Incorrect EndToEnd total_time in SimAI_analytical results
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 1.2k
- Forks
- 184
- Avg merge
- 13h 4m
- Merged PRs (30d)
- 1
Description
1. Environment Info
- OS: [ Ubuntu 22.04.4 LTS]
- SimAI Commit: [e5d1251]
- Python Version: [Python 3.11.12]
2.Steps to Reproduce
Step 1: Generate Workload
- Working Directory:
/root/simai/SimAI/aicb/ - Command Executed:
uv run -m workload_generator.SimAI_training_workload_generator \ --frame=Megatron \ --world_size=1472 \ --tensor_model_parallel_size=8 \ --pipeline_model_parallel=8 \ --global_batch=32 \ --micro_batch=1 \ --epoch_num=1 \ --model_name=gpt_13B \ --hidden_size=5120 \ --num_layers=40 \ --seq_length=4096 \ --num_attention_heads=40 \ --vocab_size=50257 \ --max_position_embeddings=4096 \ --ffn_hidden_size=11008 \ --dtype=bfloat16 \ --enable_sequence_parallel \ --swiglu \ --make_vocab_size_divisible_by=128 \ --workload_only \ --output_filename=two_phase_opt_phase1/ws1472-tp8-pp8 \ --aiob_enable \ --comp_filepath=/root/simai/SimAI/aicb/workload/aiob_inputs/A100_A800.txt - Generated Workload: ws1472-tp8-pp8.txt
Step 2: Run Analytical Simulation
- Working Directory:
/root/simai/SimAI/ - Command Executed:
/root/simai/SimAI/bin/SimAI_analytical \ -w /root/simai/SimAI/aicb/results/workload/two_phase_opt_phase1/ws1472-tp8-pp8.txt \ -g 1472 \ -g_p_s 16 \ -n_p_s 5 \ -r two_phase_opt_phase1/ws1472-tp8-pp8 \ -g_type A800 \ -nic 37.47811737589823 \ -dp_o 0.5 \ -tp_o 0.7 \ -ep_o 0.8 \ -pp_o 0.5 - Result File: ws1472-tp8-pp8EndToEnd.csv
3. The Issue
In the generated EndToEnd.csv file, the Expose DP comm time is exceptionally large,This occurs consistently in many experiments
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Reproduce the workload generation and analytical simulation using the commands in the issue, then inspect the generated ws1472-tp8-pp8EndToEnd.csv file. Focus on the exceptionally large Expose DP comm value and compare it with the provided workload and results from other experiments. Done means the EndToEnd total_time reports a validated Expose DP comm value.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- performance
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 30/100