Conversation
Signed-off-by: Ao Tang <aot@nvidia.com>
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
| if self.entire_gpu: | ||
| self.gpus = 1.0 |
There was a problem hiding this comment.
logic: When entire_gpu=True, self.gpus is set to 1.0 after validation checks. If user also sets gpu_memory_gb > 0, the validation on line 52 will fail before reaching this code, which is correct. However, if user sets both entire_gpu=True and gpus > 0, this assignment will silently overwrite their explicit gpus value without raising an error. Should there be a validation check to prevent setting both entire_gpu=True and gpus > 0 explicitly?
There was a problem hiding this comment.
I honestly think the best is to drop entire_gpu and only provide gpu_memory_gb and gpus. @ayushdg @praateekmahajan what do you think
There was a problem hiding this comment.
I agree with this, entire_gpu has been confusing for me the whole time it existed. I think it was there for nvenc/nvdec reasons (which I never understood well). So if we're nuking nvenc/nvdec support this should be good to go
There was a problem hiding this comment.
Agreed let's just drop this
There was a problem hiding this comment.
I will make a seperate issue for this to track. Let's keep it seperate. Wanna make sure we do it properly across documentation and code and stuff
|
/ok to test d9f705a |
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> Signed-off-by: Abhinav Garg <abhinavg@stanford.edu>
|
/ok to test d9834e6 |
There was a problem hiding this comment.
Additional Comments (1)
-
docs/about/concepts/video/abstractions.md, line 78-80 (link)syntax: Documentation still references
entire_gpugiving access to NVDEC/NVENC and mentionsnvdecs/nvencsfields which have been removed in this PR
7 files reviewed, 2 comments
|
/ok to test 73d9dd3 |
There was a problem hiding this comment.
Additional Comments (1)
-
nemo_curator/__init__.py, line 34 (link)logic: import path needs updating to match the new cosmos-xenna 0.1.8 structure -
cosmos_xenna.ray_utils.clusterwas moved. Check if this should becosmos_xenna.pipelines.private.clusteror ifAPI_LIMITis now in a different module
7 files reviewed, 2 comments
| if self.entire_gpu: | ||
| self.gpus = 1.0 |
| if self.entire_gpu: | ||
| self.gpus = 1.0 |
There was a problem hiding this comment.
Agreed let's just drop this
Signed-off-by: Abhinav Garg <abhinavg@stanford.edu>
|
/ok to test ec9a212 |
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> Signed-off-by: Abhinav Garg <abhinavg@stanford.edu>
|
/ok to test 84fb624 |
| Attributes: | ||
| cpus: Number of CPU cores required | ||
| gpu_memory_gb: GPU memory required in GB (Only for single-GPU stages) | ||
| nvdecs: Number of NVDEC units required | ||
| nvencs: Number of NVENC units required | ||
| entire_gpu: Whether to allocate entire GPU regardless of memory (This also gives you nvdecs and nvencs of that GPU) | ||
| entire_gpu: Whether to allocate entire GPU regardless of memory | ||
| gpus: Number of GPUs required (Only for multi-GPU stages) | ||
| """ |
There was a problem hiding this comment.
style: Check that documentation is updated. Several doc files still reference removed nvdecs/nvencs fields:
docs/curate-video/tutorials/pipeline-customization/add-cust-stage.mddocs/curate-video/tutorials/pipeline-customization/add-cust-model.mddocs/curate-video/tutorials/pipeline-customization/add-cust-env.mddocs/curate-video/index.mddocs/about/concepts/video/abstractions.md
There was a problem hiding this comment.
@suiyoubi @praateekmahajan Can you take a look at this? Not addressing this in this PR.
|
/ok to test 84fb624 |
|
/ok to test 85814d2 |
Greptile's behavior is changing!From now on, if a review finishes with no comments, we will not post an additional "statistics" comment to confirm that our review found nothing to comment on. However, you can confirm that we reviewed your changes in the status check section. This feature can be toggled off in your Code Review Settings by deselecting "Create a status check for each PR". |
5 participants
Description
Linked issue: #1201
For Video Transcoding stage, use similar logic as https://github.com/nvidia-cosmos/cosmos-curate/blob/main/cosmos_curate/pipelines/video/clipping/clip_extraction_stages.py#L198 to partition gpu based on _nb_streams_per_gpu.
Usage
# Add snippet demonstrating usageChecklist