Skip to content

[bugfix] fix gsa coredump #265

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

Merged
merged 3 commits into from
Sep 30, 2025
Merged

[bugfix] fix gsa coredump #265

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f41286d
[bugfix] fix gsa coredump
HaoLi980405 Sep 30, 2025
5d7df37
[bugfix] fix build_sparse_meta param lost
HaoLi980405 Sep 30, 2025
18133b8
clean code
HaoLi980405 Sep 30, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 2 additions & 7 deletions ucm/sparse/gsa/gsa.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -497,9 +497,7 @@ def init_topk_cal(
self.gsa_offload_ops = gsa_offload_ops.CalKpreAndTopk(
self.layer_num, block_size, MAX_BS, att_num_heads, head_size
)
self.gsa_offload_ops.set_kpre_method_param(
int(max_model_len / block_size) * MAX_BS, kv_num_heads, 1
)
self.gsa_offload_ops.set_kpre_method_param(kv_num_heads, 1)
self.gsa_offload_ops.set_kpre_cache(prefetch_engine.kpre_caches)
self.is_cal_kpre = [False] * self.layer_num
self.gsa_q_cache = torch.zeros(
Expand Down Expand Up @@ -868,10 +866,7 @@ def execute_finished(self):
)

def build_sparse_meta(
self,
scheduler_output: SchedulerOutput,
requests,
input_batch,
self, scheduler_output: SchedulerOutput, requests, input_batch, attn_metadata
) -> None:
self.gsa_metadata = self.build_gsa_metadata(
scheduler_output, requests, input_batch
Expand Down
2 changes: 1 addition & 1 deletion ucm/sparse/gsa/offload_ops/include/cal_kpre_and_topk.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,7 +46,7 @@ class __attribute__((visibility("hidden"))) CalKpreAndTopk
public:
CalKpreAndTopk(uint32_t layerNum, uint32_t blockSize, uint32_t maxBs, uint32_t numHeads, uint32_t headSize);
~CalKpreAndTopk();
void SetKpreMethodParam(uint32_t maxBlockNum, uint32_t numHeads, uint32_t numKpre);
void SetKpreMethodParam(uint32_t numHeads, uint32_t numKpre);
void SetKpreCache(std::vector<torch::Tensor>& kpreCache);
void SetTopkCache(std::vector<torch::Tensor>& topkCache, std::vector<uint32_t>& topkLens);
void SetCommonParam(std::vector<uint32_t>& calTopkIdx, std::vector<bool>& isDecode);
Expand Down
11 changes: 3 additions & 8 deletions ucm/sparse/gsa/offload_ops/src/cal_kpre_and_topk.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,15 +27,10 @@ CalKpreAndTopk::CalKpreAndTopk(uint32_t layerNum, uint32_t blockSize, uint32_t m
m_count = 0;
}

void CalKpreAndTopk::SetKpreMethodParam(uint32_t maxBlockNum, uint32_t numHeads, uint32_t numKpre)
void CalKpreAndTopk::SetKpreMethodParam(uint32_t numHeads, uint32_t numKpre)
{
// m_kNumHeads = numHeads;
// m_numKpre = numKpre;
// auto optionsForKCache = torch::TensorOptions().device("cpu").dtype(torch::kFloat32);
// for (uint32_t i = 0; i < m_layerNum; i++) {
// torch::Tensor layerKCache = torch::zeros({maxBlockNum, m_kNumHeads, m_blockSize, m_headSize}, optionsForKCache);
// m_kCache.push_back(layerKCache);
// }
m_kNumHeads = numHeads;
m_numKpre = numKpre;
}

void CalKpreAndTopk::SetKpreCache(std::vector<torch::Tensor>& kpreCache)
Expand Down