Elementwise
Elementwise ops operate on a per element basis. They don't change the shape of the tensor.
Unary Ops (math)¤
logical_not
¤
logical_not()
Computes the logical NOT of the tensor element-wise.
print(Tensor([False, True]).logical_not().numpy())
[ True False]
Source code in tinygrad/tensor.py
2350 2351 2352 2353 2354 2355 2356 2357 2358 |
|
neg
¤
neg()
Negates the tensor element-wise.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).neg().numpy())
[ 3. 2. 1. -0. -1. -2. -3.]
Source code in tinygrad/tensor.py
2359 2360 2361 2362 2363 2364 2365 2366 2367 |
|
log
¤
log()
Computes the natural logarithm element-wise.
See: https://en.wikipedia.org/wiki/Logarithm
print(Tensor([1., 2., 4., 8.]).log().numpy())
[0. 0.6931 1.3863 2.0794]
Source code in tinygrad/tensor.py
2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 |
|
log2
¤
log2()
Computes the base-2 logarithm element-wise.
See: https://en.wikipedia.org/wiki/Logarithm
print(Tensor([1., 2., 4., 8.]).log2().numpy())
[0. 1. 2. 3.]
Source code in tinygrad/tensor.py
2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 |
|
exp
¤
exp()
Computes the exponential function element-wise.
See: https://en.wikipedia.org/wiki/Exponential_function
print(Tensor([0., 1., 2., 3.]).exp().numpy())
[ 1. 2.7183 7.3891 20.0855]
Source code in tinygrad/tensor.py
2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 |
|
exp2
¤
exp2()
Computes the base-2 exponential function element-wise.
See: https://en.wikipedia.org/wiki/Exponential_function
print(Tensor([0., 1., 2., 3.]).exp2().numpy())
[1. 2. 4. 8.]
Source code in tinygrad/tensor.py
2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 |
|
sqrt
¤
sqrt()
Computes the square root of the tensor element-wise.
print(Tensor([1., 2., 3., 4.]).sqrt().numpy())
[1. 1.4142 1.7321 2. ]
Source code in tinygrad/tensor.py
2458 2459 2460 2461 2462 2463 2464 2465 2466 |
|
rsqrt
¤
rsqrt()
Computes the reciprocal of the square root of the tensor element-wise.
print(Tensor([1., 2., 3., 4.]).rsqrt().numpy())
[1. 0.7071 0.5774 0.5 ]
Source code in tinygrad/tensor.py
2467 2468 2469 2470 2471 2472 2473 2474 2475 |
|
sin
¤
sin()
Computes the sine of the tensor element-wise.
print(Tensor([0., math.pi/2, math.pi, 3*math.pi/2, 2*math.pi]).sin().numpy())
[ 0. 1. -0. -1. 0.]
Source code in tinygrad/tensor.py
2476 2477 2478 2479 2480 2481 2482 2483 2484 |
|
cos
¤
cos()
Computes the cosine of the tensor element-wise.
print(Tensor([0., math.pi/2, math.pi, 3*math.pi/2, 2*math.pi]).cos().numpy())
[ 1.0000e+00 0.0000e+00 -1.0000e+00 -2.3842e-07 1.0000e+00]
Source code in tinygrad/tensor.py
2485 2486 2487 2488 2489 2490 2491 2492 2493 |
|
tan
¤
tan()
Computes the tangent of the tensor element-wise.
print(Tensor([0., math.pi/4, math.pi/2, 3*math.pi/4, math.pi]).tan().numpy())
[ 0. 1. inf -1. 0.]
Source code in tinygrad/tensor.py
2494 2495 2496 2497 2498 2499 2500 2501 2502 |
|
asin
¤
asin()
Computes the inverse sine (arcsine) of the tensor element-wise.
print(Tensor([-0.9, -0.6, -0.3, 0., 0.3, 0.6, 0.9]).asin().numpy())
[-1.1198 -0.6435 -0.3047 0. 0.3047 0.6435 1.1198]
Source code in tinygrad/tensor.py
2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 |
|
acos
¤
acos()
Computes the inverse cosine (arccosine) of the tensor element-wise.
