refactor: separate the weight loading in the npu layer class. #489
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Clement-Wang26
force-pushed
the
final_refatcor
branch
from
eef8bfc to
aaa9676
Compare
Clement-Wang26
changed the title
XuZhang99
reviewed
Dec 5, 2025
| int weight_position) { | ||
| for (const auto& [name, tensor] : state_dict) { | ||
| if (absl::EndsWith(name, tensor_name)) { | ||
| at::Tensor mutable_tensor = tensor; |
Collaborator
There was a problem hiding this comment.
change all at:: to torch::.
|
|
||
| void BaseLoader::set_weight(const StateDict& state_dict, | ||
| const std::string& tensor_name, | ||
| int weight_position, |
Collaborator
There was a problem hiding this comment.
clarify whether it is int32_t or int64_t.
| } | ||
|
|
||
| torch::Dtype BaseLoader::string2dtype(const std::string& dtype_str) { | ||
| if (dtype_str.compare("float16") == 0) { |
|
|
||
| if (tensor.dtype() != torch::kInt8 && tensor.dtype() != torch::kInt32 && | ||
| tensor.dtype() != torch::kInt64) { | ||
| torch::Dtype dtype = string2dtype(torch_dtype_); |
Collaborator
There was a problem hiding this comment.
replace torch::Dtype with torch::ScalarType.
| const ModelContext& context) | ||
| : BaseLoader(weight_count, context) { | ||
| auto options = context.get_tensor_options(); | ||
| dtype_ = c10::typeMetaToScalarType(options.dtype()); |
| namespace layer { | ||
| class ColumParallelLinearLoader : public BaseLoader { | ||
| public: | ||
| explicit ColumParallelLinearLoader(uint64_t weight_count, |
Collaborator
There was a problem hiding this comment.
no need to add explicit when construct func has two params.
|
|
||
| void ColumParallelLinearLoader::verify_loaded_weights( | ||
| const std::string& weight_str) const { | ||
| CHECK(at_weight_tensors_[0].sizes() != std::vector<int64_t>({1})) |
| #include <torch_npu/csrc/core/npu/NPUCachingAllocator.h> | ||
| #include <torch_npu/csrc/core/npu/NPUException.h> | ||
|
|
||
| #include <map> |
Collaborator
There was a problem hiding this comment.
try # include <unordered_map>
Collaborator
There was a problem hiding this comment.
maybe we can move BaseLayer to npu dir or merge BaseLayer and NpuBaseLayer, because no other platform will use BaseLayer.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.