~/tmp/polaris-talk/2024-07-17-073327]
$ pwd
/home/foremans/tmp/polaris-talk/2024-07-17-073327

#[⭐][07:33:31 AM][foremans@x3101c0s13b0n0][~/tmp/polaris-talk/2024-07-17-073327]
$ git clone https://github.com/saforem2/ezpz ezpz && git clone https://github.com/saforem2/wordplay wordplay
Cloning into 'ezpz'...
remote: Enumerating objects: 2134, done.`
remote: Counting objects: 100% (363/363), done.
remote: Compressing objects: 100% (169/169), done.
remote: Total 2134 (delta 197), reused 265 (delta 141), pack-reused 1771
Receiving objects: 100% (2134/2134), 4.27 MiB | 25.01 MiB/s, done.
Resolving deltas: 100% (1117/1117), done.
Cloning into 'wordplay'...
remote: Enumerating objects: 869, done.
remote: Counting objects: 100% (72/72), done.
remote: Compressing objects: 100% (37/37), done.
remote: Total 869 (delta 29), reused 56 (delta 23), pack-reused 797
Receiving objects: 100% (869/869), 14.36 MiB | 46.54 MiB/s, done.
Resolving deltas: 100% (395/395), done.

Setup Python

#[⭐][07:33:53 AM][foremans@x3101c0s13b0n0][~/tmp/polaris-talk/2024-07-17-073327]
$ source ezpz/src/ezpz/bin/utils.sh && ezpz_setup_python && ezpz_setup_alcf
Unable to detect PBS or SLURM working directory info...
Using /home/foremans/tmp/polaris-talk/2024-07-17-073327 as working directory...
Using WORKING_DIR: /home/foremans/tmp/polaris-talk/2024-07-17-073327
No conda_prefix OR virtual_env found in environment...
Setting up conda...
Lmod is automatically replacing "nvhpc/23.9" with "gcc-native/12.3".
Lmod is automatically replacing "PrgEnv-nvhpc/8.5.0" with "PrgEnv-gnu/8.5.0".
Due to MODULEPATH changes, the following have been reloaded:
  1) cray-mpich/8.1.28
Found conda at: /soft/applications/conda/2024-04-29/mconda3
No VIRTUAL_ENV found in environment!
    - Trying to setup from /soft/applications/conda/2024-04-29/mconda3
    - Using VENV_DIR=/home/foremans/tmp/polaris-talk/2024-07-17-073327/venvs/2024-04-29
    - Creating a new virtual env on top of 2024-04-29 in /home/foremans/tmp/polaris-talk/2024-07-17-073327/venvs/2024-04-29
[python] Using /home/foremans/tmp/polaris-talk/2024-07-17-073327/venvs/2024-04-29/bin/python3

[ezpz/bin/utils.sh]

[2024-07-17-073407]
    • USER=foremans
    • MACHINE=polaris
    • HOST=x3101c0s13b0n0

[ezpz_setup_host]
    • Using hostfile: /var/spool/pbs/aux/2024084.polaris-pbs-01.hsn.cm.polaris.alcf.anl.gov
    • Found in environment:
        • HOSTFILE: /var/spool/pbs/aux/2024084.polaris-pbs-01.hsn.cm.polaris.alcf.anl.gov
        • Writing PBS vars to: /home/foremans/.pbsenv

[ezpz_save_pbs_env]
    • Setting:
        • HOSTFILE: /var/spool/pbs/aux/2024084.polaris-pbs-01.hsn.cm.polaris.alcf.anl.gov
        • JOBENV_FILE: /home/foremans/.pbsenv

[HOSTS]
    • [host:0] - x3101c0s13b0n0.hsn.cm.polaris.alcf.anl.gov

[DIST INFO]
    • HOSTFILE=/var/spool/pbs/aux/2024084.polaris-pbs-01.hsn.cm.polaris.alcf.anl.gov
    • NHOSTS=1
    • NGPU_PER_HOST=4
    • NGPUS=4
    • DIST_LAUNCH=mpiexec --verbose --envall -n 4 -ppn 4 --hostfile /var/spool/pbs/aux/2024084.polaris-pbs-01.hsn.cm.polaris.alcf.anl.gov --cpu-bind depth -d 16

[LAUNCH]:
    • To launch across all available GPUs, use: launch
      launch = mpiexec --verbose --envall -n 4 -ppn 4 --hostfile /var/spool/pbs/aux/2024084.polaris-pbs-01.hsn.cm.polaris.alcf.anl.gov --cpu-bind depth -d 16

Install {ezpz, wordplay}

#[⭐][07:34:13 AM][foremans@x3101c0s13b0n0][~/tmp/polaris-talk/2024-07-17-073327]
$ python3 -m pip install -e ezpz wordplay --require-virtualenv
Looking in indexes: https://pypi.org/simple, https://pypi.ngc.nvidia.com
Obtaining file:///home/foremans/tmp/polaris-talk/2024-07-17-073327/ezpz
  Installing build dependencies ... done
  Checking if build backend supports build_editable ... done
  Getting requirements to build editable ... done
  Installing backend dependencies ... done
  Preparing editable metadata (pyproject.toml) ... done

# ...[clipped]...

Successfully built ezpz
Installing collected packages: enum34, wordplay, pyinstrument, ezpz
  Attempting uninstall: ezpz
    Found existing installation: ezpz 0.1
    Not uninstalling ezpz at /home/foremans/.local/polaris/conda/2024-04-29/lib/python3.11/site-packages, outside environment /home/foremans/tmp/polaris-talk/2024-07-17-073327/venvs/2024-04-29
    Cant uninstall 'ezpz'. No files were found to uninstall.
Successfully installed enum34-1.1.10 ezpz pyinstrument-4.6.2 wordplay-1.0.0a4
[notice] A new release of pip is available: 24.0 -> 24.1.2
[notice] To update, run: pip install --upgrade pip
9.62s user 1.11s system 61% cpu 17.505s total

#[⭐][07:34:53 AM][foremans@x3101c0s13b0n0][~/tmp/polaris-talk/2024-07-17-073327]
$ python3 -m pip install --upgrade wandb
Looking in indexes: https://pypi.org/simple, https://pypi.ngc.nvidia.com
Requirement already satisfied: wandb in /soft/applications/conda/2024-04-29/mconda3/lib/python3.11/site-packages (0.16.6)
Collecting wandb
  Downloading wandb-0.17.4-py3-none-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (10 kB)
Downloading wandb-0.17.4-py3-none-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (6.9 MB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 6.9/6.9 MB 2.1 MB/s eta 0:00:00
Installing collected packages: wandb
  Attempting uninstall: wandb
    Found existing installation: wandb 0.16.6
    Not uninstalling wandb at /soft/applications/conda/2024-04-29/mconda3/lib/python3.11/site-packages, outside environment /home/foremans/tmp/polaris-talk/2024-07-17-073327/venvs/2024-04-29
    Cant uninstall 'wandb'. No files were found to uninstall.
Successfully installed wandb-0.17.4
[notice] A new release of pip is available: 24.0 -> 24.1.2
[notice] To update, run: pip install --upgrade pip

Launch ezpz.test_dist

#(👻 2024-04-29)
#[⭐][07:34:07 AM][foremans@x3101c0s13b0n0][~/tmp/polaris-talk/2024-07-17-073327][⏱ 7s]
$ which launch
launch: aliased to mpiexec --verbose --envall -n 4 -ppn 4 --hostfile /var/spool/pbs/aux/2024084.polaris-pbs-01.hsn.cm.polaris.alcf.anl.gov --cpu-bind depth -d 16

#(👻 2024-04-29)
#[⭐][07:34:11 AM][foremans@x3101c0s13b0n0][~/tmp/polaris-talk/2024-07-17-073327]
$ which python3
/home/foremans/tmp/polaris-talk/2024-07-17-073327/venvs/2024-04-29/bin/python3