print(Tensor([-0.9, -0.6, -0.3, 0., 0.3, 0.6, 0.9]).acos().numpy())
[2.6906 2.2143 1.8755 1.5708 1.2661 0.9273 0.451 ]
Source code in tinygrad/tensor.py
2517 2518 2519 2520 2521 2522 2523 2524 2525 |
|
atan
¤
atan()
Computes the inverse tangent (arctan) of the tensor element-wise.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).atan().numpy())
[-1.249 -1.1071 -0.7854 0. 0.7854 1.1071 1.249 ]
Source code in tinygrad/tensor.py
2527 2528 2529 2530 2531 2532 2533 2534 2535 |
|
trunc
¤
trunc() -> Tensor
Truncates the tensor element-wise.
print(Tensor([-3.5, -2.5, -1.5, -0.5, 0.5, 1.5, 2.5, 3.5]).trunc().numpy())
[-3. -2. -1. 0. 0. 1. 2. 3.]
Source code in tinygrad/tensor.py
2539 2540 2541 2542 2543 2544 2545 2546 2547 |
|
ceil
¤
ceil() -> Tensor
Rounds the tensor element-wise towards positive infinity.
print(Tensor([-3.5, -2.5, -1.5, -0.5, 0.5, 1.5, 2.5, 3.5]).ceil().numpy())
[-3. -2. -1. 0. 1. 2. 3. 4.]
Source code in tinygrad/tensor.py
2548 2549 2550 2551 2552 2553 2554 2555 2556 |
|
floor
¤
floor() -> Tensor
Rounds the tensor element-wise towards negative infinity.
print(Tensor([-3.5, -2.5, -1.5, -0.5, 0.5, 1.5, 2.5, 3.5]).floor().numpy())
[-4. -3. -2. -1. 0. 1. 2. 3.]
Source code in tinygrad/tensor.py
2557 2558 2559 2560 2561 2562 2563 2564 2565 |
|
round
¤
round() -> Tensor
Rounds the tensor element-wise with rounding half to even.
print(Tensor([-3.5, -2.5, -1.5, -0.5, 0.5, 1.5, 2.5, 3.5]).round().numpy())
[-4. -2. -2. 0. 0. 2. 2. 4.]
Source code in tinygrad/tensor.py
2566 2567 2568 2569 2570 2571 2572 2573 2574 |
|
isinf
¤
Checks the tensor element-wise to return True where the element is infinity, otherwise returns False
print(Tensor([1, float('inf'), 2, float('-inf'), float('nan')]).isinf().numpy())
[False True False True False]
Source code in tinygrad/tensor.py
2576 2577 2578 2579 2580 2581 2582 2583 2584 |
|
isnan
¤
isnan()
Checks the tensor element-wise to return True where the element is NaN, otherwise returns False
print(Tensor([1, float('inf'), 2, float('-inf'), float('nan')]).isnan().numpy())
[False False False False True]
Source code in tinygrad/tensor.py
2585 2586 2587 2588 2589 2590 2591 2592 2593 |
|
lerp
¤
Linearly interpolates between self
and end
by weight
.
print(Tensor([1., 2., 3.]).lerp(Tensor([4., 5., 6.]), 0.5).numpy())
[2.5 3.5 4.5]
Source code in tinygrad/tensor.py
2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 |
|
square
¤
square()
Squares the tensor element-wise.
Equivalent to self*self
.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).square().numpy())
[9. 4. 1. 0. 1. 4. 9.]
Source code in tinygrad/tensor.py
2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 |
|
clamp
¤
clamp(min_=None, max_=None)
Clips (clamps) the values in the tensor between min_
and max_
element-wise.
If min_
is None
, there is no lower bound. If max_
is None, there is no upper bound.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).clip(-1, 1).numpy())
[-1. -1. -1. 0. 1. 1. 1.]
Source code in tinygrad/tensor.py
2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 |
|
clip
¤
clip(min_=None, max_=None)
Alias for Tensor.clamp
.
Source code in tinygrad/tensor.py
2630 2631 2632 2633 2634 |
|
sign
¤
sign()
Returns the sign of the tensor element-wise.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).sign().numpy())
[-1. -1. -1. 0. 1. 1. 1.]
Source code in tinygrad/tensor.py
2635 2636 2637 2638 2639 2640 2641 2642 2643 |
|
abs
¤
abs()
Computes the absolute value of the tensor element-wise.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).abs().numpy())
[3. 2. 1. 0. 1. 2. 3.]
