Skip to content

perf: customize cublastLt algo for Llamba 3.3 70B TP4 #6315

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 1 commit into from
Jul 24, 2025
+4 −1
Merged

perf: customize cublastLt algo for Llamba 3.3 70B TP4 #6315

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
2 changes: 1 addition & 1 deletion .clangd
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@ CompileFlags:
# Tweak the clangd parse settings for all files
CompileFlags:
Compiler: clang++
CompilationDatabase: .
CompilationDatabase: cpp/build
Add:
# report all errors
- "-ferror-limit=0"
Expand Down
3 changes: 3 additions & 0 deletions cpp/tensorrt_llm/thop/cublasScaledMM.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,6 +66,9 @@ AlgoListType fp8_algo_list = {
{{8, 8192, 8192}, {393, 36, 1, 0, 0, 5, 2}},
// [-algo66 -m_tile10 -m_stages36 -m_numsK1 -m_reduction0 -m_swizzle0 -m_custom1 -m_mma0 -m_cga2 -m_scheduling1]
{{8, 8192, 57344}, {10, 36, 1, 0, 0, 1, 2}},
// Llama-3.3-70B TP4 (this is the default algo on B200. Here we aim to use the same algo on GB200.)
// [-algo66 -m_tile393 -m_stages36 -m_numsK1 -m_reduction0 -m_swizzle0 -m_custom1 -m_mma0 -m_cga4 -m_scheduling1]
{{8, 8192, 14336}, {393, 36, 1, 0, 1, 1, 4}},
};

void set_algo_attr(cublasLtMatmulAlgo_t& algo, std::array<int, 7> const& attr_list)
Expand Down