feat: implement a separate `TracedRNumber` #161

avik-pal · 2024-10-04T17:14:59Z

~~currently very WIP~~

src/Reactant.jl

src/TracedRArray.jl

src/TracedRNumber.jl

src/TracedRArray.jl

github-actions

Reactant.jl Benchmarks

Benchmark suite	Current: `8a9f06c`	Previous: `f2c0e8a`	Ratio
`ViT base (256 x 256 x 3 x 32)/forward/CUDA/Reactant`	`1318556698` ns	`1315729546` ns	`1.00`
`ViT base (256 x 256 x 3 x 32)/forward/CUDA/Lux`	`213965204` ns	`212083499` ns	`1.01`
`ViT base (256 x 256 x 3 x 32)/forward/CPU/Reactant`	`6804934419` ns	`5286469750` ns	`1.29`
`ViT base (256 x 256 x 3 x 32)/forward/CPU/Lux`	`18511487331` ns	`23583347555` ns	`0.78`
`ViT small (256 x 256 x 3 x 4)/forward/CUDA/Reactant`	`1255068856` ns	`1254858296` ns	`1.00`
`ViT small (256 x 256 x 3 x 4)/forward/CUDA/Lux`	`8751128` ns	`8478570` ns	`1.03`
`ViT small (256 x 256 x 3 x 4)/forward/CPU/Reactant`	`1643973338` ns	`1636237670` ns	`1.00`
`ViT small (256 x 256 x 3 x 4)/forward/CPU/Lux`	`2060290806` ns	`2376437823` ns	`0.87`
`ViT tiny (256 x 256 x 3 x 32)/forward/CUDA/Reactant`	`1304838533` ns	`1266018905` ns	`1.03`
`ViT tiny (256 x 256 x 3 x 32)/forward/CUDA/Lux`	`93074730.5` ns	`84820407` ns	`1.10`
`ViT tiny (256 x 256 x 3 x 32)/forward/CPU/Reactant`	`2253189844` ns	`2170879105` ns	`1.04`
`ViT tiny (256 x 256 x 3 x 32)/forward/CPU/Lux`	`6064963189` ns	`4675094299` ns	`1.30`
`ViT tiny (256 x 256 x 3 x 4)/forward/CUDA/Reactant`	`1297977475` ns	`1263496480` ns	`1.03`
`ViT tiny (256 x 256 x 3 x 4)/forward/CUDA/Lux`	`7525710.5` ns	`7782824` ns	`0.97`
`ViT tiny (256 x 256 x 3 x 4)/forward/CPU/Reactant`	`1466869310` ns	`1467043032.5` ns	`1.00`
`ViT tiny (256 x 256 x 3 x 4)/forward/CPU/Lux`	`1361212717` ns	`1685775445` ns	`0.81`
`ViT tiny (256 x 256 x 3 x 16)/forward/CUDA/Reactant`	`1314689632` ns	`1306815930` ns	`1.01`
`ViT tiny (256 x 256 x 3 x 16)/forward/CUDA/Lux`	`11418734.5` ns	`11611908` ns	`0.98`
`ViT tiny (256 x 256 x 3 x 16)/forward/CPU/Reactant`	`1752787751.5` ns	`1752808523` ns	`1.00`
`ViT tiny (256 x 256 x 3 x 16)/forward/CPU/Lux`	`2629684857` ns	`2463987825.5` ns	`1.07`
`ViT small (256 x 256 x 3 x 16)/forward/CUDA/Reactant`	`1276847603.5` ns	`1325877558.5` ns	`0.96`
`ViT small (256 x 256 x 3 x 16)/forward/CUDA/Lux`	`86382264` ns	`90330187` ns	`0.96`
`ViT small (256 x 256 x 3 x 16)/forward/CPU/Reactant`	`2216207055` ns	`2213119086` ns	`1.00`
`ViT small (256 x 256 x 3 x 16)/forward/CPU/Lux`	`3548437058` ns	`4023816395` ns	`0.88`
`ViT small (256 x 256 x 3 x 32)/forward/CUDA/Reactant`	`1308980482.5` ns	`1270812264` ns	`1.03`
`ViT small (256 x 256 x 3 x 32)/forward/CUDA/Lux`	`116489754.5` ns	`113097539` ns	`1.03`
`ViT small (256 x 256 x 3 x 32)/forward/CPU/Reactant`	`3037687031` ns	`3042643080` ns	`1.00`
`ViT small (256 x 256 x 3 x 32)/forward/CPU/Lux`	`9576755622` ns	`8210106924.5` ns	`1.17`
`ViT base (256 x 256 x 3 x 16)/forward/CUDA/Reactant`	`1332370009` ns	`1324054039` ns	`1.01`
`ViT base (256 x 256 x 3 x 16)/forward/CUDA/Lux`	`126824571` ns	`127669686.5` ns	`0.99`
`ViT base (256 x 256 x 3 x 16)/forward/CPU/Reactant`	`3191878044` ns	`3203794253` ns	`1.00`
`ViT base (256 x 256 x 3 x 16)/forward/CPU/Lux`	`7124639562` ns	`11004907984` ns	`0.65`
`ViT base (256 x 256 x 3 x 4)/forward/CUDA/Reactant`	`1280201506` ns	`1299288245` ns	`0.99`
`ViT base (256 x 256 x 3 x 4)/forward/CUDA/Lux`	`83942324` ns	`96277750` ns	`0.87`
`ViT base (256 x 256 x 3 x 4)/forward/CPU/Reactant`	`1900998509` ns	`2155333265.5` ns	`0.88`
`ViT base (256 x 256 x 3 x 4)/forward/CPU/Lux`	`2374950424` ns	`2863535293.5` ns	`0.83`

