Merge pull request #102 from google:experimental

PiperOrigin-RevId: 616882521

DEVELOPERS.md

@@ -118,8 +118,7 @@ jax / pytorch / keras for NN deployments.
 
 ### Gemma struct contains all the state of the inference engine - tokenizer, weights, and activations
 
-`Gemma(...)` - constructor, creates a gemma model object, which is a wrapper
+`Gemma(...)` - constructor, creates a gemma model object. 
-around 3 things - the tokenizer object, weights, activations, and KV Cache.
 
 In a standard LLM chat app, you'll probably use a Gemma object directly, in
 more exotic data processing or research applications, you might decompose
@@ -129,11 +128,13 @@ only using a Gemma object.
 
 ### Use the tokenizer in the Gemma object (or interact with the Tokenizer object directly)
 
-You pretty much only do things with the tokenizer, call `Encode()` to go from
+The Gemma object contains contains a pointer to a Tokenizer object. The main
-string prompts to token id vectors, or `Decode()` to go from token id vector
+operations performed on the tokenizer are to load the tokenizer model from a
-outputs from the model back to strings.
+file (usually `tokenizer.spm`), call `Encode()` to go from string prompts to
+token id vectors, or `Decode()` to go from token id vector outputs from the
+model back to strings.
 
-### The main entrypoint for generation is `GenerateGemma()`
+### `GenerateGemma()` is the entrypoint for token generation
 
 Calling into `GenerateGemma` with a tokenized prompt will 1) mutate the
 activation values in `model` and 2) invoke StreamFunc - a lambda callback for
@@ -150,7 +151,7 @@ constrained decoding type of use cases where you want to force the generation
 to fit a grammar. If you're not doing this, you can send an empty lambda as a
 no-op which is what `run.cc` does.
 
-### If you want to invoke the neural network forward function directly call the `Transformer()` function
+### `Transformer()` implements the inference (i.e. `forward()` method in PyTorch or Jax) computation of the neural network
 
 For high-level applications, you might only call `GenerateGemma()` and never
 interact directly with the neural network, but if you're doing something a bit

README.md

@@ -36,7 +36,12 @@ For production-oriented edge deployments we recommend standard deployment
 pathways using Python frameworks like JAX, Keras, PyTorch, and Transformers
 ([all model variations here](https://www.kaggle.com/models/google/gemma)).
 
-Community contributions large and small are welcome. This project follows
+## Contributing
+
+Community contributions large and small are welcome. See
+[DEVELOPERS.md](https://github.com/google/gemma.cpp/blob/main/DEVELOPERS.md)
+for additional notes contributing developers and [join the discord by following
+this invite link](https://discord.gg/H5jCBAWxAe). This project follows
 [Google's Open Source Community
 Guidelines](https://opensource.google.com/conduct/).
 

examples/README.md

@@ -0,0 +1,7 @@
+# Examples
+
+In this directory are some simple examples illustrating usage of `gemma.cpp` as
+a library beyond the interactive `gemma` app implemented in `run.cc`.
+
+- `hello_world/` - minimal/template project for using `gemma.cpp` as a library.
+  It sets up the model state and generates text for a single hard coded prompt.

experimental/README.md

@@ -0,0 +1,3 @@
+# Experimental
+
+This directory is for experimental code and features.