Skip to content

Commit

Permalink
Support additional scaling
Browse files Browse the repository at this point in the history 
PiperOrigin-RevId: 631429113
  • Loading branch information
@pchx @copybara-github
pchx authored and copybara-github committed May 7, 2024
1 parent f6d02b2 commit bacba35
Show file tree
Hide file tree
Showing 2 changed files with 36 additions and 0 deletions.
3 changes: 3 additions & 0 deletions gemma/configs.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -97,6 +97,7 @@ struct ConfigGemma7B {
static constexpr int kQKVDim = 256; // query size == key size == value size
static constexpr int kTopK = gcpp::kTopK;
static constexpr bool kAbsolutePE = false;
static constexpr bool kPostNormScale = false;

// SSM config.
static constexpr int kConv1dWidth = 0;
Expand Down Expand Up @@ -128,6 +129,7 @@ struct ConfigGemma2B {
static constexpr int kQKVDim = 256; // query size == key size == value size
static constexpr int kTopK = gcpp::kTopK;
static constexpr bool kAbsolutePE = false;
static constexpr bool kPostNormScale = false;

// SSM config.
static constexpr int kConv1dWidth = 0;
Expand Down Expand Up @@ -187,6 +189,7 @@ struct ConfigGriffin2B {
static constexpr int kQKVDim = 256; // query size == key size == value size
static constexpr int kTopK = gcpp::kTopK;
static constexpr bool kAbsolutePE = false;
static constexpr bool kPostNormScale = false;

// SSM config.
static constexpr int kConv1dWidth = 4;
Expand Down
33 changes: 33 additions & 0 deletions gemma/gemma.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -87,6 +87,7 @@ struct Layer {
static constexpr size_t kGatingEinsumWSize = 2 * kFFHiddenDim * kModelDim;
static constexpr size_t kConv1dWidth = TConfig::kConv1dWidth;
static constexpr bool kFFBiases = TConfig::kFFBiases;
static constexpr bool kPostNormScale = TConfig::kPostNormScale;
static constexpr size_t kAOBiasDim =
TConfig::kSoftmaxAttnOutputBiases ? kModelDim : 0;
static constexpr size_t kGriffinDim =
Expand Down Expand Up @@ -121,6 +122,8 @@ struct Layer {
ArrayT<float, kModelDim * kFFHiddenDim> linear_w;
ArrayT<float, kModelDim> pre_attention_norm_scale;
ArrayT<float, kModelDim> pre_ffw_norm_scale;
ArrayT<float, kPostNormScale ? kModelDim : 0> post_attention_norm_scale;
ArrayT<float, kPostNormScale ? kModelDim : 0> post_ffw_norm_scale;

ArrayT<float, kFFBiases ? 2 * kFFHiddenDim : 0> ffw_gating_biases;
ArrayT<float, kFFBiases ? kModelDim : 0> ffw_output_biases;
Expand Down Expand Up @@ -269,6 +272,10 @@ hwy::AlignedFreeUniquePtr<uint8_t[]> LoadWeights(
SCALE_WEIGHTS(linear_w);
READ_WEIGHTS(pre_attention_norm_scale);
READ_WEIGHTS(pre_ffw_norm_scale);
if (TConfig::kPostNormScale) {
READ_WEIGHTS(post_attention_norm_scale);
READ_WEIGHTS(post_ffw_norm_scale);
}
if (TConfig::kFFBiases) {
READ_WEIGHTS(ffw_gating_biases);
READ_WEIGHTS(ffw_output_biases);
Expand Down Expand Up @@ -311,6 +318,7 @@ struct CompressedLayer {
static constexpr size_t kGatingEinsumWSize = TLayer::kGatingEinsumWSize;
static constexpr size_t kConv1dWidth = TLayer::kConv1dWidth;
static constexpr bool kFFBiases = TLayer::kFFBiases;
static constexpr bool kPostNormScale = TConfig::kPostNormScale;
static constexpr size_t kAOBiasDim = TLayer::kAOBiasDim;
static constexpr size_t kGriffinDim = TLayer::kGriffinDim;

