NVIDIA / NVIDIA/TensorRT-LLM

tiny llama 1.1B test shows extra memory consumption after loading the weights and before compilation

Open
#6,335 0 comments 0 reactions 1 assignee View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
14.7k
Forks
2.8k
Avg merge
2d 23h
Merged PRs (30d)
489

Description

In this new unit test - test_trtllm_bench_backend_comparison , when commenting out the num_hidden_layers (set to 10 to shorten the test duration), I see that the consumed memory is
Found free_mem_pre_mb: 76943
Total GPU memory: 81007MB

the model params should consume 2.2GB but we see 4GB are consumed.

pytest -s -v tests/unittest/_torch/auto_deploy/unit/singlegpu/test_ad_trtllm_bench.py::test_trtllm_bench_backend_comparison

MR: https://github.com/nv-auto-deploy/TensorRT-LLM/pull/102

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.