#(👻 2024-04-29)
#[⭐][07:35:21 AM][foremans@x3101c0s13b0n0][~/tmp/polaris-talk/2024-07-17-073327][⏱ 14s]
$ launch python3 -m ezpz.test_dist | tee ezpz-test-dist-DDP.log
Connected to tcp://x3101c0s13b0n0.hsn.cm.polaris.alcf.anl.gov:7919
Found executable /home/foremans/tmp/polaris-talk/2024-07-17-073327/venvs/2024-04-29/bin/python3
Launching application cff755ee-557e-4df2-a987-db85a8b7dbe7
[2024-07-17 07:35:30.304306][INFO][__init__:156] - Setting logging level to 'INFO' on 'RANK == 0'
[2024-07-17 07:35:30.307036][INFO][__init__:157] - Setting logging level to 'CRITICAL' on all others 'RANK != 0'
[2024-07-17 07:35:30.307494][INFO][__init__:160] - To disable this behavior, and log from ALL ranks (not recommended), set: 'export LOG_FROM_ALL_RANKS=1'  in your environment, and re-run.
[2024-07-17 07:35:32.116037][INFO][dist:358] - [device='cuda'][rank=2/3][local_rank=2/3][node=0/0]
[2024-07-17 07:35:32.116089][INFO][dist:358] - [device='cuda'][rank=3/3][local_rank=3/3][node=0/0]
[2024-07-17 07:35:32.116940][INFO][dist:358] - [device='cuda'][rank=1/3][local_rank=1/3][node=0/0]
[2024-07-17 07:35:32.122726][INFO][dist:95] -
[dist_info]:
  • DEVICE=cuda
  • DEVICE_ID=cuda:0
  • DISTRIBUTED_BACKEND=nccl
  • GPUS_PER_NODE=4
  • HOSTS=['x3101c0s13b0n0.hsn.cm.polaris.alcf.anl.gov']
  • HOSTFILE=/var/spool/pbs/aux/2024084.polaris-pbs-01.hsn.cm.polaris.alcf.anl.gov
  • HOSTNAME=x3101c0s13b0n0.hsn.cm.polaris.alcf.anl.gov
  • LOCAL_RANK=0
  • MACHINE=Polaris
  • NUM_NODES=1
  • NGPUS=4
  • NGPUS_AVAILABLE=4
  • NODE_ID=0
  • RANK=0
  • SCHEDULER=PBS
  • WORLD_SIZE_TOTAL=4
  • WORLD_SIZE_IN_USE=4
  • LAUNCH_CMD=mpiexec --verbose --envall -n 4 -ppn 4 --hostfile /var/spool/pbs/aux/2024084.polaris-pbs-01.hsn.cm.polaris.alcf.anl.gov --cpu-bind depth -d 16
[2024-07-17 07:35:32.124800][INFO][dist:725] - [0/4] Using device='cuda' with backend='DDP' + 'nccl' for distributed training.
[2024-07-17 07:35:32.129169][INFO][dist:358] - [device='cuda'][rank=0/3][local_rank=0/3][node=0/0]
[2024-07-17 07:35:32.129674][WARNING][dist:364] - Using [4 / 4] available "cuda" devices !!
[2024-07-17 07:35:32.130219][INFO][dist:874] - Setting up wandb from rank: 0
[2024-07-17 07:35:32.130638][INFO][dist:875] - Using: WB PROJECT: ezpz.test_dist
wandb: Using wandb-core as the SDK backend. Please refer to https://wandb.me/wandb-core for more information.
wandb: Currently logged in as: foremans (aurora_gpt). Use `wandb login --relogin` to force relogin
wandb: Tracking run with wandb version 0.17.4
wandb: Run data is saved locally in /home/foremans/tmp/polaris-talk/2024-07-17-073327/wandb/run-20240717_073532-p49rzxtv
wandb: Run `wandb offline` to turn off syncing.
wandb: Syncing run vibrant-river-284
wandb: ⭐️ View project at https://wandb.ai/aurora_gpt/ezpz.test_dist
wandb: 🚀 View run at https://wandb.ai/aurora_gpt/ezpz.test_dist/runs/p49rzxtv
[2024-07-17 07:35:33.171085][INFO][dist:905] - W&B RUN: [vibrant-river-284](https://wandb.ai/aurora_gpt/ezpz.test_dist/runs/p49rzxtv)
[2024-07-17 07:35:33.182307][INFO][dist:312] - Updating wandb.run: vibrant-river-284 config with "DIST_INFO"
[2024-07-17 07:35:33.186499][INFO][dist:938] - Running on machine='Polaris'
[2024-07-17 07:35:33.187790][INFO][dist:95] -
[timers_import]:
  • os=1.082196831703186e-06
  • logging=4.507601261138916e-07
  • typing=2.9457733035087585e-06
  • pathlib=1.3122335076332092e-06
  • ezpz=6.109476089477539e-07
  • torch=2.9457733035087585e-06
  • torch_ddp=2.314336597919464e-06
  • wandb=1.842435449361801e-05
  • total=3.0086375772953033e-05

[2024-07-17 07:35:33.188979][INFO][dist:95] -

[CONFIG]:
  • warmup=0
  • log_freq=1
  • batch_size=64
  • input_size=128
  • output_size=128
  • dtype=torch.float32
  • device=cuda
  • world_size=4
  • train_iters=100