This comment was automatically generated by workflow using github-action-benchmark.

github-actions · 2024-10-05T03:52:59Z

Benchmark Results

	main	`4a81556`...	main/4a81556a8fc685...
comptime/basics/2D sum	27.4 ± 2.3 ms	27.5 ± 1.2 ms	0.997
comptime/basics/Basic grad cos	0.0455 ± 0.00079 s	0.0457 ± 0.00073 s	0.995
comptime/basics/cos.(x)	0.0323 ± 0.0014 s	0.0321 ± 0.00059 s	1.01
comptime/lux neural networks/ViT base	5.88 s	5.83 s	1.01
comptime/lux neural networks/ViT tiny	6.04 s	5.85 s	1.03
comptime/lux neural networks/vgg11 bn=false	0.4 ± 0.017 s	0.409 ± 0.0068 s	0.977
comptime/lux neural networks/vgg13 bn=false	0.439 ± 0.017 s	0.47 ± 0.015 s	0.934
comptime/lux neural networks/vgg16 bn=false	0.534 ± 0.012 s	0.553 ± 0.0049 s	0.964
comptime/lux neural networks/vgg19 bn=false	0.616 ± 0.0099 s	0.635 ± 0.024 s	0.969
runtime/lux neural networks/ViT base (compiled)	6.38 s	6.37 s	1
runtime/lux neural networks/ViT tiny (compiled)	1.71 s	1.69 s	1.01
runtime/lux neural networks/vgg11 bn=false (compiled)	2.14 s	2.13 s	1
runtime/lux neural networks/vgg13 bn=false (compiled)	2.97 s	2.95 s	1.01
runtime/lux neural networks/vgg16 bn=false (compiled)	3.78 s	3.77 s	1
runtime/lux neural networks/vgg19 bn=false (compiled)	4.55 s	4.55 s	1
time_to_load	1.45 ± 0.0048 s	1.34 ± 0.006 s	1.08

Benchmark Plots

A plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR.
Go to "Actions"->"Benchmark a pull request"->[the most recent run]->"Artifacts" (at the bottom).

mofeing · 2024-10-05T03:56:24Z

Let me just merge #163 after CI checks pass and rebase this PR on top of it because it adds some tests that should be passed by TracedRNumbers too.

src/Reactant.jl

wsmoses

Lgtm only minor comment: would defining traced value = union{tracedrarrsy, travednumber} simplify some things

avik-pal · 2024-10-05T15:54:56Z

Do we also need a ConcreteRScalar? If we want to extend to_rarray to all of ReactantPrimitives

wsmoses · 2024-10-05T16:21:57Z

For now we can leave concrete as is

mofeing · 2024-10-05T18:06:09Z

I don't think we need a concrete scalar type, because indexing on a concrete array is just a normal number.

avik-pal · 2024-10-05T18:21:05Z

I don't think we need a concrete scalar type, because indexing on a concrete array is just a normal number.

