Skip to content

Pass runtime_config_ to the KVCache constructor within GemmaEnv in benchmark_helper.cc. This ensures that kv_cache_type and attention_impl settings are properly propagated when using non-default attention implementations such as flash_transposed_qs_bf16. #956

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
copybara-service merged 1 commit into from
Jul 17, 2026
+6 −4
Merged

Pass runtime_config_ to the KVCache constructor within GemmaEnv in benchmark_helper.cc. This ensures that kv_cache_type and attention_impl settings are properly propagated when using non-default attention implementations such as flash_transposed_qs_bf16. #956

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
10 changes: 6 additions & 4 deletions evals/benchmark_helper.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,8 +42,6 @@ GemmaEnv::GemmaEnv(const GemmaArgs& args)
env_(ctx_),
gemma_(args, ctx_) {
const ModelConfig& config = gemma_.Config();
// Only allocate one for starters because GenerateBatch might not be called.
kv_caches_.push_back(KVCache(config, args.inference, ctx_.allocator));

if (args.inference.verbosity >= 2) {
ShowConfig(args, config, gemma_.WeightReadMode(), ctx_);
Expand All @@ -57,6 +55,10 @@ GemmaEnv::GemmaEnv(const GemmaArgs& args)
.verbosity = args.inference.verbosity,
};
args.inference.CopyTo(runtime_config_);

// Only allocate one for starters because GenerateBatch might not be called.
kv_caches_.push_back(
KVCache(config, args.inference, runtime_config_, ctx_.allocator));
}

QueryResult GemmaEnv::QueryModel(const std::vector<int>& tokens) {
Expand Down Expand Up @@ -130,8 +132,8 @@ QueryResultAndMetrics GemmaEnv::BatchQueryModelWithMetrics(

// Ensure we have at least one KVCache per query.
while (kv_caches_.size() < num_queries) {
kv_caches_.push_back(
KVCache(gemma_.Config(), gemma_.Inference(), ctx_.allocator));
kv_caches_.push_back(KVCache(gemma_.Config(), gemma_.Inference(),
runtime_config_, ctx_.allocator));
}
const hwy::Span<KVCache> kv_caches(&kv_caches_[0], num_queries);

Expand Down
Loading