issues
search
ml-explore
/
mlx
MLX: An array framework for Apple silicon
https://ml-explore.github.io/mlx/
MIT License
14.83k
stars
845
forks
source link
issues
Oldest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Update cosine decay to not clamp output, match other implementations
#1138
jlwitthuhn
opened
1 day ago
0
Fix CPU copy bug with many dims
#1137
awni
closed
1 day ago
0
Update README.md
#1136
KPCOFGS
opened
1 day ago
0
Choose the right MLX bf16 for extensions
#1135
awni
closed
1 day ago
0
[BUG] mlx gets stuck with high-dimensional array on Linux
#1134
cccyf
closed
1 day ago
1
More jitting
#1132
awni
opened
2 days ago
0
Patch bump
#1131
awni
closed
2 days ago
0
Optimization Plans for Conv2D CPU Execution
#1130
abdussamettrkr
opened
3 days ago
0
Add groups to 2-D convolutions
#1129
Rifur13
opened
3 days ago
0
[BUG] AttributeError in mlx.core.conj and mlx.core.conjugate functions
#1127
profConradi
closed
3 days ago
3
Fix extensions
#1126
awni
closed
2 days ago
0
[Fix] minor typo in default argument for argpartition's "axis" parameter
#1125
PRESIDENT810
closed
3 days ago
0
Block sparse qmm
#1124
angeloskath
closed
2 days ago
1
Fused GEMM
#1123
jagrit06
closed
3 days ago
0
Fix scatter index bug
#1122
awni
closed
4 days ago
0
[BUG] Passing `axis=None` into `argpartition` causes `TypeError`
#1121
hegyibalint
closed
3 days ago
1
Allow conversion to dlpack
#1120
awni
closed
2 days ago
3
Implemented Cholesky on CPU #1026
#1119
arn4
closed
1 day ago
0
Move args in conv_general
#1118
zcbenz
closed
5 days ago
0
The tile op does not accept None as reps
#1117
zcbenz
closed
5 days ago
0
Use compiled function in Sigmoid module
#1116
zcbenz
closed
5 days ago
0
Use correct module type in quantized.py
#1115
zcbenz
closed
5 days ago
0
padding_lo is passed as padding_hi in conv_general
#1114
zcbenz
closed
5 days ago
1
Fill vector with constructor instead of fill_n
#1113
zcbenz
closed
5 days ago
0
Remove unused code in mlx_scatter_args_slice
#1112
zcbenz
closed
4 days ago
1
Pass missing stream arg in array.flatten
#1111
zcbenz
closed
5 days ago
0
Add missing docs for some ops
#1110
zcbenz
closed
5 days ago
0
Detect metal version and propagate correctly for JIT
#1109
awni
closed
3 days ago
0
[Feature] something like `mlx.scipy.stats`
#1107
PaulScemama
closed
5 days ago
2
No CPU option for binary minimization
#1105
awni
closed
5 days ago
0
Fix typo in a variable name in example code.
#1104
jaju
closed
6 days ago
0
Feature complete Metal FFT
#1102
barronalex
opened
1 week ago
3
Improve error messaging in eval
#1101
awni
closed
1 week ago
0
Add conjugate operator
#1100
barronalex
closed
1 week ago
0
v0.13
#1099
awni
closed
1 week ago
0
Add compiler flags to disable safetensors and gguf
#1098
awni
closed
1 week ago
0
[WIP] Comms
#1097
angeloskath
opened
1 week ago
3
[Feature] Build flag to make safetensor and GGUF dependencies optional
#1096
awni
closed
1 week ago
0
[BUG] mx.radians & mx.degrees - unexpected behavior when the input is not an array
#1094
Crear12
closed
1 week ago
0
Allow unary ops to accept array like
#1093
awni
closed
1 week ago
1
IO thread-pool and device
#1092
angeloskath
opened
1 week ago
0
JIT compile option for binary minimization
#1091
awni
opened
1 week ago
2
[Feature] Expose something like custom VJP in Python
#1090
awni
opened
1 week ago
2
Enhancement - Use M series iPads as extra GPU power
#1089
petergreis
closed
1 week ago
1
[Feature] Rotary Positional Embeddings for Nomic Embed
#1088
zanussbaum
closed
1 week ago
2
Update gemm block offset adjustment to be in size_t
#1087
jagrit06
closed
1 week ago
0
[Feature] Leak memory on exit
#1086
zcbenz
opened
1 week ago
6
Split encoders in non-concurrent context with a max ops per encoder
#1085
awni
closed
1 week ago
1
[BUG] Bad result for GPU matmul for specific shape
#1084
awni
closed
1 week ago
1
[BUG] give better diagnostic message when calling compiled code with an eval in it -- currently "Attempting to eval an array without a primitive"
#1083
davidkoski
closed
1 week ago
8
Next