happyz
/
gemma.cpp
mirror of https://github.com/google/gemma.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Assorted small cleanups. 

PiperOrigin-RevId: 720548132
This commit is contained in:
Daniel Keysers 2025-01-28 06:09:08 -08:00 committed by Copybara-Service
parent a248f76245
commit bcdb0d65bd
4 changed files with 5 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
compression/compress.h
View File

@ -629,7 +629,7 @@ class ReadFromBlobStore {
// reader_ invalid or any Enqueue failed // reader_ invalid or any Enqueue failed
if (err_ != 0) return err_; if (err_ != 0) return err_;
// Setup the model_memory. // Setup the model_memory.
for (int b = 0; b < model_toc_.size(); ++b) { for (size_t b = 0; b < model_toc_.size(); ++b) {
const std::string& file_key = file_keys_[b]; const std::string& file_key = file_keys_[b];
MatPtr* blob = model_toc_[b]; MatPtr* blob = model_toc_[b];
if (!file_toc_.Empty()) { if (!file_toc_.Empty()) {

6
compression/python/compression_clif_aux.cc
View File

@ -139,8 +139,9 @@ class SbsWriterImpl : public WriterInterface {
void AddTokenizer(const std::string& tokenizer_path) override { void AddTokenizer(const std::string& tokenizer_path) override {
Path path(tokenizer_path); Path path(tokenizer_path);
GemmaTokenizer tokenizer(path); GemmaTokenizer tokenizer(path);
tokenizer_proto_ = tokenizer.Serialize(); std::string tokenizer_proto = tokenizer.Serialize();
compressor_.AddTokenizer(tokenizer_proto_); HWY_ASSERT(!tokenizer_proto.empty());
compressor_.AddTokenizer(tokenizer_proto);
} }
// Returns the number of blobs added. // Returns the number of blobs added.
@ -159,7 +160,6 @@ class SbsWriterImpl : public WriterInterface {
std::vector<MatStorage> model_memory_; std::vector<MatStorage> model_memory_;
std::vector<float> scales_; std::vector<float> scales_;
CompressorMode mode_; CompressorMode mode_;
std::string tokenizer_proto_;
}; };
WriterInterface* NewSbsWriter(CompressorMode mode) { WriterInterface* NewSbsWriter(CompressorMode mode) {

2
gemma/gemma-inl.h
View File

@ -690,7 +690,6 @@ HWY_NOINLINE void FFWNoVit(Activations& activations, size_t num_interleaved,
PROFILER_ZONE("Gen.FFW"); PROFILER_ZONE("Gen.FFW");
const size_t model_dim = layer_weights->layer_config.model_dim; const size_t model_dim = layer_weights->layer_config.model_dim;
const size_t ffh_hidden_dim = layer_weights->layer_config.ff_hidden_dim; const size_t ffh_hidden_dim = layer_weights->layer_config.ff_hidden_dim;
using WeightType = T;
HWY_DASSERT(num_interleaved <= activations.bf_pre_ffw_rms_out.BatchSize()); HWY_DASSERT(num_interleaved <= activations.bf_pre_ffw_rms_out.BatchSize());
const bool add_bias = layer_weights->layer_config.ff_biases; const bool add_bias = layer_weights->layer_config.ff_biases;
@ -746,7 +745,6 @@ HWY_NOINLINE void FFWVit(Activations& activations, size_t num_interleaved,
const LayerWeightsPtrs<T>* layer_weights) { const LayerWeightsPtrs<T>* layer_weights) {
PROFILER_ZONE("Gen.FFW"); PROFILER_ZONE("Gen.FFW");
const size_t ff_hidden_dim = layer_weights->layer_config.ff_hidden_dim; const size_t ff_hidden_dim = layer_weights->layer_config.ff_hidden_dim;
using WeightType = typename LayerWeightsPtrs<T>::WeightF32OrBF16;
HWY_DASSERT(num_interleaved <= activations.bf_pre_ffw_rms_out.BatchSize()); HWY_DASSERT(num_interleaved <= activations.bf_pre_ffw_rms_out.BatchSize());
const bool add_bias = layer_weights->layer_config.ff_biases; const bool add_bias = layer_weights->layer_config.ff_biases;

2
gemma/weights.cc
View File

@ -130,7 +130,7 @@ BlobError ModelWeightsStorage::Save(const std::string& tokenizer,
writer.AddTokenizer(tokenizer); writer.AddTokenizer(tokenizer);
int err = writer.WriteAll(weights, &config_); int err = writer.WriteAll(weights, &config_);
if (err != 0) { if (err != 0) {
fprintf(stderr, "Failed to load model weights: %d\n", err); fprintf(stderr, "Failed to write model weights: %d\n", err);
return err; return err;
} }
return 0; return 0;