Source code in tinygrad/tensor.py
2644 2645 2646 2647 2648 2649 2650 2651 2652 |
|
reciprocal
¤
reciprocal()
Compute 1/x
element-wise.
print(Tensor([1., 2., 3., 4.]).reciprocal().numpy())
[1. 0.5 0.3333 0.25 ]
Source code in tinygrad/tensor.py
2653 2654 2655 2656 2657 2658 2659 2660 2661 |
|
Unary Ops (activation)¤
relu
¤
relu()
Applies the Rectified Linear Unit (ReLU) function element-wise.
- Described: https://paperswithcode.com/method/relu
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).relu().numpy())
[0. 0. 0. 0. 1. 2. 3.]
Source code in tinygrad/tensor.py
2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 |
|
sigmoid
¤
sigmoid()
Applies the Sigmoid function element-wise.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).sigmoid().numpy())
[0.0474 0.1192 0.2689 0.5 0.7311 0.8808 0.9526]
Source code in tinygrad/tensor.py
2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 |
|
hardsigmoid
¤
Applies the Hardsigmoid function element-wise.
NOTE: default alpha
and beta
values is taken from torch
- Described: https://paperswithcode.com/method/hard-sigmoid
- See: https://pytorch.org/docs/stable/generated/torch.nn.functional.hardsigmoid.html
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).hardsigmoid().numpy())
[0. 0.1667 0.3333 0.5 0.6667 0.8333 1. ]
Source code in tinygrad/tensor.py
2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 |
|
elu
¤
elu(alpha=1.0)
Applies the Exponential Linear Unit (ELU) function element-wise.
- Described: https://paperswithcode.com/method/elu
- Paper: https://arxiv.org/abs/1511.07289v5
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).elu().numpy())
[-0.9502 -0.8647 -0.6321 0. 1. 2. 3. ]
Source code in tinygrad/tensor.py
2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 |
|
celu
¤
celu(alpha=1.0)
Applies the Continuously differentiable Exponential Linear Unit (CELU) function element-wise.
- Described: https://paperswithcode.com/method/celu
- Paper: https://arxiv.org/abs/1704.07483
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).celu().numpy())
[-0.9502 -0.8647 -0.6321 0. 1. 2. 3. ]
Source code in tinygrad/tensor.py
2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 |
|
selu
¤
selu(alpha=1.67326, gamma=1.0507)
Applies the Scaled Exponential Linear Unit (SELU) function element-wise.
- Described: https://paperswithcode.com/method/selu
- Paper: https://arxiv.org/abs/1706.02515v5
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).selu().numpy())
[-1.6706 -1.5202 -1.1113 0. 1.0507 2.1014 3.1521]
Source code in tinygrad/tensor.py
2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 |
|
swish
¤
swish()
See .silu()
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).swish().numpy())
[-0.1423 -0.2384 -0.2689 0. 0.7311 1.7616 2.8577]
Source code in tinygrad/tensor.py
2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 |
|
silu
¤
silu()
Applies the Sigmoid Linear Unit (SiLU) function element-wise.
- Described: https://paperswithcode.com/method/silu
- Paper: https://arxiv.org/abs/1606.08415
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).silu().numpy())
[-0.1423 -0.2384 -0.2689 0. 0.7311 1.7616 2.8577]
Source code in tinygrad/tensor.py
2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 |
|
relu6
¤
relu6()
Applies the ReLU6 function element-wise.
- Described: https://paperswithcode.com/method/relu6
- Paper: https://arxiv.org/abs/1704.04861v1
print(Tensor([-9., -6., -3., 0., 3., 6., 9.]).relu6().numpy())
[0. 0. 0. 0. 3. 6. 6.]
