GitLab

To improve benchmark debugging, the following debug methods were added:

pytorch profiler in model benchmark

- SB_ENABLE_PYTORCH_PROFILER: switch to enable/disable
- SB_TORCH_PROFILER_TRACE_DIR: log path
These 2 runtime variables need to be configured in SB config file.

nsys in SB runner

- SB_ENABLE_NSYS: switch to enable/disable 
- SB_NSYS_TRACE_DIR: log path
These 2 runtime variables need to be configured in runner's ENV

---------
Co-authored-by: Hongtao Zhang <hongtaozhang@microsoft.com>

**Description**
Fix executor for Benchmark Execution Without Explicit Framework Field

Add 10-hpcx.sh to /etc/profile.d
Update the Docker exec command to ensure a persistent HPCX environment.

Update `omegaconf` version to
[2.3.0](https://pypi.org/project/omegaconf/2.3.0/) as omegaconf 2.0.6
has a non-standard dependency specifier PyYAML>=5.1.*. pip 24.1 will
enforce this behaviour change.
Discussion can be found at https://github.com/pypa/pip/issues/12063.

This PR has following changes
- torch.distributed.launch changed to torchrun. torch.distributed.launch
is deprecated in latest Pytorch and is recommended to move to torchrun -
https://pytorch.org/docs/stable/elastic/run.html


- Changes to AMD GPU detection logic. The AMD GPU detection logic throws
warning when containers have only renderD in /dev/dri, this change would
resolve those warnings

---------
Co-authored-by: Yuting Jiang <yutingjiang@microsoft.com>

**Description**
Add runner for sys info to automatically collect on multiple nodes and
update related docs.

**Major Revision**
- add runner for sys info which will check docker status and run `sb
node info` on all nodes' docker and fetch results from all nodes

**Minor Revision**
- update cli and system-info doc
- update sb node info to save output info output-dir/sys-info.json

Add signal handler in runner to gracefully exit when receiving SIGINT
(<kbd>Ctrl</kbd>+<kbd>C</kbd>) or SIGTERM during benchmark execution.

**Description**

Cherry-pick bug fixes from v0.8.0 to main.

**Major Revisions**

* Monitor - Fix the cgroup version checking logic (#502)
* Benchmark - Fix matrix size overflow issue in cuBLASLt GEMM (#503)
* Fix wrong torch usage in communication wrapper for Distributed
Inference Benchmark (#505)
* Analyzer: Fix bug in python3.8 due to pandas api change (#504)
* Bug - Fix bug to get metric from cmd when error happens (#506)
* Monitor - Collect realtime GPU power when benchmarking (#507)
* Add num_workers argument in model benchmark (#511)
* Remove unreachable condition when write host list (#512)
* Update cuda11.8 image to cuda12.1 based on nvcr23.03 (#513)
* Doc - Fix wrong unit of cpu-memory-bw-latency in doc (#515)
* Docs - Upgrade version and release note (#508)
Co-authored-by: guoshzhao <guzhao@microsoft.com>
Co-authored-by: Ziyue Yang <ziyyang@microsoft.com>
Co-authored-by: Yuting Jiang <yutingjiang@microsoft.com>

**Description**
Support SuperBench Executor running on Windows.

**Major Revision**
- Lazy import ansible related module

**Description**

Cherry-pick bug fixes from v0.7.0 to main.

**Major Revisions**

* Benchmarks - Fix missing include in FP8 benchmark (#460)
* Fix bug in TE BERT model (#461)
* Doc - Update benchmark doc (#465)
* Bug: Fix bug for incorrect datatype judgement in cublas-function
source code (#464)
* Support `sb deploy` without pulling image (#466)
* Docs - Upgrade version and release note (#467)
Co-authored-by: Russell J. Hewett <russell.j.hewett@gmail.com>
Co-authored-by: Yuting Jiang <yutingjiang@microsoft.com>

**Major Revision**

- Add an option for pattern to generate mpi_pattern.txt file if
specified the path.
- In mpi pattern, serial_index and parallel_index will add in each
benchmark as environment variables.

**Minor Revision**
- Fix typo

**Description**
Support the following patterns  in `mpi` mode:
* `k-batch`
* `topo-aware`

* Extract pair-wise pattern from ib_validation

* add mpi-parallels mode

* update according to comments

* fix and update doc

* update

* merge into 'mpi' mode

* udpate according to comments

* fix testcases

* fix ansible

* regard pattern as field

* udpate

* fix flake8 version

* add flake8 range

* remove map-by from host config

* udpate comments

Add non-zero return code for `sb deploy` and `sb run` command when
there're Ansible failures in control plane.
Return code is set to count of failure.

For failures caused by benchmarks, return code is still set per benchmark
in results json file.

