8000 DataParallel should support multiple inputs · Issue #649 · pytorch/pytorch · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

DataParallel should support multiple inputs #649

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
apaszke opened this issue Jan 30, 2017 · 7 comments
Closed
8000

DataParallel should support multiple inputs #649

apaszke opened this issue Jan 30, 2017 · 7 comments
Assignees
@apaszke

Comments

@apaszke
Copy link
Contributor
apaszke commented Jan 30, 2017

No description provided.

@everwind
Copy link
everwind commented Feb 6, 2017

I implement the data_parallel with two inputs, but it does not work

def data_parallel2(module, input1, input2, device_ids, output_device=None):
"""Evaluates module(input) in parallel across the GPUs given in device_ids.

This is the functional version of the DataParallel module. 

Args:   
    module: the module to evaluate in parallel
    input: input to the module
    device_ids: GPU ids on which to replicate module
    output_device: GPU location of the output  Use -1 to indicate the CPU.
        (default: device_ids[0])
Returns:
    a Variable containing the result of module(input) located on
    output_device
"""
if not device_ids:
    return module(input1, input2) 

if output_device is None:
    output_device = device_ids[0]

replicas = replicate(module, device_ids)
input1s = scatter(input1, device_ids)
input2s = scatter(input2, device_ids)
replicas = replicas[:len(input1s)]
outputs = parallel_apply2(replicas, input1s, input2s)
return gather(outputs, output_device)

def parallel_apply2(modules, input1s, input2s):
assert len(modules) == len(input1s)
# Fast track
if len(modules) == 1:
return (modules[0](input1s[0], input2s[0]),)

lock = threading.Lock()
results = {}

def _worker(module, input1, input2, results, lock):
    var_input1 = input1
    var_input2 = input2
    while not isinstance(var_input1, Variable):
        var_input1 = var_input1[0]
    while not isinstance(var_input2, Variable):
        var_input2 = var_input2[0]
    try:    
        with torch.cuda.device_of(var_input1):
            output = module(input1, input2) 
        with lock:
            results[input1] = output
    except Exception as e:
        with lock:
            results[input1] = e

threads = [threading.Thread(target=_worker,
                            args=(module, input1, input2, results, lock))
            for module, input1, input2 in zip(modules, input1s, input2s)]

@soumith
Copy link
Member
soumith commented Feb 6, 2017

wdym by "it does not work"

@everwind
Copy link
everwind commented Feb 6, 2017

the error info:
Traceback (most recent call last):
File "translate.py", line 136, in
data_parallel2(model,batch_enc_inputs, batch_dec_inputs, device_ids=[0, 1] )
File "/data1/plat/peakzeng/workspace/seq2seq-pytorch-example-master/data_parallel.py", line 115, in data_parallel2
outputs = parallel_apply2(replicas, input1s, input2s)
File "/data1/plat/peakzeng/workspace/seq2seq-pytorch-example-master/parallel_apply.py", line 86, in parallel_apply2
raise output
RuntimeError: equal number of batches expected at /data/plat/peakzeng/solfware/pytorch/torch/lib/THC/generic/THCTensorMathBlas.cu:441

@soumith
Copy link
Member
soumith commented Feb 6, 2017 •
edited
Loading

looks like input1 and input2 don't have equal sizes maybe? either ways looks like an implementation bug on your side

@everwind
Copy link
everwind commented Feb 6, 2017 •
edited
Loading

yes, input1 and input2 don't have equal sizes, input1.size =( input_len1, batch_size, hidden_size) and the input2 .size() = (input_len2 , batch_size, hidden_size ) , if i make input1.size = (batch_size, hidden_size, input_len1) and the input2.size() = (batch_size, hidden_size, input_len2), can it works ?

@apaszke
Copy link
Contributor Author
apaszke commented Feb 6, 2017

Yes, it expects the batch dimension to be first.

