clab / clab/dynet

ComputationGraph::clear() not working properly -> Crash in exec.cc line 980

Open
#1,612 2 comments 0 reactions 0 assignees View on GitHub
Dominant language
C++
Stars
3.4k
Forks
701
PR merge metrics
No merged PRs in 30d

Description

Hello,
I've run into a crash which I need help to track it down.
I'm running a Qt (5.12) application using dynet with MKL (2020.0.166) on ubuntu 18 with auto-batching enabled.
I get a reproducible crash in the above mentioned line, but only with a release build. I tried to debug into it but in this case all relevant information is optimized. And with a debug build it doesn't occur.
But I could definitely find out that in this line
```size_t final_size = batched_ndEdfs.back().d.size();```
the container batched_ndEdfs in some rare cases (once in about 50000 cycles) is empty.
Could you help me to track it down or maybe already have a clue?

Contributor guide

No contributing guide indexed for this repository

Research direction

Start at exec.cc line 980 and trace where batched_ndEdfs is populated before batched_ndEdfs.back() is called. Compare release and debug behavior with auto-batching enabled, using the reported empty-container case as the failure condition. Done means identifying why the container can be empty and verifying a correction against the reproducible crash.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
backend, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.