**Description**

Cherry-pick bug fixes from v0.6.0 to main.

**Major Revisions**

* Enable latency test in ib traffic validation distributed benchmark (#396)
* Enhance parameter parsing to allow spaces in value (#397)
* Update apt packages in dockerfile (#398)
* Upgrade colorlog for NO_COLOR support (#404)
* Analyzer - Update error handling to support exit code of sb result diagnosis (#403)
* Analyzer - Make baseline file optional in data diagnosis and fix bugs (#399)
* Enhance timeout cleanup to avoid possible hanging (#405)
* Auto generate ibstat file by pssh (#402)
* Analyzer - Format int type and unify empty value to N/A in diagnosis output file (#406)
* Docs - Upgrade version and release note (#407)
* Docs - Fix issues in document (#408)
Co-authored-by: Yang Wang <yangwang1@microsoft.com>
Co-authored-by: Yuting Jiang <yutingjiang@microsoft.com>

Fix minimum timeout: use 60s if config is shorter.

* Gracefully exit when timeout, add corresponding log and return code.
* Set minimum timeout to 1 minute and enlarge Ansible timeout.

Fix several issues in ib validation benchmark:
* continue running when timeout in the middle, instead of aborting whole mpi process
* make timeout parameter configurable, set default to 120 seconds
* avoid mixture of stdio and iostream when print to stdout
* set default message size to 8M which will saturate ib in most cases
* fix hostfile path issue so that it can be auto found in different cases

Support `node_num: 1` in mpi mode, so that we can run mpi benchmarks in
both 1 node and all nodes in one config by changing `node_num`.
Update docs and add test case accordingly.

**Description**

Support multiple IB/GPU devices run simultaneously in ib validation benchmark.

**Major Revisions**
- Revise ib_validation_performance.cc so that multiple processes per node could be used to launch multiple perftest commands simultaneously. For each node pair in the config, number of processes per node will run in parallel.
- Revise ib_validation_performance.py to correct file paths and adjust parameters to specify different NICs/GPUs/NUMA nodes.
- Fix env issues in Dockerfile for end-to-end test.
- Update ib-traffic configuration examples in config files.
- Update unit tests and docs accordingly.

Closes #326.

Fix sudo issue when running without Docker, user account could be
arbitrary in such case.

**Description**

Support `sb run` on host directly without Docker

**Major Revisions**
- Add `--no-docker` argument for `sb run`.
- Run on host directly if `--no-docker` if specified.
- Update docs and tests correspondingly.

Fix env file path to absolute path in `docker exec`, in case there're mixed ssh and local connections or different users are used.

Fix env file path for `docker run`.

Support customized env for all modes in configuration.

**Description**
Please write a brief description and link the related issue if have.

**Major Revision**
- Sync (do allreduce max) the E2E training results among all workers.
- Avoid using ':0' in metric name if there has only one rank having output.

**Description**
Add timeout feature for each benchmark.

**Major Revision**
- Add `timeout` config for each benchmark. In current config files, only set the timeout for kernel-launch as example. Other benchmarks can be set in the future.
- Set the timeout config for `ansible_runner.run()`. Runner will get the return code 254:
   [ansible.py:80][WARNING] Run failed, return code 254.
- Using `timeout` command to terminate the client process.

__Description__

Update benchmark naming to support annotations.

__Major Revisions__
- Update name for `create_benchmark_context` in executor.
- Backward compatibility for model benchmarks using "_models" suffix.
- Update documents.

__Description__

Cherry-pick  bug fixes from v0.4.0 to main.

__Major Revisions__

* Bug - Fix issues for Ansible and benchmarks (#267)
* Tests - Refine test cases for microbenchmark (#268)
* Bug - Build openmpi with ucx support in rocm dockerfiles (#269)
* Benchmarks: Fix Bug - Fix fio build issue (#272)
* Docs - Unify metric and add doc for cublas and cudnn functions (#271)
* Monitor: Revision - Add 'monitor/' prefix to monitor metrics in result summary (#274)
* Bug - Fix bug of detecting if gpu_index is none (#275)
* Bug - Fix bugs in data diagnosis (#273)
* Bug - Fix issue that the root mpi rank may not be the first in the hostfile (#270)
* Benchmarks: Configuration - Update inference and network benchmarks in configs (#276)
* Docs - Upgrade version and release note (#277)
Co-authored-by: Yuting Jiang <v-yutjiang@microsoft.com>

**Description**
Integrate monitor into Superbench.

**Major Revision**
- Initialize, start and stop monitor in SB executor.
- Parse the monitor data in SB runner and merge into benchmark results.
- Specify ReduceType for monitor metrics, such as MAX, MIN and LAST.
- Add monitor configs into config file.