I meant more from a tracing perspective. What should to_rarray(::ReactantPrimitives) return? If it is Number then we can't convert it to a TracedRNumber. If we want to convert it then we need something like a ConcreteRNumber so that users are aware that it is going to be traced.

wsmoses · 2024-10-05T18:28:36Z

So I think this is two different questions.

Should we upgrade scalars into traced values (like we do for arrays) or compile with respect to them (like we do constants).

For now I'm fine leaving them as is which is compile as constants

Can we represent a ConcreteScalar, if desired. I think this will be necessary in the future.

For example we may want to have a function with a user defined index offset.

e.g.

A = ConcreteRarray
v = ConcreteScalar
f = @compile getindex(A, v)

without a concrete scalar we need to recompile for each index. I do think we need such a scalar, even if the default behavior for conversion is only converting arrays

wsmoses · 2024-10-05T18:30:07Z

Even more simply though, we may need as the return type of a function with array inputs and a scalar output.

i.e.

x = ConcreteArray(ones(4,4))
f = @compile sum(x)
f(x) # this should return a scalar now, but we cannot actually represent this as a regularfloat64, since it will be the result of a future/etc

avik-pal · 2024-10-05T19:44:18Z

Looking at the above example I also realized the mapreduce semantics is incorrect at-present. For example sum(x) should return TracedRNumber and not TracedRArray{T,0}

Even more simply though, we may need as the return type of a function with array inputs and a scalar output.

We could store it like an array in the ConcreteRArray but expose it to the enduser as a scalar. JuliaGPU/GPUArrays.jl#550 is something similar in the GPUArrays world. But let's keep this PR simple and avoid this for now

codecov-commenter · 2024-10-05T22:20:16Z

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

Attention: Patch coverage is 60.13072% with 61 lines in your changes missing coverage. Please review.

Project coverage is 30.62%. Comparing base (9904590) to head (841376d).

Files with missing lines	Patch %	Lines
src/TracedRNumber.jl	50.81%	30 Missing ⚠️
src/TracedRArray.jl	62.79%	16 Missing ⚠️
ext/ReactantNNlibExt.jl	0.00%	7 Missing ⚠️
src/Tracing.jl	80.64%	6 Missing ⚠️
src/ConcreteRArray.jl	0.00%	1 Missing ⚠️
src/XLA.jl	0.00%	1 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #161      +/-   ##
==========================================
- Coverage   33.09%   30.62%   -2.47%     
==========================================
  Files          37       38       +1     
  Lines        5107     5175      +68     
==========================================
- Hits         1690     1585     -105     
- Misses       3417     3590     +173

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

src/utils.jl

wsmoses

a small comment but otherwise lgtm

mofeing · 2024-10-06T01:05:33Z

now that #163 has been merged, we should test concatenation on numbers.

~~@avik-pal should I do it or do you prefer to do it yourself?~~

mofeing · 2024-10-06T02:10:57Z

There's a bug on concatenation of TracedRNumber. In particular, if we run this kernel:

using Reactant

x = fill(true)
x_concrete = Reactant.to_rarray(x)

function traced_vcat(x)
	a = x[]
	[a; a; a]
end

f = @compile traced_vcat(x_concrete)

f(x_concrete)

It fails with the following error:

ERROR: BoundsError: attempt to access 3-element Vector{ConcreteRArray{Bool}} at index [1]
Stacktrace:
 [1] traced_getfield
   @ ~/Developer/Reactant.jl/src/Compiler.jl:18 [inlined]
 [2] macro expansion
   @ ~/Developer/Reactant.jl/src/Compiler.jl:649 [inlined]
 [3] (::Reactant.Compiler.Thunk{Symbol("##test_vcat_reactant#229")})(args::ConcreteRArray{Bool, 0})
   @ Reactant.Compiler ~/Developer/Reactant.jl/src/Compiler.jl:665
 [4] top-level scope
   @ REPL[10]:1