Source code in tinygrad/tensor.py
2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 |
|
hardswish
¤
hardswish()
Applies the Hardswish function element-wise.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).hardswish().numpy())
[-0. -0.3333 -0.3333 0. 0.6667 1.6667 3. ]
Source code in tinygrad/tensor.py
2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 |
|
tanh
¤
tanh()
Applies the Hyperbolic Tangent (tanh) function element-wise.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).tanh().numpy())
[-0.9951 -0.964 -0.7616 0. 0.7616 0.964 0.9951]
Source code in tinygrad/tensor.py
2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 |
|
sinh
¤
sinh()
Applies the Hyperbolic Sine (sinh) function element-wise.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).sinh().numpy())
[-10.0179 -3.6269 -1.1752 0. 1.1752 3.6269 10.0179]
Source code in tinygrad/tensor.py
2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 |
|
cosh
¤
cosh()
Applies the Hyperbolic Cosine (cosh) function element-wise.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).cosh().numpy())
[10.0677 3.7622 1.5431 1. 1.5431 3.7622 10.0677]
Source code in tinygrad/tensor.py
2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 |
|
atanh
¤
atanh()
Applies the Inverse Hyperbolic Tangent (atanh) function element-wise.
print(Tensor([-0.9, -0.6, -0.3, 0., 0.3, 0.6, 0.9]).atanh().numpy())
[-1.4722 -0.6931 -0.3095 0. 0.3095 0.6931 1.4722]
Source code in tinygrad/tensor.py
2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 |
|
asinh
¤
asinh()
Applies the Inverse Hyperbolic Sine (asinh) function element-wise.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).asinh().numpy())
[-1.8184 -1.4436 -0.8814 0. 0.8814 1.4436 1.8184]
Source code in tinygrad/tensor.py
2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 |
|
acosh
¤
acosh()
Applies the Inverse Hyperbolic Cosine (acosh) function element-wise.
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).acosh().numpy())
[ nan nan nan nan 0. 1.317 1.7627]
Source code in tinygrad/tensor.py
2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 |
|
hardtanh
¤
hardtanh(min_val=-1, max_val=1)
Applies the Hardtanh function element-wise.
print(Tensor([-1.5, -1.0, -0.5, 0., 0.5, 1.0, 1.5]).hardtanh().numpy())
[-1. -1. -0.5 0. 0.5 1. 1. ]
Source code in tinygrad/tensor.py
2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 |
|
erf
¤
erf()
Applies error function element-wise.
- Described: https://en.wikipedia.org/wiki/Error_function
print(Tensor([-1.5, -1.0, -0.5, 0., 0.5, 1.0, 1.5]).erf().numpy())
[-0.9661 -0.8427 -0.5205 0. 0.5205 0.8427 0.9661]
Source code in tinygrad/tensor.py
2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 |
|
gelu
¤
gelu()
Applies the Gaussian Error Linear Unit (GELU) function element-wise.
- Described: https://paperswithcode.com/method/gelu
- Paper: https://arxiv.org/abs/1606.08415v5
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).gelu().numpy())
[-0.0036 -0.0454 -0.1588 0. 0.8412 1.9546 2.9964]
Source code in tinygrad/tensor.py
2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 |
|
quick_gelu
¤
quick_gelu()
Applies the Sigmoid GELU approximation element-wise.
- Described: https://paperswithcode.com/method/gelu
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).quick_gelu().numpy())
[-0.0181 -0.0643 -0.1542 0. 0.8458 1.9357 2.9819]
Source code in tinygrad/tensor.py
2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 |
|
leakyrelu
¤
leakyrelu(neg_slope=0.01)
Applies the Leaky ReLU function element-wise.
- Described: https://paperswithcode.com/method/leaky-relu
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).leakyrelu().numpy())
[-0.03 -0.02 -0.01 0. 1. 2. 3. ]
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).leakyrelu(neg_slope=0.42).numpy())
[-1.26 -0.84 -0.42 0. 1. 2. 3. ]
Source code in tinygrad/tensor.py
2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 |
|
mish
¤
mish()
Applies the Mish function element-wise.
- Described: https://paperswithcode.com/method/mish
- Paper: https://arxiv.org/abs/1908.08681v3
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).mish().numpy())
[-0.1456 -0.2525 -0.3034 0. 0.8651 1.944 2.9865]
Source code in tinygrad/tensor.py
2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 |
|
softplus
¤
softplus(beta=1)
Applies the Softplus function element-wise.
- Described: https://paperswithcode.com/method/softplus
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).softplus().numpy())
[0.0486 0.1269 0.3133 0.6931 1.3133 2.1269 3.0486]
Source code in tinygrad/tensor.py
2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 |
|
softsign
¤
softsign()
Applies the Softsign function element-wise.
