Merge pull request #58 from google:dev-cleanup

PiperOrigin-RevId: 610942948

.clang-format

@@ -0,0 +1 @@
+BasedOnStyle: Google

DEVELOPERS.md

@@ -71,6 +71,18 @@ The implementation code is roughly split into 4 layers, from high to low level:
 4.  Backend (`highway`) - Low-level hardware interface (SIMD in the case of
     highway) supporting the implementations in (3).
 
+Besides these layers, supporting utilities are:
+
+- `compression/` - model compression operations. The 8-bit switched floating
+  point model conversion is here.
+- `util/` - command line argument handling and any other utilities.
+
+## Style and Formatting
+
+A `.clang-format` configuration is provided with our defaults, please run source
+files through `clang-format` (or a formatter that produces equivalent behavior)
+before finalizing PR for submission.
+
 ## Compile-Time Flags (Advanced)
 
 There are several compile-time flags to be aware of (note these may or may not

gemma.h

@@ -26,15 +26,19 @@
 
 // copybara:import_next_line:gemma_cpp
 #include "compression/compress.h"  // SfpStream/NuqStream
+// copybara:end
 // copybara:import_next_line:gemma_cpp
 #include "configs.h"  // kSeqLen
+// copybara:end
 // copybara:import_next_line:gemma_cpp
 #include "util/args.h"  // ArgsBase
+// copybara:end
 #include "hwy/aligned_allocator.h"
 #include "hwy/base.h"  // hwy::bfloat16_t
 #include "hwy/contrib/thread_pool/thread_pool.h"
 // copybara:import_next_line:sentencepiece
 #include "src/sentencepiece_processor.h"
+// copybara:end
 
 namespace gcpp {
 
@@ -115,7 +119,8 @@ struct LoaderArgs : public ArgsBase<LoaderArgs> {
   Path cache;  // compressed weights
   std::string model_type;
 
-  template <class Visitor> void ForEach(const Visitor &visitor) {
+  template <class Visitor>
+  void ForEach(const Visitor& visitor) {
     visitor(tokenizer, "tokenizer", Path(),
             "Path name of tokenizer model file. (required)");
     visitor(
@@ -175,7 +180,8 @@ struct InferenceArgs : public ArgsBase<InferenceArgs> {
     return nullptr;
   }
 
-  template <class Visitor> void ForEach(const Visitor &visitor) {
+  template <class Visitor>
+  void ForEach(const Visitor& visitor) {
     visitor(max_tokens, "max_tokens", size_t{3072},
             "Maximum number of tokens in prompt + generation.");
     visitor(max_generated_tokens, "max_generated_tokens", size_t{2048},
@@ -186,7 +192,8 @@ struct InferenceArgs : public ArgsBase<InferenceArgs> {
             "Make top-k sampling deterministic", 2);
     visitor(multiturn, "multiturn", false,
             "Multiturn mode (if 0, this clears the KV cache after every "
-            "interaction without quitting)\n    Default = 0 (conversation resets every turn)");
+            "interaction without quitting)\n    Default = 0 (conversation "
+            "resets every turn)");
   }
 };