[2024-07-17 07:35:34.761945][INFO][test_dist:183] - model=Network(
  (layers): Sequential(
    (0): Linear(in_features=128, out_features=1024, bias=True)
    (1): Linear(in_features=1024, out_features=512, bias=True)
    (2): Linear(in_features=512, out_features=256, bias=True)
    (3): Linear(in_features=256, out_features=128, bias=True)
    (4): Linear(in_features=128, out_features=128, bias=True)
  )
)
[2024-07-17 07:35:36.943300][INFO][test_dist:274] - iter=1, loss=2152.41, sps=1.697e+04, dt=0.00377066, dtf=0.001003, dtb=0.002768
[2024-07-17 07:35:36.948048][INFO][test_dist:274] - iter=2, loss=1577.24, sps=3.611e+04, dt=0.00177221, dtf=0.0005256, dtb=0.001247
[2024-07-17 07:35:36.952085][INFO][test_dist:274] - iter=3, loss=1201.25, sps=3.59e+04, dt=0.00178271, dtf=0.0004875, dtb=0.001295
[2024-07-17 07:35:36.956071][INFO][test_dist:274] - iter=4, loss=1034.03, sps=3.704e+04, dt=0.0017279, dtf=0.0005082, dtb=0.00122
[2024-07-17 07:35:36.959944][INFO][test_dist:274] - iter=5, loss=875.796, sps=3.825e+04, dt=0.00167313, dtf=0.0005121, dtb=0.001161
[2024-07-17 07:35:36.963806][INFO][test_dist:274] - iter=6, loss=817.544, sps=3.804e+04, dt=0.00168248, dtf=0.0004651, dtb=0.001217
[2024-07-17 07:35:36.967806][INFO][test_dist:274] - iter=7, loss=734.838, sps=3.536e+04, dt=0.0018099, dtf=0.0004969, dtb=0.001313
[2024-07-17 07:35:36.971741][INFO][test_dist:274] - iter=8, loss=741.583, sps=3.682e+04, dt=0.00173809, dtf=0.0004537, dtb=0.001284
[2024-07-17 07:35:36.975672][INFO][test_dist:274] - iter=9, loss=738.157, sps=3.717e+04, dt=0.0017217, dtf=0.0004635, dtb=0.001258
[2024-07-17 07:35:36.979537][INFO][test_dist:274] - iter=10, loss=727.255, sps=3.857e+04, dt=0.00165911, dtf=0.0004897, dtb=0.001169
[2024-07-17 07:35:36.983367][INFO][test_dist:274] - iter=11, loss=715.534, sps=3.979e+04, dt=0.00160845, dtf=0.0004246, dtb=0.001184
[2024-07-17 07:35:36.987262][INFO][test_dist:274] - iter=12, loss=693.96, sps=3.791e+04, dt=0.00168827, dtf=0.0004543, dtb=0.001234
[2024-07-17 07:35:36.991156][INFO][test_dist:274] - iter=13, loss=693.518, sps=3.815e+04, dt=0.00167748, dtf=0.0004182, dtb=0.001259
[2024-07-17 07:35:36.994942][INFO][test_dist:274] - iter=14, loss=675.289, sps=4.003e+04, dt=0.00159879, dtf=0.0004048, dtb=0.001194
[2024-07-17 07:35:36.999681][INFO][test_dist:274] - iter=15, loss=677.706, sps=4.062e+04, dt=0.0015755, dtf=0.0004248, dtb=0.001151
[2024-07-17 07:35:37.003599][INFO][test_dist:274] - iter=16, loss=671.639, sps=3.754e+04, dt=0.00170499, dtf=0.000416, dtb=0.001289
[2024-07-17 07:35:37.007565][INFO][test_dist:274] - iter=17, loss=652.219, sps=3.704e+04, dt=0.00172777, dtf=0.0004208, dtb=0.001307
[2024-07-17 07:35:37.011753][INFO][test_dist:274] - iter=18, loss=633.308, sps=3.191e+04, dt=0.00200554, dtf=0.0004193, dtb=0.001586
[2024-07-17 07:35:37.015595][INFO][test_dist:274] - iter=19, loss=635.459, sps=3.845e+04, dt=0.0016645, dtf=0.0004236, dtb=0.001241
[2024-07-17 07:35:37.019356][INFO][test_dist:274] - iter=20, loss=626.979, sps=4.033e+04, dt=0.00158685, dtf=0.0004225, dtb=0.001164
[2024-07-17 07:35:37.023081][INFO][test_dist:274] - iter=21, loss=612.352, sps=4.105e+04, dt=0.00155914, dtf=0.0004169, dtb=0.001142
[2024-07-17 07:35:37.026861][INFO][test_dist:274] - iter=22, loss=609.89, sps=4.004e+04, dt=0.00159827, dtf=0.0004155, dtb=0.001183
[2024-07-17 07:35:37.030555][INFO][test_dist:274] - iter=23, loss=602.673, sps=4.258e+04, dt=0.00150295, dtf=0.0004166, dtb=0.001086
[2024-07-17 07:35:37.034382][INFO][test_dist:274] - iter=24, loss=613.106, sps=3.918e+04, dt=0.00163367, dtf=0.0004164, dtb=0.001217
[2024-07-17 07:35:37.038129][INFO][test_dist:274] - iter=25, loss=644.755, sps=4.173e+04, dt=0.00153368, dtf=0.0004175, dtb=0.001116
[2024-07-17 07:35:37.041943][INFO][test_dist:274] - iter=26, loss=789.106, sps=4.049e+04, dt=0.00158053, dtf=0.0004397, dtb=0.001141
[2024-07-17 07:35:37.045705][INFO][test_dist:274] - iter=27, loss=691.36, sps=4.166e+04, dt=0.00153641, dtf=0.0004157, dtb=0.001121
[2024-07-17 07:35:37.049496][INFO][test_dist:274] - iter=28, loss=657.228, sps=4.018e+04, dt=0.00159288, dtf=0.0004209, dtb=0.001172
[2024-07-17 07:35:37.053229][INFO][test_dist:274] - iter=29, loss=633.212, sps=4.19e+04, dt=0.0015274, dtf=0.0004288, dtb=0.001099
[2024-07-17 07:35:37.057013][INFO][test_dist:274] - iter=30, loss=640.29, sps=4.012e+04, dt=0.00159538, dtf=0.0004144, dtb=0.001181
[2024-07-17 07:35:37.060722][INFO][test_dist:274] - iter=31, loss=604.287, sps=4.21e+04, dt=0.00152018, dtf=0.000398, dtb=0.001122
[2024-07-17 07:35:37.064489][INFO][test_dist:274] - iter=32, loss=640.15, sps=4.079e+04, dt=0.00156912, dtf=0.0004007, dtb=0.001168
[2024-07-17 07:35:37.068206][INFO][test_dist:274] - iter=33, loss=585.789, sps=4.238e+04, dt=0.00151007, dtf=0.0004199, dtb=0.00109
[2024-07-17 07:35:37.071974][INFO][test_dist:274] - iter=34, loss=591.99, sps=4.053e+04, dt=0.00157917, dtf=0.000434, dtb=0.001145
[2024-07-17 07:35:37.075702][INFO][test_dist:274] - iter=35, loss=618.223, sps=4.168e+04, dt=0.00153538, dtf=0.0004152, dtb=0.00112
[2024-07-17 07:35:37.079496][INFO][test_dist:274] - iter=36, loss=572.365, sps=3.998e+04, dt=0.0016008, dtf=0.0004108, dtb=0.00119
[2024-07-17 07:35:37.083250][INFO][test_dist:274] - iter=37, loss=573.749, sps=4.276e+04, dt=0.00149675, dtf=0.0004123, dtb=0.001084
[2024-07-17 07:35:37.086969][INFO][test_dist:274] - iter=38, loss=580.662, sps=4.136e+04, dt=0.00154751, dtf=0.0004129, dtb=0.001135
[2024-07-17 07:35:37.090636][INFO][test_dist:274] - iter=39, loss=568.836, sps=4.311e+04, dt=0.0014847, dtf=0.000409, dtb=0.001076
[2024-07-17 07:35:37.094396][INFO][test_dist:274] - iter=40, loss=551.294, sps=4.145e+04, dt=0.00154388, dtf=0.0004118, dtb=0.001132
[2024-07-17 07:35:37.098103][INFO][test_dist:274] - iter=41, loss=573.647, sps=4.352e+04, dt=0.00147048, dtf=0.0003977, dtb=0.001073
[2024-07-17 07:35:37.101867][INFO][test_dist:274] - iter=42, loss=545.584, sps=4.257e+04, dt=0.00150354, dtf=0.000433, dtb=0.001071
[2024-07-17 07:35:37.105639][INFO][test_dist:274] - iter=43, loss=544.877, sps=4.322e+04, dt=0.00148085, dtf=0.0004117, dtb=0.001069
[2024-07-17 07:35:37.109471][INFO][test_dist:274] - iter=44, loss=559.886, sps=4.028e+04, dt=0.00158879, dtf=0.0004254, dtb=0.001163
[2024-07-17 07:35:37.113186][INFO][test_dist:274] - iter=45, loss=534.895, sps=4.311e+04, dt=0.00148444, dtf=0.0004153, dtb=0.001069
[2024-07-17 07:35:37.116972][INFO][test_dist:274] - iter=46, loss=536.457, sps=4.099e+04, dt=0.00156151, dtf=0.0004113, dtb=0.00115
[2024-07-17 07:35:37.120710][INFO][test_dist:274] - iter=47, loss=548.508, sps=4.183e+04, dt=0.00152993, dtf=0.0004151, dtb=0.001115
[2024-07-17 07:35:37.124552][INFO][test_dist:274] - iter=48, loss=532.186, sps=4.051e+04, dt=0.0015798, dtf=0.0004379, dtb=0.001142
[2024-07-17 07:35:37.128266][INFO][test_dist:274] - iter=49, loss=519.254, sps=4.272e+04, dt=0.0014981, dtf=0.0004164, dtb=0.001082
[2024-07-17 07:35:37.131975][INFO][test_dist:274] - iter=50, loss=535.535, sps=4.16e+04, dt=0.00153862, dtf=0.0004304, dtb=0.001108
[2024-07-17 07:35:37.135717][INFO][test_dist:274] - iter=51, loss=520.722, sps=4.136e+04, dt=0.00154757, dtf=0.0004158, dtb=0.001132
[2024-07-17 07:35:37.139451][INFO][test_dist:274] - iter=52, loss=513.063, sps=4.147e+04, dt=0.00154317, dtf=0.0004138, dtb=0.001129
[2024-07-17 07:35:37.143231][INFO][test_dist:274] - iter=53, loss=514.546, sps=4.038e+04, dt=0.0015848, dtf=0.0004149, dtb=0.00117
[2024-07-17 07:35:37.146971][INFO][test_dist:274] - iter=54, loss=506.488, sps=4.137e+04, dt=0.00154701, dtf=0.0004132, dtb=0.001134
[2024-07-17 07:35:37.150659][INFO][test_dist:274] - iter=55, loss=503.01, sps=4.319e+04, dt=0.0014817, dtf=0.000415, dtb=0.001067
[2024-07-17 07:35:37.154441][INFO][test_dist:274] - iter=56, loss=506.116, sps=4.06e+04, dt=0.00157637, dtf=0.0004211, dtb=0.001155
[2024-07-17 07:35:37.158180][INFO][test_dist:274] - iter=57, loss=485.523, sps=4.287e+04, dt=0.00149301, dtf=0.000414, dtb=0.001079
[2024-07-17 07:35:37.161931][INFO][test_dist:274] - iter=58, loss=489.076, sps=4.185e+04, dt=0.00152915, dtf=0.0004162, dtb=0.001113
[2024-07-17 07:35:37.165759][INFO][test_dist:274] - iter=59, loss=484.844, sps=4.134e+04, dt=0.00154802, dtf=0.0004119, dtb=0.001136
[2024-07-17 07:35:37.169483][INFO][test_dist:274] - iter=60, loss=496.104, sps=4.209e+04, dt=0.00152069, dtf=0.0003993, dtb=0.001121
[2024-07-17 07:35:37.173190][INFO][test_dist:274] - iter=61, loss=467.599, sps=4.221e+04, dt=0.00151621, dtf=0.0004142, dtb=0.001102
[2024-07-17 07:35:37.176950][INFO][test_dist:274] - iter=62, loss=480.055, sps=4.187e+04, dt=0.00152868, dtf=0.0004138, dtb=0.001115
[2024-07-17 07:35:37.181194][INFO][test_dist:274] - iter=63, loss=483.146, sps=3.656e+04, dt=0.00175062, dtf=0.0006253, dtb=0.001125
[2024-07-17 07:35:37.185018][INFO][test_dist:274] - iter=64, loss=479.273, sps=4.099e+04, dt=0.00156151, dtf=0.0004447, dtb=0.001117
[2024-07-17 07:35:37.188752][INFO][test_dist:274] - iter=65, loss=464.753, sps=4.189e+04, dt=0.00152781, dtf=0.0004161, dtb=0.001112
[2024-07-17 07:35:37.192464][INFO][test_dist:274] - iter=66, loss=462.583, sps=4.188e+04, dt=0.00152824, dtf=0.0004138, dtb=0.001114
[2024-07-17 07:35:37.196126][INFO][test_dist:274] - iter=67, loss=461.665, sps=4.272e+04, dt=0.00149801, dtf=0.0004293, dtb=0.001069
[2024-07-17 07:35:37.199838][INFO][test_dist:274] - iter=68, loss=465.25, sps=4.118e+04, dt=0.00155412, dtf=0.0004298, dtb=0.001124
[2024-07-17 07:35:37.203602][INFO][test_dist:274] - iter=69, loss=460.897, sps=4.01e+04, dt=0.00159593, dtf=0.0004131, dtb=0.001183
[2024-07-17 07:35:37.207372][INFO][test_dist:274] - iter=70, loss=456.136, sps=4.106e+04, dt=0.00155887, dtf=0.00041, dtb=0.001149
[2024-07-17 07:35:37.211089][INFO][test_dist:274] - iter=71, loss=447.565, sps=4.158e+04, dt=0.00153923, dtf=0.0004113, dtb=0.001128
[2024-07-17 07:35:37.214861][INFO][test_dist:274] - iter=72, loss=444.733, sps=4.05e+04, dt=0.00158026, dtf=0.0004127, dtb=0.001168
[2024-07-17 07:35:37.218601][INFO][test_dist:274] - iter=73, loss=459.152, sps=4.123e+04, dt=0.00155234, dtf=0.0004201, dtb=0.001132
[2024-07-17 07:35:37.222334][INFO][test_dist:274] - iter=74, loss=444.6, sps=4.226e+04, dt=0.00151444, dtf=0.0004371, dtb=0.001077
[2024-07-17 07:35:37.226042][INFO][test_dist:274] - iter=75, loss=439.884, sps=4.29e+04, dt=0.001492, dtf=0.0004154, dtb=0.001077
[2024-07-17 07:35:37.229838][INFO][test_dist:274] - iter=76, loss=438.578, sps=4.086e+04, dt=0.00156632, dtf=0.0004418, dtb=0.001125
[2024-07-17 07:35:37.233560][INFO][test_dist:274] - iter=77, loss=431.993, sps=4.327e+04, dt=0.00147909, dtf=0.0004096, dtb=0.00107
[2024-07-17 07:35:37.237367][INFO][test_dist:274] - iter=78, loss=422.338, sps=4.057e+04, dt=0.00157754, dtf=0.0004468, dtb=0.001131
[2024-07-17 07:35:37.241117][INFO][test_dist:274] - iter=79, loss=427.973, sps=4.288e+04, dt=0.00149254, dtf=0.000415, dtb=0.001077
[2024-07-17 07:35:37.244895][INFO][test_dist:274] - iter=80, loss=418.703, sps=4.06e+04, dt=0.00157617, dtf=0.0004137, dtb=0.001162
[2024-07-17 07:35:37.248740][INFO][test_dist:274] - iter=81, loss=427.645, sps=4.031e+04, dt=0.00158766, dtf=0.000415, dtb=0.001173
[2024-07-17 07:35:37.252447][INFO][test_dist:274] - iter=82, loss=417.629, sps=4.227e+04, dt=0.00151406, dtf=0.0004149, dtb=0.001099
[2024-07-17 07:35:37.256190][INFO][test_dist:274] - iter=83, loss=411.667, sps=4.189e+04, dt=0.00152778, dtf=0.0004357, dtb=0.001092
[2024-07-17 07:35:37.259935][INFO][test_dist:274] - iter=84, loss=409.366, sps=4.144e+04, dt=0.0015445, dtf=0.0004575, dtb=0.001087
[2024-07-17 07:35:37.263677][INFO][test_dist:274] - iter=85, loss=409.511, sps=4.232e+04, dt=0.00151228, dtf=0.0004035, dtb=0.001109
[2024-07-17 07:35:37.267463][INFO][test_dist:274] - iter=86, loss=409.593, sps=4.101e+04, dt=0.00156049, dtf=0.0004028, dtb=0.001158
[2024-07-17 07:35:37.271174][INFO][test_dist:274] - iter=87, loss=408.794, sps=4.3e+04, dt=0.00148828, dtf=0.0004006, dtb=0.001088
[2024-07-17 07:35:37.274926][INFO][test_dist:274] - iter=88, loss=403.151, sps=4.091e+04, dt=0.00156441, dtf=0.000415, dtb=0.001149
[2024-07-17 07:35:37.278633][INFO][test_dist:274] - iter=89, loss=402.182, sps=4.26e+04, dt=0.00150243, dtf=0.0004147, dtb=0.001088
[2024-07-17 07:35:37.282372][INFO][test_dist:274] - iter=90, loss=387.829, sps=4.216e+04, dt=0.00151793, dtf=0.0004411, dtb=0.001077
[2024-07-17 07:35:37.286102][INFO][test_dist:274] - iter=91, loss=393.108, sps=4.308e+04, dt=0.00148558, dtf=0.0004167, dtb=0.001069
[2024-07-17 07:35:37.289904][INFO][test_dist:274] - iter=92, loss=389.039, sps=4.103e+04, dt=0.00155996, dtf=0.0004359, dtb=0.001124
[2024-07-17 07:35:37.293618][INFO][test_dist:274] - iter=93, loss=383.54, sps=4.322e+04, dt=0.00148092, dtf=0.0004147, dtb=0.001066
[2024-07-17 07:35:37.297401][INFO][test_dist:274] - iter=94, loss=384.459, sps=4.1e+04, dt=0.00156106, dtf=0.0004164, dtb=0.001145
[2024-07-17 07:35:37.301172][INFO][test_dist:274] - iter=95, loss=376.397, sps=4.191e+04, dt=0.0015272, dtf=0.0004129, dtb=0.001114
[2024-07-17 07:35:37.304924][INFO][test_dist:274] - iter=96, loss=389.544, sps=4.091e+04, dt=0.00156433, dtf=0.0004139, dtb=0.00115
[2024-07-17 07:35:37.308641][INFO][test_dist:274] - iter=97, loss=365.041, sps=4.343e+04, dt=0.00147362, dtf=0.0004165, dtb=0.001057
[2024-07-17 07:35:37.312398][INFO][test_dist:274] - iter=98, loss=358.427, sps=4.134e+04, dt=0.00154796, dtf=0.0004143, dtb=0.001134
[2024-07-17 07:35:37.561881][INFO][test_dist:274] - iter=99, loss=375.596, sps=258.9, dt=0.247161, dtf=0.1969, dtb=0.05026

