Skip to content

Better spreading of work for tiled attention #948

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

Open
copybara-service wants to merge 1 commit into
base: dev
Choose a base branch
from
+167 −120
Open

Better spreading of work for tiled attention #948

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
14 changes: 14 additions & 0 deletions gemma/gemma_args.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -211,6 +211,7 @@ struct InferenceArgs : public ArgsBase<InferenceArgs> {
Path prompt_file;
std::string eot_line;
std::string attention_impl;
std::string kv_cache_type;

template <class Visitor>
void ForEach(const Visitor& visitor) {
Expand Down Expand Up @@ -266,6 +267,10 @@ struct InferenceArgs : public ArgsBase<InferenceArgs> {
2);
visitor(attention_impl, "attention_impl", std::string("flash"),
"Attention implementation to use. See configs.cc for options.", 2);
visitor(kv_cache_type, "kv_cache_type", std::string(""),
"KV cache data type (f32, bf16, int8). If empty, deduced from "
"attention_impl.",
2);
}

void CopyTo(RuntimeConfig& runtime_config) const {
Expand All @@ -288,6 +293,15 @@ struct InferenceArgs : public ArgsBase<InferenceArgs> {
runtime_config.temperature = temperature;
runtime_config.top_k = top_k;
runtime_config.attention_impl = GetAttentionImpl(attention_impl);
if (!kv_cache_type.empty()) {
if (kv_cache_type == "int8" || kv_cache_type == "i8") {
runtime_config.kv_cache_type = Type::kInt8;
} else if (kv_cache_type == "bf16") {
runtime_config.kv_cache_type = Type::kBF16;
} else if (kv_cache_type == "f32" || kv_cache_type == "float") {
runtime_config.kv_cache_type = Type::kF32;
}
}
}
};

Expand Down
Loading
Loading