The error seems to be that it's not doing any XLA call and it's setting result returning to 3 empty buffers. Check out the generated Julia code of f:

quote
    #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:639 =#
    #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:640 =#
    nothing
    #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:647 =#
    usbuf_1 = (getindex(args, 1)).data
    sbuf_1 = XLA.synced_buffer(usbuf_1)
    #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:648 =#
    ()
    #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:649 =#
    result = (ConcreteRArray{Bool})[ConcreteRArray{Bool, 0}(Reactant.XLA.AsyncBuffer(Reactant.XLA.Buffer(Ptr{Nothing} @0x0000000000000000), nothing), ()), ConcreteRArray{Bool, 0}(Reactant.XLA.AsyncBuffer(Reactant.XLA.Buffer(Ptr{Nothing} @0x0000000000000000), nothing), ()), ConcreteRArray{Bool, 0}(Reactant.XLA.AsyncBuffer(Reactant.XLA.Buffer(Ptr{Nothing} @0x0000000000000000), nothing), ())]
    (traced_getfield(result, $(Expr(:quote, 1)))).data = (args[1]).data
    (traced_getfield(result, $(Expr(:quote, 2)))).data = (args[1]).data
    (traced_getfield(result, $(Expr(:quote, 3)))).data = (args[1]).data
    #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:650 =#
    return result
end

wsmoses · 2024-10-06T02:13:57Z

That actually makes sense here since it’s making an array of values all available as args. What’s the issue?

…

On Sat, Oct 5, 2024 at 9:11 PM Sergio Sánchez Ramírez < ***@***.***> wrote: There's a bug on concatenation of TracedRNumber. In particular, if we run this kernel: using Reactant x = fill(true) x_concrete = Reactant.to_rarray(x) function traced_vcat(x) a = x[] [a; a; a]end f = @compile traced_vcat(x_concrete) f(x_concrete) It fails with the following error: ERROR: BoundsError: attempt to access 3-element Vector{ConcreteRArray{Bool}} at index [1] Stacktrace: [1] traced_getfield @ ~/Developer/Reactant.jl/src/Compiler.jl:18 [inlined] [2] macro expansion @ ~/Developer/Reactant.jl/src/Compiler.jl:649 [inlined] [3] (::Reactant.Compiler.Thunk{Symbol("##test_vcat_reactant#229")})(args::ConcreteRArray{Bool, 0}) @ Reactant.Compiler ~/Developer/Reactant.jl/src/Compiler.jl:665 [4] top-level scope @ REPL[10]:1 The error seems to be that it's not doing any XLA call and it's setting result returning to 3 empty buffers. Check out the generated Julia code of f: quote #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:639 =# #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:640 =# nothing #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:647 =# usbuf_1 = (getindex(args, 1)).data sbuf_1 = XLA.synced_buffer(usbuf_1) #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:648 =# () #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:649 =# result = (ConcreteRArray{Bool})[ConcreteRArray{Bool, 0}(Reactant.XLA.AsyncBuffer(Reactant.XLA.Buffer(Ptr{Nothing} @0x0000000000000000), nothing), ()), ConcreteRArray{Bool, 0}(Reactant.XLA.AsyncBuffer(Reactant.XLA.Buffer(Ptr{Nothing} @0x0000000000000000), nothing), ()), ConcreteRArray{Bool, 0}(Reactant.XLA.AsyncBuffer(Reactant.XLA.Buffer(Ptr{Nothing} @0x0000000000000000), nothing), ())] (traced_getfield(result, $(Expr(:quote, 1)))).data = (args[1]).data (traced_getfield(result, $(Expr(:quote, 2)))).data = (args[1]).data (traced_getfield(result, $(Expr(:quote, 3)))).data = (args[1]).data #= /Users/mofeing/Developer/Reactant.jl/src/Compiler.jl:650 =# return resultend — Reply to this email directly, view it on GitHub <#161 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJTUXFKSBYO4L2Z573JTE3Z2CL4RAVCNFSM6AAAAABPMI4XLGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGOJVGI3DCOJQGU> . You are receiving this because your review was requested.Message ID: ***@***.***>

mofeing · 2024-10-06T03:00:47Z

That actually makes sense here since it’s making an array of values all available as args. What’s the issue?

I don't get it. Its behavior should be like the one on 0-dim array test since we are passing a 0-dim array and should return a vector.

