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

1049 lines
44 KiB
C++
Raw Blame History

// Copyright 2024 Google LLC
// SPDX-License-Identifier: Apache-2.0
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// Lightweight C++ implementation of the gemma model.
// Compiles this file for multiple architectures via "foreach_target.h", to
// which we pass the filename via macro 'argument'.
#undef HWY_TARGET_INCLUDE
#define HWY_TARGET_INCLUDE "gemma/gemma.cc" // NOLINT
#include "hwy/foreach_target.h" // IWYU pragma: keep
// Must come after foreach_target.h to avoid redefinition errors.
#include "gemma/ops.h"
#include "hwy/contrib/matvec/matvec-inl.h"
#include "hwy/highway.h"
// Non-SIMD includes and types. Note that HWY_ONCE is only true on the last
// compile pass, whereas we want this defined in the first.
#ifndef GEMMA_ONCE
#define GEMMA_ONCE
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <algorithm>
#include <array>
#include <string>
#include <utility> // std::move
#include <vector>
#include "compression/io.h" // Path
#include "gemma/common.h"
#include "gemma/configs.h"
#include "gemma/gemma.h"
#include "gemma/weights.h"
// Placeholder for internal test1, do not remove
// Placeholder for internal test4, do not remove
#include "hwy/aligned_allocator.h"
#include "hwy/base.h"
#include "hwy/contrib/thread_pool/thread_pool.h"
#include "hwy/profiler.h"
#include "hwy/timer.h"
namespace gcpp {
// Must be aligned.
template <class TConfig, size_t kBatchSize>
struct Activations {
static constexpr size_t kModelDim = TConfig::kModelDim;
static constexpr size_t kQKVDim = TConfig::kQKVDim;
static constexpr size_t kHeads = TConfig::kHeads;
static constexpr size_t kKVHeads = TConfig::kKVHeads;
static constexpr bool kIsMHA = kHeads == kKVHeads; // Multi-Head Attention
// Stride between subsequent queries. Each of Q, K, V are of length kQKVDim,
// but for MHA we store them as Q,K,V, Q,K,V, .. instead of Q..Q, K..K, V..V.
static constexpr size_t kQStride = kQKVDim * (kIsMHA ? 3 : 1);
std::array<float, kBatchSize * kModelDim> x; // input
std::array<float, kBatchSize * kModelDim> pre_att_rms_out;
std::array<float, kBatchSize * kHeads * kQStride>
q; // query vector
std::array<float, kBatchSize * kHeads * TConfig::kSeqLen>
att; // attention vector
std::array<float, kBatchSize * kHeads * kQKVDim>
att_out; // attention output
std::array<float, kHeads * kBatchSize * kModelDim>
att_post1; // attention output after linear transformation, per head
std::array<float, kBatchSize * kModelDim>
att_post2; // accumulation of attention outputs over heads
std::array<hwy::bfloat16_t, kBatchSize * kModelDim>
bf_pre_ffw_rms_out;
std::array<float, kBatchSize * TConfig::kFFHiddenDim * 2>
ffw_hidden;
// For FFW MatMul.
std::array<float, kBatchSize * TConfig::kFFHiddenDim> C1;
std::array<float, kBatchSize * TConfig::kFFHiddenDim> C2;
// bf_ version can't be used until GeluMulToBF16 issue in FFW() is resolved.
// std::array<hwy::bfloat16_t, kBatchSize * 2 * TConfig::kFFHiddenDim>
// bf_ffw_hidden;
std::array<float, kBatchSize * kModelDim> ffw_out;
std::array<float, kBatchSize * TConfig::kVocabSize> logits;
// For bf16/f32 vectors * bf16 matrix: faster to unpack once beforehand, into
// per-thread storage.
std::array<float, kModelDim * kMaxThreads> even_odd;
// Griffin layer internal activations
static constexpr size_t kGriffinDim =
TConfig::kGriffinLayers > 0 ? kModelDim : 0;
std::array<float, kBatchSize * kGriffinDim> griffin_x;
std::array<float, kBatchSize * kGriffinDim> griffin_y;
std::array<float, kBatchSize * kGriffinDim> griffin_gate_x;
std::array<float, kBatchSize * kGriffinDim>
griffin_multiplier;
};
template <typename TConfig>
struct AllocateState {
void operator()(ByteStorageT& prefill, ByteStorageT& decode) const {
// When batching queries, the prefill batch size is reduced by a factor
// of kBatchedQueryBatchSize
prefill =
AllocateSizeof<Activations<TConfig, kMinAdjustedPrefillBatchSize *
kBatchedQueryBatchSize>>();
decode = AllocateSizeof<
Activations<TConfig, kDecodeBatchSize * kBatchedQueryBatchSize>>();
}
};
template <class TConfig, size_t kBatchSize>
Activations<TConfig, kBatchSize>& GetActivations(const ByteStorageT& state_u8) {
return *reinterpret_cast<Activations<TConfig, kBatchSize>*>(state_u8.get());
}
// Placeholder for internal test2, do not remove
} // namespace gcpp
#endif // GEMMA_ONCE
// SIMD code, compiled once per target.