@apaszke apaszke self-assigned this Feb 19, 2017
@apaszke apaszke mentioned this issue Feb 20, 2017
Closed
@soumith soumith closed this as completed Feb 21, 2017
@JianboTang
Copy link

@apaszke the second dimension should be equal? I encounter this problem while the first dimension is equal

Traceback (most recent call last):
File "/home/tangjianbo/anaconda2/lib/python2.7/pdb.py", line 1314, in main
pdb._runscript(mainpyfile)
File "/home/tangjianbo/anaconda2/lib/python2.7/pdb.py", line 1233, in _runscript
self.run(statement)
File "/home/tangjianbo/anaconda2/lib/python2.7/bdb.py", line 400, in run
exec cmd in globals, locals
File "", line 1, in
File "main_SimpleAoA.py", line 1, in
import argparse
File "main_SimpleAoA.py", line 352, in main
train_loss_epoch, train_prob_epoch = train(epoch)
File "main_SimpleAoA.py", line 300, in train
loss, prob = AoA(q1_input, q2_input, lb_input)
File "/home/tangjianbo/anaconda2/lib/python2.7/site-packages/torch/nn/modules/module.py", line 224, in call
result = self.forward(*input, **kwargs)
File "/home/tangjianbo/anaconda2/lib/python2.7/site-packages/torch/nn/parallel/data_parallel.py", line 60, in forward
outputs = self.parallel_apply(replicas, inputs, kwargs)
File "/home/tangjianbo/anaconda2/lib/python2.7/site-packages/torch/nn/parallel/data_parallel.py", line 70, in parallel_apply
return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)])
File "/home/tangjianbo/anaconda2/lib/python2.7/site-packages/torch/nn/parallel/parallel_apply.py", line 67, in parallel_apply
raise output
RuntimeError: invalid argument 7: equal number of batches expected at /pytorch/torch/lib/THC/generic/THCTensorMathBlas.cu:447
Uncaught exception. Entering post mortem debugging
Running 'cont' or 'step' will restart the program

/home/tangjianbo/anaconda2/lib/python2.7/site-packages/torch/nn/parallel/parallel_apply.py(67)parallel_apply()
-> raise output
(Pdb) up
/home/tangjianbo/anaconda2/lib/python2.7/site-packages/torch/nn/parallel/data_parallel.py(70)parallel_apply()
-> return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)])
(Pdb)
/home/tangjianbo/anaconda2/lib/python2.7/site-packages/torch/nn/parallel/data_parallel.py(60)forward()
-> outputs = self.parallel_apply(replicas, inputs, kwargs)
(Pdb)
/home/tangjianbo/anaconda2/lib/python2.7/site-packages/torch/nn/modules/module.py(224)call()
-> result = self.forward(*input, **kwargs)
(Pdb)
/data/tangjianbo/question_normalization/AoA/char/main_SimpleAoA.py(300)train()
-> loss, prob = AoA(q1_input, q2_input, lb_input)
(Pdb) q1_input.size()
torch.Size([64, 20, 128])
(Pdb) q2_input.size()
torch.Size([64, 38, 128])
(Pdb) lb_input.size()
torch.Size([64, 1])

mrshenli pushed a commit to mrshenli/pytorch that referenced this issue Apr 11, 2020
@SethHWeidman
Merge pull request pytorch#649 from zhangguanheng66/transformer_tutorial
f9d5be7 
[WIP] Transformer tutorial
jjsjann123 pushed a commit to jjsjann123/pytorch that referenced this issue Apr 11, 2021
@naoyam
TensorView cleanup (pytorch#649)
5661076 
Remove relative_compute_at_axis, getComputeAtRelPos, TensorView::compute_at_view_, Expose ComputeAtMap so that it can be used in the C++ tests
KyleCZH pushed a commit to KyleCZH/pytorch that referenced this issue Sep 20, 2021
@jeffdaily
manylinux/build_common.sh should also rewrite caffe2 deps (pytorch#649)
4990182
akashveramd pushed a commit to akashveramd/pytorch that referenced this issue Apr 9, 2025
@aska-0096
[Navi3x] Fix Gridwise_multiple_d operation (pytorch#649)
e5376be 
* Add CMake Option "USE_OPT_NAVI3X"

* fix bug
akashveramd pushed a commit to akashveramd/pytorch that referenced this issue Apr 9, 2025
@zjing14 @aska-0096
@samjwu @rocking5566 @geyyer @carlushuang @junliume @illsilin
Navi3 rel (pytorch#1176)
1837040 
* wmma_op + unit test

* add arch limitation to wmma test

* change arch limitation

* Refactor + Add all type unit test(int4 compile failed)