                            train/dt [2024-07-17-073537]
     ┌─────────────────────────────────────────────────────────────────────────┐
0.247┤                                                                        ▝│
     │                                                                         │
     │                                                                         │
0.206┤                                                                         │
     │                                                                         │
     │                                                                         │
0.165┤                                                                         │
     │                                                                         │
0.124┤                                                                         │
     │                                                                         │
     │                                                                         │
0.083┤                                                                         │
     │                                                                         │
     │                                                                         │
0.042┤                                                                         │
     │                                                                         │
     │                                                                         │
0.001┤▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▄▖▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▄▗▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▄▗▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▖│
     └┬─────────────────┬─────────────────┬─────────────────┬─────────────────┬┘
     1.0              25.5              50.0              74.5             99.0
train/dt                                iter
[2024-07-17 07:35:37.589287][INFO][plot:156] - Appending plot to: /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/dt.txt
text saved in /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/dt.txt
                            train/dtf [2024-07-17-073537]
     ┌─────────────────────────────────────────────────────────────────────────┐
0.197┤                                                                        ▝│
     │                                                                         │
     │                                                                         │
0.164┤                                                                         │
     │                                                                         │
     │                                                                         │
0.131┤                                                                         │
     │                                                                         │
0.099┤                                                                         │
     │                                                                         │
     │                                                                         │
0.066┤                                                                         │
     │                                                                         │
     │                                                                         │
0.033┤                                                                         │
     │                                                                         │
     │                                                                         │
0.000┤▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▄▖▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▄▗▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▄▗▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▄▗▖▖│
     └┬─────────────────┬─────────────────┬─────────────────┬─────────────────┬┘
     1.0              25.5              50.0              74.5             99.0
train/dtf                               iter
[2024-07-17 07:35:37.603242][INFO][plot:156] - Appending plot to: /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/dtf.txt
text saved in /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/dtf.txt
                             train/dtb [2024-07-17-073537]
      ┌────────────────────────────────────────────────────────────────────────┐
0.0503┤                                                                       ▝│
      │                                                                        │
      │                                                                        │
0.0421┤                                                                        │
      │                                                                        │
      │                                                                        │
0.0339┤                                                                        │
      │                                                                        │
0.0257┤                                                                        │
      │                                                                        │
      │                                                                        │
0.0175┤                                                                        │
      │                                                                        │
      │                                                                        │
0.0093┤                                                                        │
      │                                                                        │
      │                                                                        │
0.0011┤▚▗▖▄▗▖▄▗▖▄▖▄▗▖▄▗▖▄▖▄▗▖▄▗▖▄▗▄▗▖▄▗▖▄▗▖▄▖▄▗▖▄▗▖▄▖▄▗▖▄▗▖▄▗▄▗▖▄▗▖▄▗▄▗▖▄▗▖▄▗▖▖│
      └┬─────────────────┬─────────────────┬────────────────┬─────────────────┬┘
      1.0              25.5              50.0             74.5             99.0
train/dtb                                iter
[2024-07-17 07:35:37.615896][INFO][plot:156] - Appending plot to: /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/dtb.txt
text saved in /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/dtb.txt
                            train/loss [2024-07-17-073537]
      ┌────────────────────────────────────────────────────────────────────────┐
2152.4┤▘                                                                       │
      │                                                                        │
      │                                                                        │
1853.4┤                                                                        │
      │                                                                        │
      │▗                                                                       │
1554.4┤                                                                        │
      │                                                                        │
1255.4┤                                                                        │
      │ ▗                                                                      │
      │                                                                        │
 956.4┤  ▘                                                                     │
      │   ▖                                                                    │
      │   ▝              ▖                                                     │
 657.4┤    ▝▘▀▝▘▚▖▄     ▗ ▄                                                    │
      │            ▝▘▀▝▘▘  ▝▘▀▗▘▚▗▄▗▖▄▗ ▗                                      │
      │                                ▘▘▝▘▀▘▀▝▘▞▗▘▄▖▄▗▖▄▗▖▄▗▄                 │
 358.4┤                                                       ▝▘▀▝▘▀▝▀▝▘▀▝▖▚▝▖▄│
      └┬─────────────────┬─────────────────┬────────────────┬─────────────────┬┘
      1.0              25.5              50.0             74.5             99.0
train/loss                               iter
[2024-07-17 07:35:37.655339][INFO][plot:156] - Appending plot to: /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/loss.txt
text saved in /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/loss.txt
                           train/iter [2024-07-17-073537]
    ┌──────────────────────────────────────────────────────────────────────────┐
99.0┤                                                                      ▗▗▖▀│
    │                                                                   ▄▝▘▘   │
    │                                                              ▗▖▞▝▘       │
82.7┤                                                          ▄▗▘▀            │
    │                                                      ▖▄▝▘                │
    │                                                 ▗▗▖▀▝                    │
66.3┤                                              ▄▝▘▘                        │
    │                                         ▗▖▞▝▘                            │
50.0┤                                     ▄▗▘▀                                 │
    │                                 ▖▄▝▘                                     │
    │                            ▗▗▖▀▝                                         │
33.7┤                         ▄▝▘▘                                             │
    │                    ▗▖▞▝▘                                                 │
    │                ▄▗▘▀                                                      │
17.3┤            ▖▄▝▘                                                          │
    │       ▗▗▖▀▝                                                              │
    │    ▄▝▘▘                                                                  │
 1.0┤▖▞▝▘                                                                      │
    └┬─────────────────┬──────────────────┬─────────────────┬─────────────────┬┘
    1.0              25.5               50.0              74.5             99.0
train/iter                              iter
[2024-07-17 07:35:37.669214][INFO][plot:156] - Appending plot to: /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/iter.txt
text saved in /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/iter.txt
                             train/sps [2024-07-17-073537]
       ┌───────────────────────────────────────────────────────────────────────┐
43523.3┤                ▖▗  ▖▗ ▖▗ ▖▝ ▚▘▝ ▖▗    ▘▗▖▗▖▖ ▖▄    ▗▖▝ ▖ ▗▖▗ ▘▗▞ ▘▗ ▘ │
       │       ▖ ▗▘  ▗▝▖  ▀▗ ▖▝▝ ▖▝ ▘  ▖▝ ▘▝▀▗▘▝ ▝   ▝  ▘▞▝▘▘ ▘▝ ▚ ▝ ▘▝  ▝ ▘▝ ▘│
       │  ▖▀ ▖▞ ▞  ▄ ▘  ▝                                                      │
36312.5┤▝▝  ▗                                       ▝                          │
       │            ▖                                                          │
       │                                                                       │
29101.8┤                                                                       │
       │                                                                       │
21891.1┤                                                                       │
       │                                                                       │
       │▖                                                                      │
14680.4┤                                                                       │
       │                                                                       │
       │                                                                       │
 7469.7┤                                                                       │
       │                                                                       │
       │                                                                       │
  258.9┤                                                                      ▗│
       └┬─────────────────┬────────────────┬─────────────────┬────────────────┬┘
       1.0              25.5             50.0              74.5            99.0
train/sps                                iter
[2024-07-17 07:35:37.681268][INFO][plot:156] - Appending plot to: /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/sps.txt
text saved in /home/foremans/tmp/polaris-talk/2024-07-17-073327/test-dist-plots/train/sps.txt