Expand Down Expand Up @@ -346,6 +354,9 @@ struct CompressedLayer {
// We don't yet have an RMSNorm that accepts all WeightT.
ArrayT<hwy::bfloat16_t, kModelDim> pre_attention_norm_scale;
ArrayT<hwy::bfloat16_t, kModelDim> pre_ffw_norm_scale;
ArrayT<hwy::bfloat16_t, kPostNormScale ? kModelDim : 0>
post_attention_norm_scale;
ArrayT<hwy::bfloat16_t, kPostNormScale ? kModelDim : 0> post_ffw_norm_scale;

ArrayT<float, kFFBiases ? 2 * kFFHiddenDim : 0> ffw_gating_biases;
ArrayT<float, kFFBiases ? kModelDim : 0> ffw_output_biases;
Expand Down Expand Up @@ -949,6 +960,11 @@ HWY_NOINLINE void Prefill(const int* tokens, size_t num_tokens, size_t pos,

pool.Run(0, num_tokens, [&](const uint64_t token_idx,
size_t /*thread*/) HWY_ATTR {
if (TConfig::kPostNormScale) {
RMSNormInplace(layer_weights->post_attention_norm_scale.data(),
activations.att_post2.data() + token_idx * kModelDim,
kModelDim);
}
AddFrom(activations.att_post2.data() + token_idx * kModelDim,
activations.x.data() + token_idx * kModelDim, kModelDim);
RMSNorm(activations.x.data() + token_idx * kModelDim,
Expand All @@ -958,6 +974,11 @@ HWY_NOINLINE void Prefill(const int* tokens, size_t num_tokens, size_t pos,
});
FFW<kBatchSize>(activations, num_tokens, layer_weights, pool);
for (size_t token_idx = 0; token_idx < num_tokens; ++token_idx) {
if (TConfig::kPostNormScale) {
RMSNormInplace(layer_weights->post_ffw_norm_scale.data(),
activations.ffw_out.data() + token_idx * kModelDim,
kModelDim);
}
AddFrom(activations.ffw_out.data() + token_idx * kModelDim,
activations.x.data() + token_idx * kModelDim, kModelDim);
}
Expand Down Expand Up @@ -1005,10 +1026,18 @@ void Transformer(int token, size_t pos, const WeightArrayT& weights,
GriffinRecurrent<1>(pos, 1, layer_of_type, activations, layer_weights,
kv_cache, pool);
}
if (TConfig::kPostNormScale) {
RMSNormInplace(layer_weights->post_attention_norm_scale.data(),
activations.att_post2.data(), kModelDim);
}
AddFrom(activations.att_post2.data(), activations.x.data(), kModelDim);
RMSNorm(activations.x.data(), layer_weights->pre_ffw_norm_scale.data(),
activations.bf_pre_ffw_rms_out.data(), kModelDim);
FFW<1>(activations, /* num_tokens = */ 1, layer_weights, pool);
if (TConfig::kPostNormScale) {
RMSNormInplace(layer_weights->post_ffw_norm_scale.data(),
activations.ffw_out.data(), kModelDim);
}
AddFrom(activations.ffw_out.data(), activations.x.data(), kModelDim);
if (layers_output != nullptr) {
std::string block_name = "blocks." + std::to_string(layer);
Expand Down Expand Up @@ -1336,6 +1365,10 @@ void ForEachTensor(const Weights<TConfig>* weights,
CALL_FUNC("gr_a", griffin.a);
}
CALL_FUNC("pre_att_ns", pre_attention_norm_scale);
if (TConfig::kPostNormScale) {
CALL_FUNC("post_att_ns", post_attention_norm_scale);
CALL_FUNC("post_ff_ns", post_ffw_norm_scale);
}

if (TConfig::kFFBiases) {
CALL_FUNC("ffw_gat_b", ffw_gating_biases);
Expand Down

0 comments on commit bacba35

Please sign in to comment.