geohot
Follow
Highlights
- Pro
Pinned
1,460 contributions in the last year
Contribution activity
July 2023
Created 15 commits in 1 repository
Created a pull request in tinygrad/tinygrad that received 2 comments
Opened 6 other pull requests in 1 repository
tinygrad/tinygrad
6
merged
Reviewed 20 pull requests in 1 repository
tinygrad/tinygrad
20 pull requests
- ensure Tensor.__init__ implementation matches type annotations
- implement common subexpression elimination
- constant fold 1-d numpy arrays
- [READY] perf: view as named tuples
- Add WHERE ternary (or trinary?) op
- Tensor Cores 2: Local Buffers Edition
- Dedup kernel args (bufs) that have the same underlying RawBuffer
- Fix ShapeTracker mismatch in LazyBuffer.fromCPU
- Fix Metal tests [WIP]
- fix Tensor.manual_seed() default to wrong type
- helpers.py: improved test coverage + exception handling
- Renaming FusedOps
- Fold expand preceding reduce if the reduction is on the same axis as the expansion
- Remove POW llop and add SQRT llop
- Make triton backend work again
- Make cross_process use cloudpickle
- Broken Sigmoid backward: Add test and mlop for Sigmoid
- Remove float64
- dtypes: add int16, priority change, refactor
- Over 90% on CIFAR with examples/hlb_cifar10.py
Created an issue in tinygrad/tinygrad that received 5 comments
Switch nvcc -> cicc in CUDA backend
Apparently it's faster to compile? Does someone want to try this and benchmark?
Replace from pycuda.compiler import compile as cuda_compile with so…
5
comments
Opened 8 other issues in 1 repository
tinygrad/tinygrad
6
open
2
closed
- Perpetual Bounty: llama Python runtime 20%+ faster on M1 -- $200+
- Common subexpression elimination in the Linearizer
- Use GitHub actions to cache files we download in CI
- Fix constant folding for Tensor([3])
- Move datasets -> extra/datasets
- Add WHERE TrinaryOp/mlop
- Fix view merging for masked views
- Norm Fusion ($2000 bounty)