Fix issues for distributed runs:
* fix config for memory bandwidth benchmarks
* add throttling for high concurrency docker pull
* update rsync path and exclude directories
* handle exceptions when creating summary
* tune for logging

**Description**

Cherry-pick  bug fixes from v0.3.0 to main.

**Major Revisions**
* Docs - Upgrade version and release note (#209)
* Benchmarks: Build Pipeline - Update rccl-test git submodule to dc1ad48 (#210)
* Benchmarks: Update - Update benchmarks in configuration file (#208)
* CI/CD - Update GitHub Action VM (#211)
* Benchmarks: Fix Bug - Fix wrong parameters for gpu-sm-copy-bw in configuration examples (#203)
* CI/CD - Fix bug in build image for push event (#205)
* Benchmark: Fix Bug - fix error message of communication-computation-overlap (#204)
* Tool: Fix bug - Fix function naming issue in system info  (#200)
* CI/CD - Push images in GitHub Action (#202)
* Bug - Fix torch.distributed command for single node (#201)
* CLI - Integrate system info for node (#199)
* Benchmarks: Code Revision - Revise CMake files for microbenchmarks. (#196)
* CI/CD - Add ROCm image build in GitHub Actions (#194)
* Bug: Fix bug - fix bug of hipBusBandwidth build (#193)
* Benchmarks: Build Pipeline - Restore rocblas build logic (#197)
* Bug: Fix Bug - Add barrier before 'destroy_process_group' in model benchmarks (#198)
* Bug - Revise 'docker run' in sb deploy (#195)
* Bug - Fix Bug : fix bug of error param operations to operation in rccl-bw of hpe config (#190)
Co-authored-by: Yuting Jiang <v-yujiang@microsoft.com>
Co-authored-by: Guoshuai Zhao <guzhao@microsoft.com>
Co-authored-by: Ziyue Yang <ziyyang@microsoft.com>

**Description**
Generate the summarized output files from all nodes. For each metric, do the reduce operation according to the `reduce_op`

**Major Revision**
- Generate the summarized json file per node:
For microbenchmark, the format is `{benchmark_name}/[{run_count}/]{metric_name}[:rank]`
For modelbenchmark, the format is `{benchmark_name}/{sub_benchmark_name}/[{run_count}/]{metric_name}`
`[]` means optional.
```
{
  "kernel-launch/overhead_event:0": 0.00583,
  "kernel-launch/overhead_event:1": 0.00545,
  "kernel-launch/overhead_event:2": 0.00581,
  "kernel-launch/overhead_event:3": 0.00572,
  "kernel-launch/overhead_event:4": 0.00559,
  "kernel-launch/overhead_event:5": 0.00591,
  "kernel-launch/overhead_event:6": 0.00562,
  "kernel-launch/overhead_event:7": 0.00586,
  "resnet_models/pytorch-resnet50/steptime-train-float32": 544.0827468410134,
  "resnet_models/pytorch-resnet50/throughput-train-float32": 353.7607016465773,
  "resnet_models/pytorch-resnet50/steptime-train-float16": 425.40482617914677,
  "resnet_models/pytorch-resnet50/throughput-train-float16": 454.0142363793973,
  "pytorch-sharding-matmul/0/allreduce": 10.561786651611328,
  "pytorch-sharding-matmul/1/allreduce": 10.561786651611328,
  "pytorch-sharding-matmul/0/allgather": 10.088025093078613,
  "pytorch-sharding-matmul/1/allgather": 10.088025093078613
}
```
- Generate the summarized jsonl file for all nodes, each line is the result from one node in json format.

Support mpi mode in runner:
* concate mpirun command
* support mca and env config
* prepare hostfile and update Ansible host pattern
Co-authored-by: Peng Cheng <chengpeng5555@outlook.com>

Support both NVIDIA and AMD GPU and check GPU vendor during deployment and execution.

* Add GPU environment check in sb deploy.
* Check GPU vendor in executor.

Fetch benchmarks results on all nodes, will rsync after each benchmark.
The results directory structure on control node is as follows:

```
outputs/
└── datetime
    ├── nodes
    │   └── node-0
    │       ├── benchmarks
    │       │   ├── benchmark-0
    │       │   │   ├── rank-0
    │       │   │   │   └── results.json
    │       └── sb-exec.log
    ├── sb-run.log
    └── sb.config.yaml
```

* Support custom output directory.
* Update document.

* Add `sb deploy` command content.
* Fix inline if-expression syntax in playbook.
* Fix quote escape issue in bash command.
* Add custom env in config.
* Update default config for multi GPU benchmarks.
* Update MANIFEST.in to include jinja2 template.
* Require jinja2 minimum version.
* Fix occasional duplicate output in Ansible runner.
* Fix mixed color from Ansible and Python colorlog.
* Update according to comments.
* Change superbench.env from list to dict in config file.