Megvii Engine Team
|
d811dc5478
|
docs(mge/distributed): add document for distributed.backend
GitOrigin-RevId: 6cdcf7af77
|
3 years ago |
Megvii Engine Team
|
9526ee521b
|
docs(distributed.functional): add return type for all_reduce_min
GitOrigin-RevId: 9f734902fe
|
3 years ago |
Megvii Engine Team
|
f2ac4c345b
|
docs(distributed.functional.all_reduce_sum): googlestring and examples
GitOrigin-RevId: a456dfde24
|
3 years ago |
Megvii Engine Team
|
1f0436967c
|
refactor(mge/distributed): using nccl as default in distributed training
GitOrigin-RevId: 81268e84bc
|
3 years ago |
Megvii Engine Team
|
b17a02d44a
|
feat(mge/distributed): deprecate get_device_count_by_fork
GitOrigin-RevId: 6147c3ae90
|
3 years ago |
Megvii Engine Team
|
207a346351
|
chore(mge): run get_device_count("gpu") in subprocess
GitOrigin-RevId: 0f0dc001cf
|
4 years ago |
Megvii Engine Team
|
551cc701c6
|
docs(distributed.functional): add return type for all_reduce_max (jira #MGE-2706)
GitOrigin-RevId: a29f1f1880
|
3 years ago |
Megvii Engine Team
|
497ef6c337
|
fix(mge/dist): fix gl oom error
GitOrigin-RevId: 4ba3d2cfd7
|
3 years ago |
Megvii Engine Team
|
df79334cae
|
feat(mge/distributed): add user_pop function to save device memory
BREAKING CHANGE:
GitOrigin-RevId: 0a8e406da5
|
3 years ago |
Megvii Engine Team
|
809d5056cd
|
feat(mge/distributed): enable pt shm allreduce
GitOrigin-RevId: 1dd5a02a51
|
4 years ago |
Megvii Engine Team
|
933dd9a497
|
feat(mge/distributed): add cuda env check before forked thread
style(core/comp_node): reformat code
GitOrigin-RevId: 372452a8eb
|
4 years ago |
Megvii Engine Team
|
aa4e847692
|
fix(interpreter): release gil when interpreter blocking
GitOrigin-RevId: c48e9efa5b
|
4 years ago |
Megvii Engine Team
|
90d3905799
|
feat(mge): add mge._exit
GitOrigin-RevId: e7f348a677
|
4 years ago |
Megvii Engine Team
|
0cf4ff702e
|
perf(dist): add fastpath for bcast params
GitOrigin-RevId: aa40b3cd72
|
4 years ago |
Megvii Engine Team
|
a7581986bc
|
perf(dist): speed up bcast_val
GitOrigin-RevId: 21c4123b09
|
4 years ago |
Megvii Engine Team
|
aea5ac1363
|
fix(mge/distributed): fix gather scatter reduce broadcast autodiff
GitOrigin-RevId: 1c2250a079
|
4 years ago |
Megvii Engine Team
|
13e6ea349d
|
feat(imperative/opr): rebase rng refactoring to dev & add python module
GitOrigin-RevId: ee5984c52d
|
4 years ago |
Megvii Engine Team
|
1a7112997c
|
feat(opr-mm): add backend argument for remote send/recv
GitOrigin-RevId: 841a0e45ab
|
4 years ago |
Megvii Engine Team
|
dee5a10af9
|
feat(distributed): auto detect device and backend when init group
GitOrigin-RevId: 90be2d5b4d
|
4 years ago |
Megvii Engine Team
|
dcff115ef8
|
fix(distributed/launcher): fetch data early to avoid dead lock in launcher
GitOrigin-RevId: 9abcc956ef
|
4 years ago |
Megvii Engine Team
|
dcfb6a537e
|
refactor(mge/functional): move functional api
GitOrigin-RevId: 9cd3e09996
|
4 years ago |
Megvii Engine Team
|
fa4bf16800
|
feat(mge/functional): add repeat and tile opr
GitOrigin-RevId: a20d4b6fb0
|
4 years ago |
Megvii Engine Team
|
4f3875eb4f
|
docs(mge): add missing docstring and fix sphinx build warnings
GitOrigin-RevId: 4ce73cfd80
|
4 years ago |
Megvii Engine Team
|
17323dbd73
|
feat(dist): collect return values in dist.launcher
GitOrigin-RevId: 519e768ce9
|
4 years ago |
Megvii Engine Team
|
364afec033
|
chore(mge): update copyright years
GitOrigin-RevId: 3c0690bcc1
|
4 years ago |
Megvii Engine Team
|
10d9d77e41
|