HWY_BEFORE_NAMESPACE();
namespace gcpp {
namespace HWY_NAMESPACE {
namespace {
template <class TConfig, size_t kBatchSize, size_t kQueryBatchSize>
HWY_NOINLINE void GriffinRecurrent(
size_t batch_start, size_t num_tokens, size_t num_queries, size_t layer,
Activations<TConfig, kBatchSize * kQueryBatchSize>& activations,
const CompressedLayer<TConfig>* layer_weights,
const std::vector<KVCache*>& kv_caches, hwy::ThreadPool& pool) {
PROFILER_ZONE("Gen.Griffin");
static_assert(kQueryBatchSize == 1,
"Griffin does not support batched queries.");
HWY_DASSERT(num_queries == 1); // TODO: add batch query support for Griffin.
KVCache& kv_cache = *kv_caches[0];
namespace hn = hwy::HWY_NAMESPACE;
using D = hn::ScalableTag<float>;
HWY_DASSERT(num_tokens <= kBatchSize);
static constexpr size_t kModelDim =
gcpp::Activations<TConfig, kBatchSize * kQueryBatchSize>::kModelDim;
static constexpr size_t kConv1dWidth = TConfig::kConv1dWidth;
static constexpr size_t kHeads = TConfig::kHeads;
// X / Y linear layers.
for (size_t batch_idx = 0; batch_idx < num_tokens; ++batch_idx) {
const size_t batch_offset = batch_idx * kModelDim;
float* HWY_RESTRICT y = activations.griffin_y.data() + batch_offset;
float* HWY_RESTRICT x = activations.griffin_x.data() + batch_offset;
TwoMatVecAdd<kModelDim, kModelDim>(
layer_weights->griffin.linear_x_w, layer_weights->griffin.linear_y_w, 0,
activations.pre_att_rms_out.data() + batch_offset,
/*add0=*/layer_weights->griffin.linear_x_biases.data(),
/*add1=*/layer_weights->griffin.linear_y_biases.data(), /*out0=*/x,
/*out1=*/y, pool);
Gelu(y, kModelDim);
}
// Conv1D.
for (size_t batch_idx = 0; batch_idx < num_tokens; ++batch_idx) {
const size_t batch_offset = batch_idx * kModelDim;
const size_t pos = batch_start + batch_idx;
float* HWY_RESTRICT x = activations.griffin_x.data() + batch_offset;
HWY_FULL(float) df;
HWY_DASSERT(kModelDim % hn::Lanes(df) == 0);
const size_t layer_offset = layer * kModelDim * (kConv1dWidth - 1);
// cache[i] = input at time t-i.
float* HWY_RESTRICT cache[HWY_MAX(kConv1dWidth, 1)];
cache[0] = x;
for (size_t i = 1; i < kConv1dWidth; i++) {
cache[i] =
kv_cache.conv1d_cache.get() + layer_offset +
((pos + kConv1dWidth - 1 - i) % (kConv1dWidth - 1)) * kModelDim;
}
for (size_t i = 0; i < kModelDim; i += hn::Lanes(df)) {
auto xv = hn::Load(df, x + i);
auto accum0 =
hn::Load(df, layer_weights->griffin.conv_biases.data() + i);
auto accum1 = hn::Zero(df);
static_assert(kConv1dWidth % 2 == 0, "Conv width must be even");
for (size_t l = 0; 2 * l < kConv1dWidth; l++) {
auto wv0 = hn::Load(df, layer_weights->griffin.conv_w.data() +
(kConv1dWidth - 1 - 2 * l) * kModelDim + i);
auto wv1 = hn::Load(df, layer_weights->griffin.conv_w.data() +
(kConv1dWidth - 2 - 2 * l) * kModelDim + i);
accum0 = hn::MulAdd(wv0, hn::Load(df, cache[l * 2] + i), accum0);
accum1 = hn::MulAdd(wv1, hn::Load(df, cache[l * 2 + 1] + i), accum1);
}
hn::Store(hn::Add(accum0, accum1), df, x + i);
hn::Store(xv, df, cache[HWY_MAX(kConv1dWidth, 1) - 1] + i);
}
}
// RGLRU
for (size_t batch_idx = 0; batch_idx < num_tokens; ++batch_idx) {
const size_t batch_offset = batch_idx * kModelDim;
const size_t pos = batch_start + batch_idx;
float* HWY_RESTRICT y = activations.griffin_y.data() + batch_offset;
float* HWY_RESTRICT x = activations.griffin_x.data() + batch_offset;
float* HWY_RESTRICT gate_x =
activations.griffin_gate_x.data() + batch_offset;
float* HWY_RESTRICT a =
activations.griffin_multiplier.data() + batch_offset;
float* HWY_RESTRICT rnn_state =