* Add f32_16x16x16_bf16 unit test

* tempsave

* tempsave

* tempsave

* runtime bug, cannot find symbol

* workaround for incorrect HIP warpSize return value

* debugging

* tempsave

* Correctness OK, waiting for optimization

* Tidy up + format

* temp save

* temp save, reproduce the v_bfi_b32 issue

* add inline asm for wmmaop test

* tidy up

* clean some debug purpose code

* discard some codes

* clang format

* clang format

* compiler issue fixed + increase tile size

* navi3x_multipleD+example

* temp save

* workable

* batchedgemm[OK], groupconv[debug]

* groupconv: Sanity check[OK], Performance[Bad]

* navi3x_groupconv_need_optimization

* create necessary files

* save progress

* Add Inter-Row thread transfer

* save progress

* save debugging progress

* sanity check pass

* fix a host tensor bug and clean up flash-attn code

* format

* cancel unnecessary change

* cancel unnecessary change

* cancel unnecessary change

* temp save, add asm backend flag to amd_wmma

* Mat-A LDS Bypass sanity pass

* temp save

* gemm sanity fix

* Porting new blockwise gemm to flash attention

* Example branch provide to compiler team

* tempsave

* Fix a bug

* batched gemm ported

* conv A-skip lds ported

* Skip B-Lds real gemm

* Skip B Lds Gemm + MulD

* batched gemm, conv, skip b lds

* format

* Attn, skip b lds

* Change GridwiseOp nam

* fix a typo caused bug

* Skip A_Lds sanity pass, Skip B_Lds scratch occured

* Bug found, intra-row permute off caused

* bug found

* a fix

* disable buffer load due to incorrect 3rd dword

* update fmha config, no scratch generated

* update 3rd dword

* fmha config update

* FMHA, add support to gfx1101/gfx1102

* Merge origin dev (pytorch#2)

* [Navi3x] Fix Gridwise_multiple_d operation (pytorch#649)

* Add CMake Option "USE_OPT_NAVI3X"

* fix bug

* standardize docs (pytorch#655)

* Separate bibtex requirement from rocm-docs-core (pytorch#656)

* separate bibtex requirement from rocm-docs-core

* point requirements to source rocm-docs-core repo

* Add CMake Option "USE_OPT_NAVI3X" (pytorch#647)

* Add CMake Option "USE_OPT_NAVI3X"

* remove navi3x opt compile option from cmake script

* Conv + quantization + tanh  (pytorch#645)

* Rename file. Prepare to support another activation

* Add comment for quantization

* Extract out_elementop

* Add tanh example

* Add conv + bias + tanh quantization instance

* Add missing parameter

* Refine cmake

* Add external api and client example

* Extract variable in example

* Fix the comment

---------

Co-authored-by: zjing14 <zhangjing14@gmail.com>

* Add a denorm test fix (pytorch#603)

* Add type_convert implementations for bf16

* Add the fix for conv_fwd

* Add the fix for conv_bwd_data

* Add the fix for conv_bwd_weight

* Format

* Format

* Another format

* Add a macro to use workaround on MI200 only

* Format

---------

Co-authored-by: Rosty Geyyer <rosty.geyyer@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>

* simplify karg in device/grid of split-k op (pytorch#644)

* simplify karg in device/grid split-k op

* fix mk_kn_mn instances

* add more instances

* use name from tensor layout

* fix 3rd dword of buffer source descriptor (pytorch#659)

* add fp64 instances (pytorch#658)

Co-authored-by: root <root@ctr-ubbsmc15.amd.com>

* Issue pytorch#666: Revert "simplify karg in device/grid of split-k op (pytorch#644)" (pytorch#665)

This reverts commit bb5530a.

* Groupnorm + swish external api (pytorch#668)

* Rename to proper naming

* Add example of groupnorm + swish

* Extract duplicate code in example

* Add groupnorm + swish instances

* Ractor instance generation, split into multiple cpp file

* Add external api and client example

* Refine profiler message

* Use ck math version of exp

* Refine problem size in example

* Add host version of exp

* add a marco to turn on/off denorm fix (off by default) (pytorch#673)

* add a marco to turn off denorm fix by default

* expose the marco

---------

Co-authored-by: root <root@ctr-ubbsmc15.amd.com>

* fixed quant example (pytorch#672)

Co-authored-by: root <root@ctr-ubbsmc15.amd.com>

* Add dependabot config and pin rocm-docs-core (pytorch#663)

* [gtest] suppress unsafe buffer warn (pytorch#670)

ref: ROCm/MIOpen#1912

* Add memory index guard in wmma device ops (pytorch#667)

* Add more macros to turn on/off denorm fix (pytorch#678)