- Described: https://paperswithcode.com/method/softsign
print(Tensor([-3., -2., -1., 0., 1., 2., 3.]).softsign().numpy())
[-0.75 -0.6667 -0.5 0. 0.5 0.6667 0.75 ]
Source code in tinygrad/tensor.py
2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 |
|
Elementwise Ops (broadcasted)¤
add
¤
Adds self
and x
.
Equivalent to self + x
.
Supports broadcasting to a common shape, type promotion, and integer, float, boolean inputs.
Tensor.manual_seed(42)
t = Tensor.randn(4)
print(t.numpy())
[-0.5144 1.085 0.9089 -0.0841]
print(t.add(20).numpy())
[19.4856 21.085 20.9089 19.9159]
print(t.add(Tensor([[2.0], [3.5]])).numpy())
[[1.4856 3.085 2.9089 1.9159]
[2.9856 4.585 4.4089 3.4159]]
Source code in tinygrad/tensor.py
2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 |
|
sub
¤
Subtracts x
from self
.
Equivalent to self - x
.
Supports broadcasting to a common shape, type promotion, and integer, float, boolean inputs.
Tensor.manual_seed(42)
t = Tensor.randn(4)
print(t.numpy())
[-0.5144 1.085 0.9089 -0.0841]
print(t.sub(20).numpy())
[-20.5144 -18.915 -19.0911 -20.0841]
print(t.sub(Tensor([[2.0], [3.5]])).numpy())
[[-2.5144 -0.915 -1.0911 -2.0841]
[-4.0144 -2.415 -2.5911 -3.5841]]
Source code in tinygrad/tensor.py
2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 |
|
mul
¤
Multiplies self
and x
.
Equivalent to self * x
.
Supports broadcasting to a common shape, type promotion, and integer, float, boolean inputs.
Tensor.manual_seed(42)
t = Tensor.randn(4)
print(t.numpy())
[-0.5144 1.085 0.9089 -0.0841]
print(t.mul(3).numpy())
[-1.5431 3.2549 2.7267 -0.2523]
print(t.mul(Tensor([[-1.0], [2.0]])).numpy())
[[ 0.5144 -1.085 -0.9089 0.0841]
[-1.0287 2.17 1.8178 -0.1682]]
Source code in tinygrad/tensor.py
3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 |
|
div
¤
Divides self
by x
.
Equivalent to self / x
.
Supports broadcasting to a common shape, type promotion, and integer, float, boolean inputs.
div
performs true division.
Tensor.manual_seed(42)
t = Tensor.randn(4)
print(t.numpy())
[-0.5144 1.085 0.9089 -0.0841]
print(t.div(3).numpy())
[-0.1715 0.3617 0.303 -0.028 ]
print(Tensor([1, 4, 10]).div(Tensor([2, 3, 4])).numpy())
[0.5 1.3333 2.5 ]
Source code in tinygrad/tensor.py
3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 |
|
xor
¤
Computes bitwise xor of self
and x
.
Equivalent to self ^ x
.
Supports broadcasting to a common shape, type promotion, and integer, boolean inputs.
print(Tensor([-1, -2, 3]).xor(Tensor([1, 0, 3])).numpy())
[-2 -2 0]
print(Tensor([True, True, False, False]).xor(Tensor([True, False, True, False])).numpy())
[False True True False]
Source code in tinygrad/tensor.py
3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 |
|
lshift
¤
lshift(x: int)
Computes left arithmetic shift of self
by x
bits. self
must have unsigned dtype.
Equivalent to self << x
.
print(Tensor([1, 3, 31], dtype=dtypes.uint8).lshift(2).numpy())
[ 4 12 124]
Source code in tinygrad/tensor.py
3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 |
|
rshift
¤
rshift(x: int)
Computes right arithmetic shift of self
by x
bits. self
must have unsigned dtype.
Equivalent to self >> x
.
print(Tensor([4, 13, 125], dtype=dtypes.uint8).rshift(2).numpy())
[ 1 3 31]
Source code in tinygrad/tensor.py
3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 |
|
pow
¤
Computes power of self
with x
.
