pytorch / pytorch/TensorRT

🐛 [Bug] Llama-2-7b on a 4090 GPU

Open
#2,836 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

bug hf story: LLM & Generative AI
Dominant language
Python
Stars
3k
Forks
410
Avg merge
3d 18h
Merged PRs (30d)
78

Description

Bug Description

Currently, Torch-TRT displays the following error when compiling Llama-2-7B in FP16 on a 4090 GPU:

[05/08/2024-20:47:56] [TRT] [E] 1: [defaultAllocator.cpp::allocate::19] Error Code 1: Cuda Runtime (out of memory)
[05/08/2024-20:47:56] [TRT] [W] Requested amount of GPU memory (13476831488 bytes) could not be allocated. There may not be enough free memory for allocation to succeed.
[05/08/2024-20:47:56] [TRT] [E] 1: [graphContext.h::~MyelinGraphContext::55] Error Code 1: Myelin (No Myelin Error exists)

The model should successfully compile on a 4090 GPU, given the memory available.

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.

Research direction

Start by reproducing the FP16 Llama-2-7B compilation on a 4090 GPU and capture the exact model, Torch-TRT, PyTorch, and TensorRT configuration. Trace the reported allocator and Myelin out-of-memory failure, then verify that compilation succeeds without exceeding the GPU's available memory.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
compilers, machine-learning
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.