From the MLIR point of view, it should be calling stablehlo.concatenate on a list of 0-dim tensors.

wsmoses · 2024-10-06T06:12:14Z

ah I see, I think this is a bug on the new number and cat perhaps?

avik-pal · 2024-10-06T12:57:39Z

I think we won't do any XLA call here, see the julia fallbacks for number:

vcat(X::Number...) = hvcat_fill!(Vector{promote_typeof(X...)}(undef, length(X)), X)
hcat(X::Number...) = hvcat_fill!(Matrix{promote_typeof(X...)}(undef, 1,length(X)), X)

It is going to fill into a regular array

avik-pal · 2024-10-06T13:05:49Z

Should work now

julia> function traced_vcat(x)
               a = x[]; 
               Float64[a; a; a]
       end
traced_vcat (generic function with 1 method)

julia> @code_hlo optimize=false traced_vcat(x_concrete)
Module:
module {
  func.func @main(%arg0: tensor<i1>) -> (tensor<3xf64>, tensor<i1>) {
    %0 = stablehlo.transpose %arg0, dims = [] : (tensor<i1>) -> tensor<i1>
    %1 = stablehlo.broadcast_in_dim %0, dims = [] : (tensor<i1>) -> tensor<1xi1>
    %2 = stablehlo.broadcast_in_dim %0, dims = [] : (tensor<i1>) -> tensor<1xi1>
    %3 = stablehlo.broadcast_in_dim %0, dims = [] : (tensor<i1>) -> tensor<1xi1>
    %4 = stablehlo.convert %1 : (tensor<1xi1>) -> tensor<1xf64>
    %5 = stablehlo.convert %2 : (tensor<1xi1>) -> tensor<1xf64>
    %6 = stablehlo.convert %3 : (tensor<1xi1>) -> tensor<1xf64>
    %7 = stablehlo.concatenate %4, %5, %6, dim = 0 : (tensor<1xf64>, tensor<1xf64>, tensor<1xf64>) -> tensor<3xf64>
    %8 = stablehlo.transpose %7, dims = [0] : (tensor<3xf64>) -> tensor<3xf64>
    %9 = stablehlo.transpose %0, dims = [] : (tensor<i1>) -> tensor<i1>
    return %8, %9 : tensor<3xf64>, tensor<i1>
  }
}

julia> @code_hlo traced_vcat(x_concrete)
Module:
module attributes {transform.with_named_sequence} {
  func.func @main(%arg0: tensor<i1>) -> tensor<3xf64> {
    %0 = stablehlo.convert %arg0 : (tensor<i1>) -> tensor<f64>
    %1 = stablehlo.broadcast_in_dim %0, dims = [] : (tensor<f64>) -> tensor<3xf64>
    return %1 : tensor<3xf64>
  }
}

mofeing reviewed Oct 4, 2024

View reviewed changes

src/Reactant.jl Outdated Show resolved Hide resolved

mofeing reviewed Oct 4, 2024

View reviewed changes

src/TracedRArray.jl Outdated Show resolved Hide resolved

mofeing mentioned this pull request Oct 4, 2024

Generalize Base._cat to non-Val, typed Base._cat_t and implement typed_hcat, typed_vcat, typed_hvcat, typed_hvncat #163

Merged

avik-pal force-pushed the ap/scalar branch from 023f97a to 3d34693 Compare October 5, 2024 00:46

avik-pal changed the title ~~feat: implement a separate TracedRScalar~~ feat: implement a separate TracedRNumber Oct 5, 2024

avik-pal force-pushed the ap/scalar branch 2 times, most recently from 246cd00 to 40af781 Compare October 5, 2024 01:31

github-actions bot reviewed Oct 5, 2024

View reviewed changes

src/TracedRNumber.jl Outdated Show resolved Hide resolved

src/TracedRNumber.jl Outdated Show resolved Hide resolved

avik-pal force-pushed the ap/scalar branch 3 times, most recently from 113c2b5 to 10495fc Compare October 5, 2024 02:56

mofeing reviewed Oct 5, 2024

View reviewed changes

src/TracedRArray.jl Outdated Show resolved Hide resolved

avik-pal mentioned this pull request Oct 5, 2024

Optimize to_rarray to avoid recursion for simple inputs #140

Closed

avik-pal marked this pull request as ready for review October 5, 2024 03:30

avik-pal requested review from mofeing and wsmoses October 5, 2024 03:37

github-actions bot reviewed Oct 5, 2024

View reviewed changes

avik-pal mentioned this pull request Oct 5, 2024

feat: compile training loop automatically using reactant LuxDL/Lux.jl#969

Merged

11 tasks

wsmoses reviewed Oct 5, 2024

View reviewed changes

src/Reactant.jl Outdated Show resolved Hide resolved

wsmoses approved these changes Oct 5, 2024

View reviewed changes

feat: TracedRScalar

4a1be10

avik-pal added 2 commits October 5, 2024 16:48

fix: minor fixes

6aab7f7

feat: support logsoftmax

abc6a9e

avik-pal force-pushed the ap/scalar branch from 172369a to abc6a9e Compare October 5, 2024 21:30

