Added llama3.1-70b Benchmarking recipe on A3-Mega nodes#246
Draft
krishnakanthankam-qt wants to merge 8 commits into
Draft
Added llama3.1-70b Benchmarking recipe on A3-Mega nodes#246krishnakanthankam-qt wants to merge 8 commits into
krishnakanthankam-qt wants to merge 8 commits into
Conversation
|
Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA). View this failed invocation of the CLA check for more information. For the most up to date status, view the checks section at the bottom of the pull request. |
depksingh
reviewed
Jun 12, 2026
|
|
||
| This recipe supports the following models. Running TRTLLM inference benchmarking on these models are only tested and validated on A3-Mega GKE nodes with certain combination of TP, PP, EP, number of GPU chips, input & output sequence length, precision, etc. | ||
|
|
||
| Example model configuration YAML files included in this repo only show a certain combination of parallelism hyperparameters and configs for benchmarking purposes. Input and output length in `/home/akrishnakanth/gpu-recipes/inference/a3mega/llama3.1-70b/trtllm-gke/values.yaml` need to be adjusted according to the model and its configs. |
depksingh
reviewed
Jun 12, 2026
| rm -rf $engine_dir | ||
| rm -f $dataset_file | ||
| rm -rf $engine_dir || true | ||
| rm -f $dataset_file || true |
Priya-Quad
reviewed
Jun 16, 2026
| --backend "pytorch" \ | ||
| --kv_cache_free_gpu_mem_fraction $kv_cache_free_gpu_mem_fraction \ | ||
| $extra_args $vl_args > $output_file | ||
| $extra_args $vl_args | tee "$output_file" |
Contributor
There was a problem hiding this comment.
| tee - This change can be reverted back to orginal.
Priya-Quad
reviewed
Jun 16, 2026
| --dataset $dataset_file \ | ||
| --engine_dir $engine_dir \ | ||
| --kv_cache_free_gpu_mem_fraction $kv_cache_free_gpu_mem_fraction $extra_args >$output_file | ||
| --kv_cache_free_gpu_mem_fraction $kv_cache_free_gpu_mem_fraction $extra_args | tee $output_file |
Contributor
There was a problem hiding this comment.
| tee - This also can revert back to original.
Priya-Quad
reviewed
Jun 16, 2026
| --kv_cache_free_gpu_mem_fraction $kv_cache_free_gpu_mem_fraction $extra_args | tee $output_file | ||
| fi | ||
|
|
||
| cat $output_file |
Contributor
There was a problem hiding this comment.
add this back to the file.
Priya-Quad
reviewed
Jun 16, 2026
| serverArgs: | ||
| max-model-len: 32768 | ||
| max-num-seqs: 128 | ||
| gpu-memory-utilization: 0.90 No newline at end of file |
Contributor
There was a problem hiding this comment.
please remove gpu-memory-utilization: 0.90 from here, you are passing this value from trtllm-configs
Priya-Quad
reviewed
Jun 16, 2026
| helm install -f values.yaml \ | ||
| --set workload.benchmarks.experiments[0].isl=128 \ | ||
| --set workload.benchmarks.experiments[0].osl=128 \ | ||
| --set workload.benchmarks.experiments[0].num_requests=1000 \ |
Contributor
There was a problem hiding this comment.
Line 235 to 237 can be removed, we are passing these values from values.yaml. we don't usually hardcode any values on Readme
Priya-Quad
reviewed
Jun 16, 2026
| $REPO_ROOT/src/helm-charts/a3mega/trtllm-inference/single-node | ||
| ``` | ||
| > [!NOTE] | ||
| > You can modify the benchmark configuration at runtime by changing the values for `isl`, `osl`, and `num_requests` (number of prompts) in the Helm command to test different scenarios. |
Contributor
There was a problem hiding this comment.
Please check the other recipes to update this line.
Priya-Quad
reviewed
Jun 16, 2026
| =========================================================== | ||
| DATASET DETAILS | ||
| =========================================================== | ||
| Dataset Path: /ssd/token-norm-dist_llama3.1-70b_128_128_tp4.json |
Priya-Quad
reviewed
Jun 16, 2026
| PYTORCH BACKEND | ||
| =========================================================== | ||
| Model: nvidia/Llama3.1-70b | ||
| Model Path: /ssd/nvidia/Llama3.1-70b |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Title
Add Llama 3.1 70B Recipe and Optimized Sequential Benchmarking
Summary
Introduces a high-performance recipe for serving and benchmarking Llama 3.1 70B on A3mega GKE node pools.