refactor(mge/distributed): remove get_free_ports func
GitOrigin-RevId: b85c4885be
|
4 years ago |
Megvii Engine Team
|
0adf49b137
|
fix(mge/distributed): fix deadlock by mixing thread and fork
GitOrigin-RevId: c138cb9c28
|
4 years ago |
Megvii Engine Team
|
ea8eb4cf72
|
feat(mge/distributed): scalar support for distributed functions
GitOrigin-RevId: 53f3575baf
|
4 years ago |
Megvii Engine Team
|
d3bfb0e983
|
fix(mge): fix trace exit code and reformat
GitOrigin-RevId: 145c06b7e7
|
4 years ago |
Megvii Engine Team
|
de0742be25
|
refactor(mge): reopen passed assertions
GitOrigin-RevId: e0276e73e3
|
4 years ago |
Megvii Engine Team
|
a90c937d22
|
feat(interpreter): add command buffer for inplace
GitOrigin-RevId: 020d1e88d4
|
4 years ago |
Megvii Engine Team
|
75129cf071
|
chore(mge): clean up before merge to dev
* remove dead test
* clean some codes
* fix test_fake_quant due to change of implementation
GitOrigin-RevId: f030a9966d
|
4 years ago |
Megvii Engine Team
|
b9762d714c
|
fix(mge): make parampack run with tensor symbolic shape
GitOrigin-RevId: 6fc313785d
|
4 years ago |
Megvii Engine Team
|
4d75f691a0
|
feat(mge): restore remote send/recv
GitOrigin-RevId: 8b78fd5591
|
4 years ago |
Megvii Engine Team
|
9c92701f63
|
feat(mge): support python -m megengine.distributed.server
GitOrigin-RevId: f1e5c8e3cf
|
4 years ago |
Megvii Engine Team
|
60c7d62a4c
|
refactor(imperative): remove multidispatch, raw_tensor, register
GitOrigin-RevId: ca5a6ed8eb
|
4 years ago |
Megvii Engine Team
|
c294b9d18b
|
refactor(mge/tensor): remove old implementation
remove core.tensor, raw_tensor,TensorWrapper
avoid create tensor with zero-stride numpy ndarray
GitOrigin-RevId: 4fe5c4c5ba
|
4 years ago |
Megvii Engine Team
|
147cef52a0
|
refactor(mge/imperative): implement new tensor system
GitOrigin-RevId: 2dd4e460ac
|
4 years ago |
Megvii Engine Team
|
9d928e7f83
|
refactor(mge/distributed): sync interpreter for distribtued launcher
GitOrigin-RevId: 8a88c272a1
|
4 years ago |
Megvii Engine Team
|
69e3e32240
|
feat(imperative): auto generated opdef header and python binding
GitOrigin-RevId: d2f22ad5fe
|
4 years ago |
Megvii Engine Team
|
b9c37112a2
|
refactor(mge/distributed): skip barrier when running with single node
GitOrigin-RevId: 6cd5a2e1e2
|
4 years ago |
Megvii Engine Team
|
ee4ea7fdc8
|
test(distributed/test): make distributed test more stronger
GitOrigin-RevId: 085fd1dcfd
|
4 years ago |
Megvii Engine Team
|
3ecded74ea
|
refactor(distributed/server): use port 0 to get available port
GitOrigin-RevId: e367846b92
|
4 years ago |
Megvii Engine Team
|
638ab52fdc
|
feat(mge/imperative): simulates scalar
GitOrigin-RevId: e81630e256
|
4 years ago |
Megvii Engine Team
|
6667100638
|
feat(mge): use weakref for GradManger.attach
GitOrigin-RevId: 6df336c3c1
|
4 years ago |
Megvii Engine Team
|
b9918c329d
|
feat(mge/distributed): support distributed key-value store
GitOrigin-RevId: b4abe80014
|
4 years ago |
Megvii Engine Team
|
495472954d
|
fix(trace): link io-op to avoid deadlock
GitOrigin-RevId: 872cb6b715
|
4 years ago |
Megvii Engine Team
|
094601e834
|
feat(mge/distributed): allow remote grad by using grad manager
GitOrigin-RevId: a890c206a5
|
4 years ago |
Megvii Engine Team
|
02438ee635
|
refactor(mge/distributed): use thread.Threading to create Server
GitOrigin-RevId: fc994411bf
|
4 years ago |
Megvii Engine Team
|
a1efd4d56d
|
fix(mge/parampack): fix param pack when no param left
GitOrigin-RevId: b88b876064
|
4 years ago |