fix: bool promote rule

841376d

fix: broadcasting of closures

eb3d1db

avik-pal force-pushed the ap/scalar branch from f15350b to eb3d1db Compare October 5, 2024 23:20

avik-pal requested a review from wsmoses October 5, 2024 23:21

wsmoses reviewed Oct 6, 2024

View reviewed changes

src/utils.jl Outdated Show resolved Hide resolved

wsmoses approved these changes Oct 6, 2024

View reviewed changes

refactor: use TracedTypes

944dca8

mofeing and others added 4 commits October 5, 2024 21:23

Merge branch 'main' into ap/scalar

500d12f

Fix type of preserved_args

3ecafef

Rename TracedTypes to TracedType

c03b5e0

small testset rename

60b614b

fix: special handling for concatenation of numbers

8a9f06c

avik-pal force-pushed the ap/scalar branch from 1282ce2 to 8a9f06c Compare October 6, 2024 13:22

avik-pal linked an issue Oct 6, 2024 that may be closed by this pull request

Inconsistent semantics of 0-dim RArrays compared to Julia semantics #141

Closed

mofeing added 2 commits October 6, 2024 13:35

Reenable tests

a35d7b7

Rename ReactantPrimitives to ReactantPrimitive

4a81556

mofeing approved these changes Oct 6, 2024

View reviewed changes

mofeing merged commit 6e89952 into main Oct 6, 2024
15 of 23 checks passed

mofeing deleted the ap/scalar branch October 6, 2024 18:41

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: implement a separate `TracedRNumber` #161

feat: implement a separate `TracedRNumber` #161

avik-pal commented Oct 4, 2024 •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot commented Oct 5, 2024 •

edited

Loading

mofeing commented Oct 5, 2024

wsmoses left a comment

avik-pal commented Oct 5, 2024

wsmoses commented Oct 5, 2024

mofeing commented Oct 5, 2024 •

edited

Loading

avik-pal commented Oct 5, 2024

wsmoses commented Oct 5, 2024

wsmoses commented Oct 5, 2024

avik-pal commented Oct 5, 2024

codecov-commenter commented Oct 5, 2024

wsmoses left a comment

mofeing commented Oct 6, 2024 •

edited

Loading

mofeing commented Oct 6, 2024

wsmoses commented Oct 6, 2024 via email

mofeing commented Oct 6, 2024 •

edited

Loading

wsmoses commented Oct 6, 2024

avik-pal commented Oct 6, 2024

avik-pal commented Oct 6, 2024

feat: implement a separate TracedRNumber #161

feat: implement a separate TracedRNumber #161

Conversation

avik-pal commented Oct 4, 2024 • edited Loading

github-actions bot left a comment • edited Loading

Choose a reason for hiding this comment

Reactant.jl Benchmarks

github-actions bot commented Oct 5, 2024 • edited Loading

Benchmark Results

Benchmark Plots

mofeing commented Oct 5, 2024

wsmoses left a comment

Choose a reason for hiding this comment

avik-pal commented Oct 5, 2024

wsmoses commented Oct 5, 2024

mofeing commented Oct 5, 2024 • edited Loading

avik-pal commented Oct 5, 2024

wsmoses commented Oct 5, 2024

wsmoses commented Oct 5, 2024

avik-pal commented Oct 5, 2024

codecov-commenter commented Oct 5, 2024

Codecov Report

wsmoses left a comment

Choose a reason for hiding this comment

mofeing commented Oct 6, 2024 • edited Loading

mofeing commented Oct 6, 2024

wsmoses commented Oct 6, 2024 via email

mofeing commented Oct 6, 2024 • edited Loading

wsmoses commented Oct 6, 2024

avik-pal commented Oct 6, 2024

avik-pal commented Oct 6, 2024

feat: implement a separate `TracedRNumber` #161

feat: implement a separate `TracedRNumber` #161

avik-pal commented Oct 4, 2024 •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot commented Oct 5, 2024 •

edited

Loading

mofeing commented Oct 5, 2024 •

edited

Loading

mofeing commented Oct 6, 2024 •

edited

Loading

mofeing commented Oct 6, 2024 •

edited

Loading