Multiplication is done using int16*int16 multiplication instructions avoid expensive conversion to f32/bf16
x2 speed on zen3
PiperOrigin-RevId: 888690192
Right now multiplication is done by converting to corresponding float format.
Can yield up to 2x improvements for membw constrained shapes
PiperOrigin-RevId: 880748493
It also supports better parallelism for small batch sizes / small models.
It also is able to utilize VDPBF16PS for nice 2x improvement on avx512
PiperOrigin-RevId: 874517319
* Adds and uses a new `AttentionActivationPtrs` that holds non-owning `MatPtrs`. Acts as a view into `AttentionActivations`.
* Updates `QBatch` to hold non-owning `MatPtr`s to the kv caches.
* Enables the `MatPtrT` default constructor for simpler initializations.
* Pulls out and passes `LayerWeightsPtrs::query_norm_scale` directly. While `LayerWeightsPtrs` already held non-owning `MatPtr`s, this change avoids the need to find and construct several empty weight tensors just to construct one `query_norm_scale` tensor.
PiperOrigin-RevId: 824584177
KVCache: add ctor required for MatStorageT, remove Create; bf_pre_ffw_rms_out -> pre_ffw_rms_out
optimize_test: larger vocab_size requires more steps
shared.h: Remove unused u128 type
correctly set Activation matrix rows, avoid passing as arg
ops: pass Mat instead of pointers/sizes; vectorize LayerNorm; support any weight type
mat: add OverrideRows, used by SetBatchSize
PiperOrigin-RevId: 757790736
- use HalfRope position encodings
- zero-initialize the caches for each Generate at position 0
The lack of the latter made the tests in gemma_test dependent on each other.
PiperOrigin-RevId: 694509054
Changed CompressedLayer and CompressedWeights to be constructed with an instance of a LayerConfig and WeightsConfig respectively.
Added CompressedModel to remove ByteStorageT and get rid of most of the type casting, as well as allowing the default destructor to be used and work properly.
Adjusted WeightsWrapper and ForwardLayer etc to match.
The only remaining template arg is the weight type.
This enables all the instantiations to be deleted, apart from one per type.
It also enables (but not yet done) the config to be stored in the blob file instead of having to be specified separately.
Reduces the size of the gemma_lib and weights shared libraries by a factor of 4.3 and 3.2 respectively.
PiperOrigin-RevId: 686870060
- Allocate only the required KV caches and activation batch size
- Add flags for batch sizes
- Const-correct interface: Span of const int.
- Also clean up the KVCache arg to a span.
- Move kPrefillBatchSize into RuntimeConfig and remove related global constants.
PiperOrigin-RevId: 655893197
Krzysztof Rymski
Ray Smith
The gemma.cpp Authors
Martin Stolle
Biruk Mammo
Jan Wassenberg
Apoorv Reddy
Daniel Keysers