kv_cache.rglru_cache.get() + layer * kModelDim;
pool.Run(0, kHeads, [&](const uint64_t head, size_t /*thread*/) HWY_ATTR {
constexpr size_t kHeadDim = kModelDim / kHeads;
constexpr size_t kMatrixSize = kHeadDim * kHeadDim;
size_t head_offset = head * kHeadDim;
TwoOfsMatVecAddLoop<kHeadDim, kHeadDim>(
layer_weights->griffin.gate_w, kMatrixSize * head,
kMatrixSize * (kHeads + head), x + head_offset,
/*add0=*/layer_weights->griffin.gate_biases.data() + head_offset,
/*add1=*/layer_weights->griffin.gate_biases.data() + kModelDim +
head_offset,
/*out0=*/gate_x + head_offset, /*out1=*/a + head_offset);
Sigmoid(gate_x + head_offset, kHeadDim);
Sigmoid(a + head_offset, kHeadDim);
const auto fn_mul = [](D d, hn::Vec<D> x, hn::Vec<D> gate_x)
HWY_ATTR { return hn::Mul(x, gate_x); };
hn::Transform1(D(), a + head_offset, kHeadDim,
layer_weights->griffin.a.data() + head_offset, fn_mul);
hn::Transform1(D(), x + head_offset, kHeadDim, gate_x + head_offset,
fn_mul);
// RNN scan
HWY_FULL(float) df;
HWY_DASSERT(kHeadDim % hn::Lanes(df) == 0);
for (size_t i = 0; i < kHeadDim; i += hn::Lanes(df)) {
auto log_a = hn::Load(df, a + head_offset + i);
auto gated_x = hn::Load(df, x + head_offset + i);
auto rnn = hn::Load(df, rnn_state + head_offset + i);
auto a = hn::Exp(df, log_a);
auto x_multiplier = hn::Sqrt(hn::NegMulAdd(a, a, hn::Set(df, 1.0f)));
if (pos == 0) {
x_multiplier = hn::Set(df, 1.0f);
}
auto new_x = hn::MulAdd(x_multiplier, gated_x, hn::Mul(a, rnn));
hn::Store(new_x, df, rnn_state + head_offset + i);
// Join branches.
auto yv = hn::Load(df, y + head_offset + i);
auto pre_out = hn::Mul(yv, new_x);
hn::Store(pre_out, df, x + head_offset + i);
}
});
}
// Final linear layer.
for (size_t batch_idx = 0; batch_idx < num_tokens; ++batch_idx) {
const size_t batch_offset = batch_idx * kModelDim;
float* HWY_RESTRICT x = activations.griffin_x.data() + batch_offset;
float* out_ptr = activations.att_post2.data() + batch_idx * kModelDim;
MatVecAdd<kModelDim, kModelDim>(
layer_weights->griffin.linear_out_w, 0, x,
layer_weights->griffin.linear_out_biases.data(),
activations.even_odd.data(), out_ptr, pool);
}
}
template <class TConfig, size_t kBatchSize, size_t kQueryBatchSize>
HWY_NOINLINE void Attention(
size_t batch_and_query_start, size_t num_tokens, size_t num_queries,
size_t layer,
Activations<TConfig, kBatchSize * kQueryBatchSize>& activations,
const CompressedLayer<TConfig>* layer_weights,
const std::vector<KVCache*>& kv_caches,
hwy::ThreadPool& pool) {
PROFILER_ZONE("Gen.Attention");
HWY_DASSERT(num_tokens <= kBatchSize);
HWY_DASSERT(num_queries <= kQueryBatchSize);
HWY_DASSERT(batch_and_query_start % num_queries == 0);
using TActivations = Activations<TConfig, kBatchSize * kQueryBatchSize>;
constexpr size_t kQKVDim = TActivations::kQKVDim;
constexpr size_t kQStride = TActivations::kQStride;
constexpr size_t kCachePosSize = CachePosSize<TConfig>()();
constexpr size_t kCacheLayerSize = CacheLayerSize<TConfig>()();
constexpr size_t kModelDim = TActivations::kModelDim;
constexpr size_t kHeads = TConfig::kHeads;
constexpr size_t kKVHeads = TConfig::kKVHeads;
constexpr size_t kSeqLen = TConfig::kSeqLen;
GEMMA_CONSTEXPR_SQRT const float kQueryScale =
1.0f / Sqrt(static_cast<float>(kQKVDim));
constexpr bool kIsMHA = TActivations::kIsMHA; // Multi-Head Attention
const size_t batch_start = batch_and_query_start / num_queries;
const size_t num_tokens_and_queries = num_tokens * num_queries;
// If MHA, this also computes KV, which we copy to the KV cache below.