PyInstrument Profile

Recorded: 07:35:34  Samples:  2227
Duration: 2.948     CPU time: 5.441
PyInstrument: v4.6.2
Program: /home/foremans/tmp/polaris-talk/2024-07-17-073327/ezpz/src/ezpz/test_dist.py
2.948 <module>  ezpz/test_dist.py:1
└─ 2.946 main  ezpz/test_dist.py:217
   ├─ 2.043 build_model_and_optimizer  ezpz/test_dist.py:171
   │  └─ 2.011 Adam.__init__  torch/optim/adam.py:15
   │        [129 frames hidden]  torch, wandb, transformers, jax, func...
   ├─ 0.326 _forward_step  ezpz/test_dist.py:231
   │  ├─ 0.279 DistributedDataParallel._wrapped_call_impl  torch/nn/modules/module.py:1528
   │  │     [13 frames hidden]  torch, wandb, <built-in>
   │  │        0.273 Network._call_impl  torch/nn/modules/module.py:1534
   │  │        └─ 0.076 Network.forward  ezpz/test_dist.py:164
   │  │           └─ 0.076 Sequential._wrapped_call_impl  torch/nn/modules/module.py:1528
   │  │                 [7 frames hidden]  torch, <built-in>
   │  └─ 0.046 calc_loss  ezpz/test_dist.py:168
   ├─ 0.254 _backward_step  ezpz/test_dist.py:236
   │  ├─ 0.177 Tensor.backward  torch/_tensor.py:466
   │  │     [4 frames hidden]  torch, <built-in>
   │  └─ 0.077 wrapper  torch/optim/optimizer.py:374
   │        [5 frames hidden]  torch
   ├─ 0.119 tplot_dict  ezpz/plot.py:136
   │  └─ 0.069 show  plotext/_core.py:292
   │        [5 frames hidden]  plotext
   ├─ 0.102 Logger.info  logging/__init__.py:1479
   │     [6 frames hidden]  logging, rich
   │        0.102 RichHandler.emit  rich/logging.py:126
   │        └─ 0.100 Console.print  ezpz/log/console.py:79
   │           └─ 0.100 Console.print  rich/console.py:1624
   │                 [5 frames hidden]  rich
   └─ 0.099 Run.wrapper  wandb/sdk/wandb_run.py:418
         [13 frames hidden]  wandb, json
[2024-07-17 07:35:37.876629][INFO][profile:115] - Saving pyinstrument profile output to: /home/foremans/tmp/polaris-talk/2024-07-17-073327/ezpz_pyinstrument_profiles
[2024-07-17 07:35:37.877255][INFO][profile:123] - PyInstrument profile saved (as html) to:  /home/foremans/tmp/polaris-talk/2024-07-17-073327/ezpz_pyinstrument_profiles/pyinstrument-profile-2024-07-17-073537.html
[2024-07-17 07:35:37.877936][INFO][profile:131] - PyInstrument profile saved (as text) to:  /home/foremans/tmp/polaris-talk/2024-07-17-073327/ezpz_pyinstrument_profiles/pyinstrument-profile-2024-07-17-073537.txt
[2024-07-17 07:35:38.391628][INFO][profile:143] - Finished with pyinstrument profiler. Took: 2.94768s
[2024-07-17 07:35:38.392519][INFO][test_dist:318] - [0] runtime=8.075730s
wandb: 🚀 View run vibrant-river-284 at: https://wandb.ai/aurora_gpt/ezpz.test_dist/runs/p49rzxtv
wandb: Find logs at: wandb/run-20240717_073532-p49rzxtv/logs
Application cff755ee resources: utime=25s stime=23s maxrss=1434396KB inblock=32 oublock=4320 minflt=670179 majflt=864 nvcsw=195893 nivcsw=1331214

Example: ezpz 🍋

• Link¹ to video

Example: using 🍋 ezpz.test_dist to train a small model using DDP

1. idk why it doesn’t render correctly in the slide (seems like refreshing helps?)

Example: wordplay 🎮💬

Figure 24: The simplest, fastest repository for training / finetuning GPT based models.

Prepare Data

#[⭐][07:41:20 AM][foremans@x3101c0s13b0n0][~/tmp/polaris-talk/2024-07-17-073327][⏱ 29s]
$ python3 wordplay/data/shakespeare_char/prepare.py
Using HF_DATASETS_CACHE=/home/foremans/tmp/polaris-talk/2024-07-17-073327/wordplay/data/shakespeare_char/.cache/huggingface
length of dataset in characters: 1,115,394
all the unique characters:
 !$&\',-.3:;?ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz
vocab size: 65
train has 1,003,854 tokens
val has 111,540 tokens

Launch Training (DDP)

#(👻 2024-04-29)
#[⭐][07:42:02 AM][foremans@x3101c0s13b0n0][~/tmp/polaris-talk/2024-07-17-073327]
$ launch python3 -m wordplay train.backend=DDP train.eval_interval=100 data=shakespeare train.dtype=bf16 model.batch_size=64 model.block_size=1024 train.max_iters=1000 train.log_interval=10 train.compile=false | tee wordplay-gpt2-DDP.log
[2024-07-17 07:42:11.746540][INFO][__init__:156] - Setting logging level to 'INFO' on 'RANK == 0'
[2024-07-17 07:42:11.748763][INFO][__init__:157] - Setting logging level to 'CRITICAL' on all others 'RANK != 0'
[2024-07-17 07:42:11.749453][INFO][__init__:160] - To disable this behavior, and log from ALL ranks (not recommended), set: 'export LOG_FROM_ALL_RANKS=1'  in your environment, and re-run.
[2024-07-17 07:42:11.772718][INFO][configs:81] - Setting HF_DATASETS_CACHE to /home/foremans/tmp/polaris-talk/2024-07-17-073327/wordplay/.cache/huggingface/datasets
[2024-07-17 07:42:15.341532][INFO][dist:358] - [device='cuda'][rank=2/3][local_rank=2/3][node=0/0]
[2024-07-17 07:42:15.342381][INFO][dist:358] - [device='cuda'][rank=1/3][local_rank=1/3][node=0/0]
[2024-07-17 07:42:15.342430][INFO][dist:358] - [device='cuda'][rank=3/3][local_rank=3/3][node=0/0]
[2024-07-17 07:42:15.348657][INFO][dist:95] -

[dist_info]:
  • DEVICE=cuda
  • DEVICE_ID=cuda:0
  • DISTRIBUTED_BACKEND=nccl
  • GPUS_PER_NODE=4
  • HOSTS=['x3101c0s13b0n0.hsn.cm.polaris.alcf.anl.gov']
  • HOSTFILE=/var/spool/pbs/aux/2024084.polaris-pbs-01.hsn.cm.polaris.alcf.anl.gov
  • HOSTNAME=x3101c0s13b0n0.hsn.cm.polaris.alcf.anl.gov
  • LOCAL_RANK=0
  • MACHINE=Polaris
  • NUM_NODES=1
  • NGPUS=4
  • NGPUS_AVAILABLE=4
  • NODE_ID=0
  • RANK=0
  • SCHEDULER=PBS
  • WORLD_SIZE_TOTAL=4
  • WORLD_SIZE_IN_USE=4
  • LAUNCH_CMD=mpiexec --verbose --envall -n 4 -ppn 4 --hostfile /var/spool/pbs/aux/2024084.polaris-pbs-01.hsn.cm.polaris.alcf.anl.gov --cpu-bind depth -d 16


[2024-07-17 07:42:15.351446][INFO][dist:725] - [0/4] Using device='cuda' with backend='DDP' + 'nccl' for distributed training.
[2024-07-17 07:42:15.356169][INFO][dist:358] - [device='cuda'][rank=0/3][local_rank=0/3][node=0/0]
[2024-07-17 07:42:15.356692][WARNING][dist:364] - Using [4 / 4] available "cuda" devices !!
[2024-07-17 07:42:15.359571][INFO][configs:317] - Loading val from /home/foremans/tmp/polaris-talk/2024-07-17-073327/wordplay/data/shakespeare_char/val.bin
[2024-07-17 07:42:15.360138][INFO][configs:317] - Loading train from /home/foremans/tmp/polaris-talk/2024-07-17-073327/wordplay/data/shakespeare_char/train.bin
[2024-07-17 07:42:15.361154][INFO][configs:442] - Tokens per iteration: 262,144
[2024-07-17 07:42:15.361574][INFO][configs:465] - Using self.ptdtype=torch.float16 on self.device_type='cuda'
[2024-07-17 07:42:15.362002][INFO][configs:471] - Initializing a new model from scratch
[2024-07-17 07:42:15.362529][INFO][dist:874] - Setting up wandb from rank: 0
[2024-07-17 07:42:15.362896][INFO][dist:875] - Using: WB PROJECT: WordPlay
[2024-07-17 07:42:16.451786][INFO][dist:905] - W&B RUN: [still-frog-17](https://wandb.ai/aurora_gpt/WordPlay/runs/6by9vpcj)
[2024-07-17 07:42:16.464106][INFO][dist:312] - Updating wandb.run: still-frog-17 config with "DIST_INFO"
[2024-07-17 07:42:16.469424][INFO][dist:938] - Running on machine='Polaris'
[2024-07-17 07:42:16.471151][WARNING][__main__:89] - {
    "train": {
        "framework": "pytorch",
        "backend": "DDP",
        "device": null,
        "seed": null,
        "port": null,
        "ds_config_path": null,
        "precision": null,
        "ngpus": null,
        "use_wandb": true,
        "eval_interval": 100,
        "log_interval": 10,
        "eval_iters": 200,
        "eval_only": false,
        "always_save_checkpoint": false,
        "init_from": "scratch",
        "wandb_project": "WordPlay",
        "max_iters": 1000,
        "warmup_iters": 100,
        "dtype": "bf16",
        "compile": false
    },
    "model": {
        "n_layer": 12,
        "n_head": 12,
        "n_embd": 768,
        "batch_size": 64,
        "block_size": 1024,
        "activation": "gelu",
        "dropout": 0.0,
        "bias": false,
        "vocab_size": 65
    },
    "data": {
        "dataset": "shakespeare_char",
        "out_dir": "out-shakespeare-char",
        "root_path": null
    },
    "optimizer": {
        "gas": 1,
        "name": "AdamW",
        "learning_rate": 0.0006,
        "weight_decay": 0.1,
        "beta1": 0.9,
        "beta2": 0.95,
        "grad_clip": 1.0,
        "decay_lr": true,
        "lr_decay_iters": 600000,
        "min_lr": 6e-05
    }
}
[2024-07-17 07:42:16.474305][WARNING][__main__:90] - Output dir: /home/foremans/tmp/polaris-talk/outputs/runs/pytorch/DDP/2024-07-17/07-42-13
[2024-07-17 07:42:16.474922][INFO][trainer:246] - Initializing a new model from scratch
[2024-07-17 07:42:17.258904][INFO][model:255] - number of parameters: 85.00M
[2024-07-17 07:42:17.290004][INFO][trainer:264] - Model size: num_params=85003776
[2024-07-17 07:42:17.292626][INFO][model:445] - num decayed parameter tensors: 50, with 85,771,008 parameters
[2024-07-17 07:42:17.293296][INFO][model:449] - num non-decayed parameter tensors: 25, with 19,200 parameters
[2024-07-17 07:42:17.515324][CRITICAL][trainer:316] - "devid='cuda:1'"
[2024-07-17 07:42:17.515340][CRITICAL][trainer:316] - "devid='cuda:2'"
[2024-07-17 07:42:17.515465][CRITICAL][trainer:316] - "devid='cuda:3'"
[2024-07-17 07:42:18.431814][INFO][model:465] - using fused AdamW: True
[2024-07-17 07:42:18.432620][CRITICAL][trainer:316] - "devid='cuda:0'"
[2024-07-17 07:42:19.951020][INFO][trainer:356] - • self.model=GPT(
  (transformer): ModuleDict(
    (wte): Embedding(65, 768)
    (wpe): Embedding(1024, 768)
    (drop): Dropout(p=0.0, inplace=False)
    (h): ModuleList(
      (0-11): 12 x Block(
        (ln_1): LayerNorm()
        (attn): CausalSelfAttention(
          (c_attn): Linear(in_features=768, out_features=2304, bias=False)
          (c_proj): Linear(in_features=768, out_features=768, bias=False)
          (attn_dropout): Dropout(p=0.0, inplace=False)
          (resid_dropout): Dropout(p=0.0, inplace=False)
        )
        (ln_2): LayerNorm()
        (mlp): MLP(
          (c_fc): Linear(in_features=768, out_features=3072, bias=False)
          (act_fn): GELU(approximate='none')
          (c_proj): Linear(in_features=3072, out_features=768, bias=False)
          (dropout): Dropout(p=0.0, inplace=False)
        )
      )
    )
    (ln_f): LayerNorm()
  )
  (lm_head): Linear(in_features=768, out_features=65, bias=False)
)
[2024-07-17 07:42:19.955340][INFO][trainer:357] - • self.grad_scaler=<torch.cuda.amp.grad_scaler.GradScaler object at 0x145a38f0f090>
[2024-07-17 07:42:19.956897][INFO][trainer:358] - • self.model_engine=DistributedDataParallel(
  (module): GPT(
    (transformer): ModuleDict(
      (wte): Embedding(65, 768)
      (wpe): Embedding(1024, 768)
      (drop): Dropout(p=0.0, inplace=False)
      (h): ModuleList(
        (0-11): 12 x Block(
          (ln_1): LayerNorm()
          (attn): CausalSelfAttention(
            (c_attn): Linear(in_features=768, out_features=2304, bias=False)
            (c_proj): Linear(in_features=768, out_features=768, bias=False)
            (attn_dropout): Dropout(p=0.0, inplace=False)
            (resid_dropout): Dropout(p=0.0, inplace=False)
          )
          (ln_2): LayerNorm()
          (mlp): MLP(
            (c_fc): Linear(in_features=768, out_features=3072, bias=False)
            (act_fn): GELU(approximate='none')
            (c_proj): Linear(in_features=3072, out_features=768, bias=False)
            (dropout): Dropout(p=0.0, inplace=False)
          )
        )
      )
      (ln_f): LayerNorm()
    )
    (lm_head): Linear(in_features=768, out_features=65, bias=False)
  )
)
[2024-07-17 07:42:19.961066][INFO][trainer:359] - • self.optimizer=AdamW (
Parameter Group 0
    amsgrad: False
    betas: (0.9, 0.95)
    capturable: False
    differentiable: False
    eps: 1e-08
    foreach: None
    fused: True
    lr: 0.0006
    maximize: False
    weight_decay: 0.1

