lucylq

[executorch] Preprocess export test

3

Add python e2e test exporting preprocess. On top of: https://github.com/pytorch/executorch/pull/4603 ``` python -m unittest examples/models/flamingo/test_preprocess.py ```

Update torch checks for preprocess

2

#### Context What is the purpose of this PR? Is it to - [x] add a new feature - [ ] fix a bug - [ ] update tests and/or...

Use generated files in module test

4

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #10497 Instead of checking in the PTE/PTD files. This helps us keep up to date with aot changes, especially with incoming flat_tensor...

Update flat tensor ndm to account for named delegate data

3

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #10330 * #10328 Currently flat_tensor ndm only accounts for tensors in get_data, get_num_keys, get_key functions. Add support to return named_data values as...

Introduce PTEFile class

4

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #15801 * __->__ #15800 PTEFile class holds the components of a PTE file: the program, mutable constants and named data. Currently, the `program`...

[llm_patch] Fix out-of-bounds access in pad2d function

2

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #15832 https://github.com/pytorch/executorch/pull/15623 Add checks for pad1d and pad3d, as well as pad2d. I think what happens is, we 'broadcast' the input tensor...

Use PTEFile class in serialize_pte_binary

2

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #15801 * #15800 Take in a PTE file for serialization instead of program, mutable_segments, named_data segments separately. Differential Revision: [D86908241](https://our.internmc.facebook.com/intern/diff/D86908241/)

[llm_patch] Fix out-of-bounds access in pad2d function

2

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #15623 Add checks for pad1d and pad3d, as well as pad2d. I think what happens is, we 'broadcast' the input tensor into...

Do not restore constant_buffer, and use bytes instead of Buffer

2

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #15802 * #15801 * #15800 All constants are serialized in the segment (none in the Program). This PR: 1. Places constant data...

Deserialize segments

1

Currently, named data segments and mutable segments are lost when we deserialize a PTE file.