Co-authored-by: Rosty Geyyer <rosty.geyyer@amd.com>

* Fix a typo (pytorch#676)

* Add (pytorch#677)

* Allow using ROCm release candidate compilers. (pytorch#679)

* enable use of rocm5.5 release candidate 4

* upgrade to ROCM5.5 RC5

* try fix the PUB_KEY error, remove the cmake-data package

* upgrade to latest cmake version

* use private dockerhub repo for rocm5.5 rc5

* add missing bracket

* add vector load check

* solve conflicts

---------

Co-authored-by: Sam Wu <sjwu@ualberta.ca>
Co-authored-by: Sam Wu <sam.wu2@amd.com>
Co-authored-by: rocking5566 <ChunYu.Lai@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>
Co-authored-by: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Co-authored-by: Rosty Geyyer <rosty.geyyer@amd.com>
Co-authored-by: carlushuang <carlus.huang@amd.com>
Co-authored-by: root <root@ctr-ubbsmc15.amd.com>
Co-authored-by: Jun Liu <Liu.Jun@amd.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>

* Disable SkipLDS & Align AIT api (pytorch#3)

* fix layernorm, reduction Ops (pytorch#4)

* [Navi3x] Fix Gridwise_multiple_d operation (pytorch#649)

* Add CMake Option "USE_OPT_NAVI3X"

* fix bug

* standardize docs (pytorch#655)

* Separate bibtex requirement from rocm-docs-core (pytorch#656)

* separate bibtex requirement from rocm-docs-core

* point requirements to source rocm-docs-core repo

* Add CMake Option "USE_OPT_NAVI3X" (pytorch#647)

* Add CMake Option "USE_OPT_NAVI3X"

* remove navi3x opt compile option from cmake script

* Conv + quantization + tanh  (pytorch#645)

* Rename file. Prepare to support another activation

* Add comment for quantization

* Extract out_elementop

* Add tanh example

* Add conv + bias + tanh quantization instance

* Add missing parameter

* Refine cmake

* Add external api and client example

* Extract variable in example

* Fix the comment

---------

Co-authored-by: zjing14 <zhangjing14@gmail.com>

* Add a denorm test fix (pytorch#603)

* Add type_convert implementations for bf16

* Add the fix for conv_fwd

* Add the fix for conv_bwd_data

* Add the fix for conv_bwd_weight

* Format

* Format

* Another format

* Add a macro to use workaround on MI200 only

* Format

---------

Co-authored-by: Rosty Geyyer <rosty.geyyer@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>

* simplify karg in device/grid of split-k op (pytorch#644)

* simplify karg in device/grid split-k op

* fix mk_kn_mn instances

* add more instances

* use name from tensor layout

* fix 3rd dword of buffer source descriptor (pytorch#659)

* add fp64 instances (pytorch#658)

Co-authored-by: root <root@ctr-ubbsmc15.amd.com>

* Issue pytorch#666: Revert "simplify karg in device/grid of split-k op (pytorch#644)" (pytorch#665)

This reverts commit bb5530a.

* Groupnorm + swish external api (pytorch#668)

* Rename to proper naming

* Add example of groupnorm + swish

* Extract duplicate code in example

* Add groupnorm + swish instances

* Ractor instance generation, split into multiple cpp file

* Add external api and client example

* Refine profiler message

* Use ck math version of exp

* Refine problem size in example

* Add host version of exp

* add a marco to turn on/off denorm fix (off by default) (pytorch#673)

* add a marco to turn off denorm fix by default

* expose the marco

---------

Co-authored-by: root <root@ctr-ubbsmc15.amd.com>

* fixed quant example (pytorch#672)

Co-authored-by: root <root@ctr-ubbsmc15.amd.com>

* Add dependabot config and pin rocm-docs-core (pytorch#663)

* [gtest] suppress unsafe buffer warn (pytorch#670)

ref: ROCm/MIOpen#1912

* Add memory index guard in wmma device ops (pytorch#667)

* Add more macros to turn on/off denorm fix (pytorch#678)

Co-authored-by: Rosty Geyyer <rosty.geyyer@amd.com>

* Fix a typo (pytorch#676)

* Add (pytorch#677)

* Allow using ROCm release candidate compilers. (pytorch#679)

* enable use of rocm5.5 release candidate 4

* upgrade to ROCM5.5 RC5

* try fix the PUB_KEY error, remove the cmake-data package

* upgrade to latest cmake version

* use private dockerhub repo for rocm5.5 rc5

* add missing bracket

* Disable SkipLDS & Align AIT api

* Update dependabot config (pytorch#682)