Equivalent to self ** x
.
print(Tensor([-1, 2, 3]).pow(2).numpy())
[1 4 9]
print(Tensor([-1, 2, 3]).pow(Tensor([-1.5, 0.5, 1.5])).numpy())
[ nan 1.4142 5.1962]
print((2 ** Tensor([-1, 2, 3])).numpy())
[0.5 4. 8. ]
Source code in tinygrad/tensor.py
3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 |
|
maximum
¤
Computes element-wise maximum of self
and x
.
print(Tensor([-1, 2, 3]).maximum(1).numpy())
[1 2 3]
print(Tensor([-1, 2, 3]).maximum(Tensor([-4, -2, 9])).numpy())
[-1 2 9]
Source code in tinygrad/tensor.py
3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 |
|
minimum
¤
Computes element-wise minimum of self
and x
.
print(Tensor([-1, 2, 3]).minimum(1).numpy())
[-1 1 1]
print(Tensor([-1, 2, 3]).minimum(Tensor([-4, -2, 9])).numpy())
[-4 -2 3]
Source code in tinygrad/tensor.py
3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 |
|
where
¤
Return a tensor of elements selected from either x
or y
, depending on self
.
output_i = x_i if self_i else y_i
.
cond = Tensor([[True, True, False], [True, False, False]])
print(cond.where(1, 3).numpy())
[[1 1 3]
[1 3 3]]
Tensor.manual_seed(42)
cond = Tensor.randn(2, 3)
print(cond.numpy())
[[ 0.9779 0.4678 0.5526]
[-0.3288 -0.8555 0.2753]]
print((cond > 0).where(cond, -float("inf")).numpy())
[[0.9779 0.4678 0.5526]
[ -inf -inf 0.2753]]
Source code in tinygrad/tensor.py
3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 |
|
Casting Ops¤
cast
¤
cast(dtype: DTypeLike) -> Tensor
Casts self
to the given dtype
.
t = Tensor([-1, 2.5, 3], dtype=dtypes.float)
print(t.dtype, t.numpy())
dtypes.float [-1. 2.5 3. ]
t = t.cast(dtypes.int32)
print(t.dtype, t.numpy())
dtypes.int [-1 2 3]
Source code in tinygrad/tensor.py
3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 |
|
bitcast
¤
bitcast(dtype: DTypeLike) -> Tensor
Bitcasts self
to the given dtype
of the same itemsize.
self
must not require a gradient.
t = Tensor([-1, 2, 3], dtype=dtypes.int32)
print(t.dtype, t.numpy())
dtypes.int [-1 2 3]
t = t.bitcast(dtypes.uint32)
print(t.dtype, t.numpy())
dtypes.uint [4294967295 2 3]
Source code in tinygrad/tensor.py
3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 |
|
float
¤
float() -> Tensor
Convenience method to cast self
to a float32
Tensor.
t = Tensor([-1, 2, 3], dtype=dtypes.int32)
print(t.dtype, t.numpy())
dtypes.int [-1 2 3]
t = t.float()
print(t.dtype, t.numpy())
dtypes.float [-1. 2. 3.]
Source code in tinygrad/tensor.py
3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 |
|
half
¤
half() -> Tensor
Convenience method to cast self
to a float16
Tensor.
t = Tensor([-1, 2, 3], dtype=dtypes.int32)
print(t.dtype, t.numpy())
dtypes.int [-1 2 3]
t = t.half()
print(t.dtype, t.numpy())
dtypes.half [-1. 2. 3.]
Source code in tinygrad/tensor.py
3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 |
|
int
¤
int() -> Tensor
Convenience method to cast self
to a int32
Tensor.
t = Tensor([-1.5, -0.5, 0.0, 0.5, 1.5])
print(t.dtype, t.numpy())
dtypes.float [-1.5 -0.5 0. 0.5 1.5]
t = t.int()
print(t.dtype, t.numpy())
dtypes.int [-1 0 0 0 1]
Source code in tinygrad/tensor.py
3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 |
|
bool
¤
bool() -> Tensor
Convenience method to cast self
to a bool
Tensor.
t = Tensor([-1, 0, 1])
print(t.dtype, t.numpy())
dtypes.int [-1 0 1]
t = t.bool()
print(t.dtype, t.numpy())
dtypes.bool [ True False True]
Source code in tinygrad/tensor.py
3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 |
|