Parameter Group 1
    amsgrad: False
    betas: (0.9, 0.95)
    capturable: False
    differentiable: False
    eps: 1e-08
    foreach: None
    fused: True
    lr: 0.0006
    maximize: False
    weight_decay: 0.0
)
[2024-07-17 07:42:19.988827][INFO][trainer:802] - Startup time: 6.7125
                Training Legend
┏━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃    abbr     ┃ desc                           ┃
┡━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│    step     │ Current training iteration     │
│    loss     │ Loss value                     │
│     dt      │ Elapsed time per training step │
│     dtf     │ Elapsed time per forward step  │
│     dtb     │ Elapsed time per backward step │
│     sps     │ Samples per second             │
│ sps_per_gpu │ Samples per second (per GPU)   │
│     tps     │ Tokens per second              │
│ tps_per_gpu │ Tokens per second (per GPU)    │
│     mfu     │ Model flops utilization        │
│ train_loss  │ Training loss value            │
│  val_loss   │ Validation loss value          │
└─────────────┴────────────────────────────────┘
[2024-07-17 07:42:21.451865][INFO][trainer:820] - ['prompt']: 'What is an LLM?'
[2024-07-17 07:42:21.452667][INFO][trainer:824] - ['response']:
What is an LLM?eelEl\'$nltPwBSWal,;PWw bbu\'HiyP\'FWwF &AhW:ygrn kk-\'\'KFlMwnlEfflkc,elpWaWtgml$Pgglhllw lglhFllzczPAFHpeAAPPSltgkrWPPhlEMgcrN ggPWt-WPSSzHSkkrzzk.FFrtSSkgMll&gFXr,hghaueaVPW-pHFF-gg,,,FF,,kbApgg gg\'aWWzzkk\'a\'CggHl$bGeA,FFk,,SF;UF,,aZ ;gglee$,k.US&kg:S,,zVzzc
[2024-07-17 07:43:01.573073][INFO][trainer:885] - step=10 loss=3.154310 dt=0.282833 dtf=0.005247 dtb=0.011417 sps=14.142633 sps_per_gpu=3.535658 tps=926851.609409 tps_per_gpu=231712.902352 mfu=46.288281 train_loss=4.125778 val_loss=4.128809
[2024-07-17 07:43:04.402750][INFO][trainer:885] - step=20 loss=2.660851 dt=0.306263 dtf=0.005233 dtb=0.011419 sps=13.060678 sps_per_gpu=3.265170 tps=855944.613638 tps_per_gpu=213986.153409 mfu=45.934162 train_loss=4.125778 val_loss=4.128809
[2024-07-17 07:43:07.237507][INFO][trainer:885] - step=30 loss=2.543283 dt=0.283021 dtf=0.005238 dtb=0.011245 sps=14.133211 sps_per_gpu=3.533303 tps=926234.088226 tps_per_gpu=231558.522057 mfu=45.966490 train_loss=4.125778 val_loss=4.128809
[2024-07-17 07:43:10.077248][INFO][trainer:885] - step=40 loss=2.503963 dt=0.285001 dtf=0.005213 dtb=0.011471 sps=14.035061 sps_per_gpu=3.508765 tps=919801.749941 tps_per_gpu=229950.437485 mfu=45.963461 train_loss=4.125778 val_loss=4.128809
[2024-07-17 07:43:12.917039][INFO][trainer:885] - step=50 loss=2.477469 dt=0.283532 dtf=0.005166 dtb=0.011294 sps=14.107763 sps_per_gpu=3.526941 tps=924566.380009 tps_per_gpu=231141.595002 mfu=45.984530 train_loss=4.125778 val_loss=4.128809
[2024-07-17 07:43:15.760749][INFO][trainer:885] - step=60 loss=2.471083 dt=0.284630 dtf=0.005140 dtb=0.011224 sps=14.053326 sps_per_gpu=3.513332 tps=920998.786204 tps_per_gpu=230249.696551 mfu=45.985675 train_loss=4.125778 val_loss=4.128809
[2024-07-17 07:43:18.602785][INFO][trainer:885] - step=70 loss=2.458894 dt=0.283926 dtf=0.005219 dtb=0.010383 sps=14.088155 sps_per_gpu=3.522039 tps=923281.352698 tps_per_gpu=230820.338174 mfu=45.998106 train_loss=4.125778 val_loss=4.128809
[2024-07-17 07:43:21.451433][INFO][trainer:885] - step=80 loss=2.489088 dt=0.285537 dtf=0.005183 dtb=0.011373 sps=14.008683 sps_per_gpu=3.502171 tps=918073.060430 tps_per_gpu=229518.265108 mfu=45.983282 train_loss=4.125778 val_loss=4.128809
[2024-07-17 07:43:24.302241][INFO][trainer:885] - step=90 loss=2.471990 dt=0.300767 dtf=0.005445 dtb=0.010290 sps=13.299337 sps_per_gpu=3.324834 tps=871585.359388 tps_per_gpu=217896.339847 mfu=45.737774 train_loss=4.125778 val_loss=4.128809
[2024-07-17 07:43:27.153275][INFO][trainer:885] - step=100 loss=2.445556 dt=0.285869 dtf=0.005182 dtb=0.011251 sps=13.992403 sps_per_gpu=3.498101 tps=917006.151328 tps_per_gpu=229251.537832 mfu=45.743655 train_loss=4.125778 val_loss=4.128809
[2024-07-17 07:43:28.182553][INFO][trainer:820] - ['prompt']: 'What is an LLM?'
[2024-07-17 07:43:28.183179][INFO][trainer:824] - ['response']:

What is an LLM?

Goupay my winghimithell bls ger t bon sinthard ht omind be,
And lereind h py balithand frd oforondof wimon me hageas thinero mand,
Thacanes,
An frift ghik med d herthecke ntore thack couthen ale, t thit ang d m t h chy me fache ag, wit my hathan glat ng
[2024-07-17 07:44:06.025837][INFO][trainer:760] - Saving checkpoint to: /home/foremans/tmp/polaris-talk/outputs/runs/pytorch/DDP/2024-07-17/07-42-13
[2024-07-17 07:44:06.026607][INFO][trainer:761] - Saving model to: /home/foremans/tmp/polaris-talk/outputs/runs/pytorch/DDP/2024-07-17/07-42-13/model.pth
[2024-07-17 07:44:07.682968][INFO][configs:141] - Appending /home/foremans/tmp/polaris-talk/outputs/runs/pytorch/DDP/2024-07-17/07-42-13 to /home/foremans/tmp/polaris-talk/2024-07-17-073327/wordplay/src/ckpts/checkpoints.log
[2024-07-17 07:44:10.519506][INFO][trainer:885] - step=110 loss=2.433923 dt=0.285038 dtf=0.005757 dtb=0.011762 sps=14.033209 sps_per_gpu=3.508302 tps=919680.367894 tps_per_gpu=229920.091974 mfu=45.762304 train_loss=2.439494 val_loss=2.478951
[2024-07-17 07:44:13.362148][INFO][trainer:885] - step=120 loss=2.429014 dt=0.284445 dtf=0.005222 dtb=0.011486 sps=14.062460 sps_per_gpu=3.515615 tps=921597.361532 tps_per_gpu=230399.340383 mfu=45.788661 train_loss=2.439494 val_loss=2.478951
[2024-07-17 07:44:16.210694][INFO][trainer:885] - step=130 loss=2.402059 dt=0.285559 dtf=0.005199 dtb=0.011765 sps=14.007633 sps_per_gpu=3.501908 tps=918004.211586 tps_per_gpu=229501.052897 mfu=45.794438 train_loss=2.439494 val_loss=2.478951
[2024-07-17 07:44:19.061546][INFO][trainer:885] - step=140 loss=2.374062 dt=0.285476 dtf=0.005239 dtb=0.011453 sps=14.011662 sps_per_gpu=3.502916 tps=918268.297093 tps_per_gpu=229567.074273 mfu=45.800956 train_loss=2.439494 val_loss=2.478951
[2024-07-17 07:44:21.917283][INFO][trainer:885] - step=150 loss=2.365385 dt=0.285846 dtf=0.005125 dtb=0.011320 sps=13.993568 sps_per_gpu=3.498392 tps=917082.475791 tps_per_gpu=229270.618948 mfu=45.800900 train_loss=2.439494 val_loss=2.478951
[2024-07-17 07:44:24.771924][INFO][trainer:885] - step=160 loss=2.317337 dt=0.280788 dtf=0.005173 dtb=0.011249 sps=14.245602 sps_per_gpu=3.561401 tps=933599.792506 tps_per_gpu=233399.948127 mfu=45.883340 train_loss=2.439494 val_loss=2.478951
[2024-07-17 07:44:27.626812][INFO][trainer:885] - step=170 loss=2.256231 dt=0.284973 dtf=0.005141 dtb=0.011299 sps=14.036416 sps_per_gpu=3.509104 tps=919890.544506 tps_per_gpu=229972.636126 mfu=45.889069 train_loss=2.439494 val_loss=2.478951
[2024-07-17 07:44:30.480952][INFO][trainer:885] - step=180 loss=2.216419 dt=0.286555 dtf=0.005180 dtb=0.011402 sps=13.958906 sps_per_gpu=3.489726 tps=914810.852170 tps_per_gpu=228702.713043 mfu=45.868857 train_loss=2.439494 val_loss=2.478951
[2024-07-17 07:44:33.337342][INFO][trainer:885] - step=190 loss=2.145123 dt=0.291456 dtf=0.005409 dtb=0.019347 sps=13.724205 sps_per_gpu=3.431051 tps=899429.467247 tps_per_gpu=224857.366812 mfu=45.773849 train_loss=2.439494 val_loss=2.478951
[2024-07-17 07:44:36.194584][INFO][trainer:885] - step=200 loss=2.068149 dt=0.285703 dtf=0.005153 dtb=0.011286 sps=14.000555 sps_per_gpu=3.500139 tps=917540.393411 tps_per_gpu=229385.098353 mfu=45.778791 train_loss=2.439494 val_loss=2.478951
[2024-07-17 07:44:37.224149][INFO][trainer:820] - ['prompt']: 'What is an LLM?'
[2024-07-17 07:44:37.224745][INFO][trainer:824] - ['response']:

What is an LLM?

