hawkinsp
Follow
hawkinsp
Follow
Highlights
- 30 discussions answered
Popular repositories
-
-
tensorflow Public
Forked from tensorflow/tensorflow
Computation using data flow graphs for scalable machine learning
C++
-
-
-
1,102 contributions in the last year
Contribution activity
June 2022
Created 26 commits in 3 repositories
Created a pull request in google/jax that received 1 comment
Opened 15 other pull requests in 1 repository
google/jax
15
merged
- Add instructions to override the TF (XLA) tree in the developer docum…
- Update XLA pin.
- Improve the comment in WORKSPACE describing how to override the TF re…
- Change the remat rule registration to register a default with a GPU s…
- Add an undocumented method on jit() functions to clear the function c…
- Improve documentation for jax.tree_util.tree_map.
- Clarify that functions passed to jax.jit must be weakly referenceable.
- [IREE] Allow backend selection via a flag.
- 2nd attempt at fixing IREE integration.
- Fix bitrot in IREE Python integration.
- Bump minimum Mac OS version to 10.14 (Mojave).
- Disallow complex dtypes for lax ops rem, lt, le, gt, and ge.
- Use util.unzipN() in more places instead of zip(*args).
- Fix autodiff of pad() of scalar inputs.
- Fix batching rule for convolution for batch dimensions of size 0.
Reviewed 18 pull requests in 2 repositories
google/jax
16 pull requests
- Make sure libtpu is initialized before initializing TPU profiling.
- [x64] make scipy_ndimage_test compatible with strict dtype promotion
- [x64] make lax_test compatible with strict dtype promotion
- [x64] make image_test compatible with strict dtype promotion
- [x64] make fft functionality compatible with strict dtype promotion
- [x64] make scipy_stats_test.py compatible with strict dtype promotion
- [x64] make sparse tests compatible with strict dtype promotion
- [x64] make api_test compatible with strict dtype promotion
- [x64] make tree_util_test compatible with strict dtype promotion
- [x64] make lax_scipy_test.py compatible with strict dtype promotion
- [x64] preserve weak types in promote_dtypes_inexact
- Update lax.py
-
Store profiler server as a global variable and add a
stop_serverfunction - Disallow complex dtypes for lax ops rem, lt, le, gt, and ge.
- Fix autodiff of pad() of scalar inputs.
- [x64] deprecate unsafe type casting in scatter-update operations
tensorflow/tensorflow
2 pull requests
Created an issue in google/iree that received 26 comments
Lowering failure for mhlo.pad of empty tensor
import iree.compiler CODE = """
#loc0 = loc(unknown)
module @jit_prim_fun.0 { func.func public @main(%arg0: tensor<0x2xi16> loc(unknown), %arg1: te…
26
comments
Opened 20 other issues in 2 repositories
google/iree
17
open
1
closed
- Fusion of f16->f32 cast into matmul
- Slow f16 matmul on Intel CPU
- CUDA_ERROR_INVALID_HANDLE from cuLaunchKernel possibly related to CUDA context handling
- ireec crash for mhlo.reduce_window of scalar
- Missing legalization for variadic mhlo.reduce_window
- ireec crash for mhlo.pad operator
-
Failed to legalize
mhlo.while - Lowering failure for i1 mhlo.gather
- Lowering failure for mhlo.dynamic-update-slice with empty update
- Lowering failure for i1 mhlo.dot_general
- Wrong output from mhlo.dynamic_slice with uint8 index
- Support float16 NumPy outputs
- MHLO lowering fails for mixed-precision mhlo.dot_general
- ireec crash for JAX cummax computation
- Failed to legalize integer mhlo.convolution
- ireec crash for MHLO convolution
- Unsigned integer outputs become signed
- Segfault when creating multiple IREE SystemContexts