First series of P0 patterns to optimize llama

[xadupre]

No description provided.

[codecov]

Codecov Report

Attention: Patch coverage is 88.77551% with 11 lines in your changes are missing coverage. Please review.

Project coverage is 77.60%. Comparing base (34e410a) to head (36eb15b).

Files	Patch %	Lines
onnxscript/rewriter/llama_rule_sets.py	86.36%	2 Missing and 4 partials ⚠️
onnxscript/rewriter/llama_rule_sets_test.py	90.74%	2 Missing and 3 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1490      +/-   ##
==========================================
+ Coverage   77.56%   77.60%   +0.03%     
==========================================
  Files         216      218       +2     
  Lines       23339    23437      +98     
  Branches     4025     4046      +21     
==========================================
+ Hits        18103    18188      +85     
- Misses       4453     4459       +6     
- Partials      783      790       +7     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[xadupre]

def example2(X: FLOAT[2, 1024]) -> FLOAT[2, 1024]:, the pattern works for any type, any 2D shape, so I would prefer to avoid to be that precise.

[xadupre]

I tried to follow the recommandations. I get the following error:

FAILED onnxscript/rewriter/llama_rule_sets_tests.py::LlamaRuleSetsTest::test_llama_p0_rule_set_identity - TypeError: Cannot convert <class 'onnxscript.ir._core.Input'> to ValuePattern
FAILED onnxscript/rewriter/llama_rule_sets_tests.py::LlamaRuleSetsTest::test_llama_p0_rule_set_transpose_transpose - TypeError: Cannot convert <class 'onnxscript.ir._core.Input'> to ValuePattern

CI is passing...

[justinchuby]

lgtm with minor comments. Thanks!