LORTESS LA:
No, sighappat selace? don downd sourciceans note cancen up sof liond
This and my man, werame, of re thee
Thise not will I on land brond sul me a fingore?

FLER:
Tisint your not nare lame o igen,-to brorst.

SamERS:
Sin:
I\'l hell she lor hen w
[2024-07-17 07:45:14.409129][INFO][trainer:760] - Saving checkpoint to: /home/foremans/tmp/polaris-talk/outputs/runs/pytorch/DDP/2024-07-17/07-42-13
[2024-07-17 07:45:14.409820][INFO][trainer:761] - Saving model to: /home/foremans/tmp/polaris-talk/outputs/runs/pytorch/DDP/2024-07-17/07-42-13/model.pth
[2024-07-17 07:45:16.366935][INFO][configs:141] - Appending /home/foremans/tmp/polaris-talk/outputs/runs/pytorch/DDP/2024-07-17/07-42-13 to /home/foremans/tmp/polaris-talk/2024-07-17-073327/wordplay/src/ckpts/checkpoints.log
[2024-07-17 07:45:19.245061][INFO][trainer:885] - step=210 loss=1.982169 dt=0.283305 dtf=0.005223 dtb=0.011284 sps=14.119042 sps_per_gpu=3.529760 tps=925305.515083 tps_per_gpu=231326.378771 mfu=45.822019 train_loss=2.045786 val_loss=2.148510
[2024-07-17 07:45:22.092430][INFO][trainer:885] - step=220 loss=1.897731 dt=0.284759 dtf=0.005217 dtb=0.011187 sps=14.046945 sps_per_gpu=3.511736 tps=920580.608106 tps_per_gpu=230145.152026 mfu=45.837327 train_loss=2.045786 val_loss=2.148510
[2024-07-17 07:45:24.942639][INFO][trainer:885] - step=230 loss=1.817213 dt=0.285266 dtf=0.005208 dtb=0.011446 sps=14.022003 sps_per_gpu=3.505501 tps=918945.985503 tps_per_gpu=229736.496376 mfu=45.842940 train_loss=2.045786 val_loss=2.148510
[2024-07-17 07:45:27.797910][INFO][trainer:885] - step=240 loss=1.779287 dt=0.285465 dtf=0.005189 dtb=0.011220 sps=14.012250 sps_per_gpu=3.503062 tps=918306.793546 tps_per_gpu=229576.698387 mfu=45.844800 train_loss=2.045786 val_loss=2.148510
[2024-07-17 07:45:30.653597][INFO][trainer:885] - step=250 loss=1.704220 dt=0.289284 dtf=0.005471 dtb=0.010346 sps=13.827253 sps_per_gpu=3.456813 tps=906182.836379 tps_per_gpu=226545.709095 mfu=45.785926 train_loss=2.045786 val_loss=2.148510
[2024-07-17 07:45:33.512769][INFO][trainer:885] - step=260 loss=1.671318 dt=0.287679 dtf=0.005125 dtb=0.011250 sps=13.904380 sps_per_gpu=3.476095 tps=911237.442617 tps_per_gpu=227809.360654 mfu=45.758182 train_loss=2.045786 val_loss=2.148510
[2024-07-17 07:45:36.373461][INFO][trainer:885] - step=270 loss=1.650952 dt=0.298661 dtf=0.005118 dtb=0.011520 sps=13.393107 sps_per_gpu=3.348277 tps=877730.651421 tps_per_gpu=219432.662855 mfu=45.565875 train_loss=2.045786 val_loss=2.148510
[2024-07-17 07:45:39.236930][INFO][trainer:885] - step=280 loss=1.573242 dt=0.285970 dtf=0.005171 dtb=0.011290 sps=13.987477 sps_per_gpu=3.496869 tps=916683.279847 tps_per_gpu=229170.819962 mfu=45.587333 train_loss=2.045786 val_loss=2.148510
[2024-07-17 07:45:42.100605][INFO][trainer:885] - step=290 loss=1.533265 dt=0.286487 dtf=0.005432 dtb=0.011288 sps=13.962259 sps_per_gpu=3.490565 tps=915030.617828 tps_per_gpu=228757.654457 mfu=45.598392 train_loss=2.045786 val_loss=2.148510
[2024-07-17 07:45:44.964424][INFO][trainer:885] - step=300 loss=1.492064 dt=0.288480 dtf=0.005355 dtb=0.011480 sps=13.865774 sps_per_gpu=3.466443 tps=908707.340870 tps_per_gpu=227176.835218 mfu=45.576766 train_loss=2.045786 val_loss=2.148510
[2024-07-17 07:45:45.995833][INFO][trainer:820] - ['prompt']: 'What is an LLM?'
[2024-07-17 07:45:45.996497][INFO][trainer:824] - ['response']:

What is an LLM?

RICHMORD:
Char stire? how in those are name the range hone.

GLOUCESTER:
Nay, in lond's time the palt are worder more
That wilt in the purpose be a pey
And thou thine onter hands, and the which broth.

ELBOWINCA:
At lie my lord with the me an arms be a s
[2024-07-17 07:46:23.549987][INFO][trainer:760] - Saving checkpoint to: /home/foremans/tmp/polaris-talk/outputs/runs/pytorch/DDP/2024-07-17/07-42-13
[2024-07-17 07:46:23.550696][INFO][trainer:761] - Saving model to: /home/foremans/tmp/polaris-talk/outputs/runs/pytorch/DDP/2024-07-17/07-42-13/model.pth
[2024-07-17 07:46:25.496559][INFO][configs:141] - Appending /home/foremans/tmp/polaris-talk/outputs/runs/pytorch/DDP/2024-07-17/07-42-13 to /home/foremans/tmp/polaris-talk/2024-07-17-073327/wordplay/src/ckpts/checkpoints.log
[2024-07-17 07:46:28.374854][INFO][trainer:885] - step=310 loss=1.444200 dt=0.299907 dtf=0.005333 dtb=0.010637 sps=13.337481 sps_per_gpu=3.334370 tps=874085.133345 tps_per_gpu=218521.283336 mfu=45.384395 train_loss=1.495372 val_loss=1.713714
[2024-07-17 07:46:31.223079][INFO][trainer:885] - step=320 loss=1.429350 dt=0.285238 dtf=0.005245 dtb=0.011485 sps=14.023353 sps_per_gpu=3.505838 tps=919034.479880 tps_per_gpu=229758.619970 mfu=45.435743 train_loss=1.495372 val_loss=1.713714
[2024-07-17 07:46:34.074957][INFO][trainer:885] - step=330 loss=1.362220 dt=0.285027 dtf=0.005165 dtb=0.011407 sps=14.033736 sps_per_gpu=3.508434 tps=919714.904826 tps_per_gpu=229928.726207 mfu=45.485355 train_loss=1.495372 val_loss=1.713714
[2024-07-17 07:46:36.929464][INFO][trainer:885] - step=340 loss=1.350888 dt=0.284436 dtf=0.005199 dtb=0.011287 sps=14.062893 sps_per_gpu=3.515723 tps=921625.744709 tps_per_gpu=230406.436177 mfu=45.539549 train_loss=1.495372 val_loss=1.713714

wordplay 🎮💬

• Link¹ to video

Example: Training a LLM to talk like Shakespeare using saforem2/wordplay 🎮💬

1. idk why it doesn’t render correctly in the slide (seems like refreshing helps?)

Thank you!

• Organizers

• ALCF Data Science & Operations

• Feel free to reach out!

🙏 Acknowledgements

This research used resources of the Argonne Leadership Computing Facility, which is a DOE Office of Science User Facility supported under Contract DE-AC02-06CH11357.

🎁 Extras

DeepSpeed ZeRO Offloading

ZeRO Offloading Stages

Figure 25

DeepSpeed ZeRO Offloading

(a) DeepSpeed + ZeRO

Figure 26

Hands-On

• Instructions

  • saforem2/wordplay 🎮💬 [web]

References

• 🔗 See also:
  • PyTorch Distributed Overview
  • Distributed Data Parallel — PyTorch master documentation
  • 🤗 Efficient Training on Multiple GPUs
  • Getting Started - DeepSpeed
• See my slides on:
  • Parallel Training Techniques for additional details
  • saforem2/llm-lunch-talk (slides)

Bibliography

Wei, Jason, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, et al. 2022. “Emergent Abilities of Large Language Models.” https://arxiv.org/abs/2206.07682.

Yang, Jingfeng, Hongye Jin, Ruixiang Tang, Xiaotian Han, Qizhang Feng, Haoming Jiang, Bing Yin, and Xia Hu. 2023. “Harnessing the Power of LLMs in Practice: A Survey on ChatGPT and Beyond.” https://arxiv.org/abs/2304.13712.

Yao, Shunyu, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. 2023. “Tree of Thoughts: Deliberate Problem Solving with Large Language Models.” https://arxiv.org/abs/2305.10601.