static_assert(!kIsMHA || TConfig::kInterleaveQKV); // MHA => interleaved
MatMul_4x4_Batch<kModelDim, kHeads * kQStride>(
num_tokens_and_queries, activations.pre_att_rms_out.data(),
layer_weights->qkv_einsum_w.data(), activations.q.data(), pool);
for (size_t batch_and_query_idx = 0;
batch_and_query_idx < num_tokens_and_queries; ++batch_and_query_idx) {
const float* x = activations.pre_att_rms_out.data() + batch_and_query_idx
* kModelDim;
const size_t query_idx = batch_and_query_idx % num_queries;
const size_t batch_idx = batch_and_query_idx / num_queries;
KVCache& kv_cache = *kv_caches[query_idx];
// QKV projections:
if constexpr (!kIsMHA) {
const size_t pos = batch_start + batch_idx;
const size_t cache_pos = pos % (kSeqLen + kPrefillBatchSize);
const size_t kv_offset =
cache_pos * kCachePosSize + layer * kCacheLayerSize;
float* HWY_RESTRICT kv = kv_cache.kv_cache.get() + kv_offset;
// TODO: requires MatMul support for offsets.
MatVec<kKVHeads * kQKVDim * 2, kModelDim>(
layer_weights->qkv_einsum_w, kHeads * kQKVDim * kModelDim, x,
activations.even_odd.data(), kv, pool);
}
}
// Positional encodings for kv:
pool.Run(
0, kKVHeads * num_tokens_and_queries,
[&](uint64_t task, size_t thread) HWY_ATTR {
const size_t head = task % kKVHeads;
const size_t batch_and_query_idx = task / kKVHeads;
const size_t query_idx = batch_and_query_idx % num_queries;
const size_t batch_idx = batch_and_query_idx / num_queries;
const size_t pos = batch_start + batch_idx;
const size_t cache_pos = pos % (kSeqLen + kPrefillBatchSize);
const size_t kv_offset = cache_pos * kCachePosSize +
layer * kCacheLayerSize + head * kQKVDim * 2;
KVCache& kv_cache = *kv_caches[query_idx];
float* HWY_RESTRICT kv = kv_cache.kv_cache.get() + kv_offset;
if constexpr (kIsMHA) {
// For MHA, copy kv into the KV cache from scratch space (see above).
const float* HWY_RESTRICT q =
activations.q.data() + (batch_and_query_idx * kHeads
+ head) * kQStride;
// Skip past the Q part of `q`, and copy KV to `kv`.
memcpy(kv, q + kQKVDim, 2 * kQKVDim * sizeof(float));
}
Rope(kv, TConfig::kUseHalfRope ? kQKVDim / 2 : kQKVDim, pos);
});
static_assert((kHeads % kKVHeads) == 0,
"query heads must be a multiple of key-value heads");
constexpr size_t kGroupHeads = kHeads / kKVHeads;
pool.Run(0, kHeads * num_tokens_and_queries,
[&](uint64_t task, size_t thread) HWY_ATTR {
const size_t head = task % kHeads;
const size_t batch_and_query_idx = task / kHeads;
const size_t query_idx = batch_and_query_idx % num_queries;
const size_t batch_idx = batch_and_query_idx / num_queries;
const size_t head_offset = (head / kGroupHeads) * kQKVDim * 2;
KVCache& kv_cache = *kv_caches[query_idx];
float* HWY_RESTRICT q =
activations.q.data() + (batch_and_query_idx * kHeads + head) * kQStride;
const size_t pos = batch_start + batch_idx;
// Calculate scores
float* HWY_RESTRICT head_att =
activations.att.data() + head * kSeqLen
+ batch_and_query_idx * kHeads * kSeqLen;
Rope(q, TConfig::kUseHalfRope ? kQKVDim / 2 : kQKVDim, pos);
MulByConst(kQueryScale, q, kQKVDim);
// