[SWAP] Implement inference mode #2696
Conversation
lhs8928
commented
Aug 5, 2024
lhs8928
commented
- This pr is simply rebase [Wait for #2286] [SWAP] Implement inference mode #2300 to current main. Please refer to [Wait for #2286] [SWAP] Implement inference mode #2300 for more details.
lhs8928
requested review from
myungjoo,
jijoongmoon,
again4you,
jaeyun-jung,
leemgs,
wooksong,
helloahn,
kparichay,
gichan-jang,
anyj0527,
zhoonit,
songgot,
jihochu,
DonghakPark,
SeoHyungjun,
baek2sm,
skykongkong8,
djeong20,
EunjuYang and
a team
as code owners
|
📝 TAOS-CI Version: 1.5.20200925. Thank you for submitting PR #2696. Please a submit 1commit/1PR (one commit per one PR) policy to get comments quickly from reviewers. Your PR must pass all verificiation processes of cibot before starting a review process from reviewers. If you are new member to join this project, please read manuals in documentation folder and wiki page. In order to monitor a progress status of your PR in more detail, visit http://ci.nnstreamer.ai/. |
lhs8928
force-pushed
the
jihochu_picogpt_swap
branch
from
9163d60 to
1e1e782
Compare
|
@DonghakPark Can you please check the clang-format version issue? |
Signed-off-by: Jiho Chu <[email protected]>
Signed-off-by: Jiho Chu <[email protected]>
This patch is for inference mode for swap device. It re-enable mmap feature, but writing time is controlled manually, due to the inference mode handling. Signed-off-by: Jiho Chu <[email protected]>
lhs8928
force-pushed
the
jihochu_picogpt_swap
branch
from
1e1e782 to
42c1e8b
Compare
| /** | ||
| * @brief Constructor | ||
| * | ||
| * @param value value to set, defaults to current directory |
There was a problem hiding this comment.
| * @param value value to set, defaults to current directory | |
| * @param value value to set, defaults to "train" mode |
| TensorLifespan var_ls = TensorLifespan::MAX_LIFESPAN; | ||
| TensorLifespan var_ls = swap_mode == "inference" | ||
| ? TensorLifespan::FORWARD_INFER_LIFESPAN | ||
| : TensorLifespan::MAX_LIFESPAN; | ||
| TensorLifespan grad_ls = TensorLifespan::BACKWARD_FUNC_LIFESPAN; |
There was a problem hiding this comment.
This is a pure question (it may be irrelevant to this PR, though).
Do we need grad_ls for the inference swap mode? Is it related to the loss thing?
| @@ -88,7 +89,7 @@ void *SwapDevice::getBuffer(off_t offset, size_t size, bool alloc_only) { | |||
| << "SwapDevice: seek file: " << dev_path; | |||
|
|
|||
| len = read(fd, ptr, size); | |||
| NNTR_THROW_IF(len != (ssize_t)size, std::runtime_error) | |||
| NNTR_THROW_IF(len != (size_t)size, std::runtime_error) | |||
There was a problem hiding this comment.
do we need to cast 'ssize_t' to 'size_t'?
because it looks 'read func' returns 'ssize_t' type.