Co-authored-by: samjwu <samjwu@users.noreply.github.com>

* update attn api

* solve type_convert bug + enable

---------

Co-authored-by: Sam Wu <sjwu@ualberta.ca>
Co-authored-by: Sam Wu <sam.wu2@amd.com>
Co-authored-by: rocking5566 <ChunYu.Lai@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>
Co-authored-by: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Co-authored-by: Rosty Geyyer <rosty.geyyer@amd.com>
Co-authored-by: carlushuang <carlus.huang@amd.com>
Co-authored-by: root <root@ctr-ubbsmc15.amd.com>
Co-authored-by: Jun Liu <Liu.Jun@amd.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
Co-authored-by: samjwu <samjwu@users.noreply.github.com>
Co-authored-by: haocwang <Haocong.WANG@amd.com>

* fix typo

* Fix attention with causal mask

* multiple fix, try ait compile

* Add A/B not use LDS pipeline

* Clang format, Add gfx1101, gfx1102 support of FMHA example

* cancel change of format script

* 1. Enable 2-stage global Prefetch ( M
5D1D
ay cause VGPR spilling)
2. Enable FP16 accumulator blockwise_gemm

* clang-format

* 1. change blockwise gemm loopover direction from kmn to mnk ( ~1% improvement)
2. change kernel timing mode to 50 warmup + 50 timed repeat

* Update low level abstration of blockwise gemm wmma

* (2/5) bilinear gemm pass, perf bug: skip a lds has lower performance than skip b lds

* (3/5) batched gemm pass, perf bug: skip a lds has lower performance than skip b lds

* (4/5) grouped conv pass

* (5/5) attention pass, todo: debug lds perf bug

* AIT Attention API refactor (pytorch#8)

* sanity pass

* sanity pass 2

* confirm significant performance regression.

* turn on all instances

* turn off instance format

* Fix bug & tunning & format

* DML meta, self_attn+cross_attn

* sanity pass

* remove useless flag

* update tile and problem size used in AIT attention

* bug fix in grouped conv supporting check

* deprecate inline asm wmma

* Bug fix: double lds skip

* clang-format

* Fix errors in
1. example, fmha
2. gridwise pipeline
3. deviceop, fmha, change some containers from vector to array

* part2 of previous commit

* clang format

* API fix of gridwisegemmpipeline

* separate array base and vector base attention tensor transformation

* fix gemm

* clang format

* add gemm fp16 instances

* Temp save

* fpAintB kernel compile pass

* Sanity pass.

* Temp save

* debug code enabled

* Fp16AInt8B_GEMM sanity

* MQA implementation

* GQA-4 example

* tempsave

* Compile pass

* New implementation of fp16Aint8B Gemm, Acheieve similar math throughput with native fp16 Gemm

* format

* Todo: fix gemm_bilinear_wmma instances compilation bug

* Solve a bug when K1=16

* remove unnecessary changes

* Remove tensor layout limitation to LDS usage in tesnor contraction

* update self-attention and cross-attention

* fix a typo of name

* Add arch limiter for fp8 gemm

* enable fp8 gemm_xdl for all gfx9 targets

* temporarily disable gemm_xdl_fp16_fp8 on MI100/200

* fix the cmake logic for gemm_xdl_fp16_fp8

* re-enable the gemm_xdl_fp16_fp8 on MI100/200

---------

Co-authored-by: aska-0096 <haocwang@amd.com>
Co-authored-by: Sam Wu <sjwu@ualberta.ca>
Co-authored-by: Sam Wu <sam.wu2@amd.com>
Co-authored-by: rocking5566 <ChunYu.Lai@amd.com>
Co-authored-by: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Co-authored-by: Rosty Geyyer <rosty.geyyer@amd.com>
Co-authored-by: carlushuang <carlus.huang@amd.com>
Co-authored-by: root <root@ctr-ubbsmc15.amd.com>
Co-authored-by: Jun Liu <Liu.Jun@amd.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
Co-authored-by: samjwu <samjwu@users.noreply.github.com>
Co-authored-by: haocwang <Haocong.WANG@amd.com>
Co-authored-by: illsilin <Illia.Silin@amd.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@apaszke apaszke

Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
4 participants
@soumith @everwind @apaszke @JianboTang
0