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Outdated suggestions cannot be applied.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.
Work-in-progress PR to add the Phi-3 model as discussed in #21.
I am running a Grayskull e75 on Ubuntu 20.04 and the output of
tt-smi -s
can be found here.Currently the pipeline runs but the output is not correct in part due to missing
sin
,cos
, andabs
operations ex.2024-07-21 17:48:48.253 | WARNING | Model - OpType 'abs' not found in params. Defaulting to nop params.
. There was also an issue where TVM could not compiletorch.bool
for which I implemented a workaround.The
sin
andcos
operations are called in the rotary embedding implementations. I tested running without these calls totorch.Tensor.sin
andtorch.Tensor.cos
in the Phi-3 model file and the warnings were not emitted. Here are the calls that trigger these missing operations:The
abs
operation was not as easy to isolate but I suspect it is lower down the stack callingtorch.Tensor.abs
in a similar way.Here is the terminal output for each of my different runs:
torch.bool
ribbon
strategyAt this point I will raise an issue in the
tt-buda
repository to try and resolve the missing operations before continuing.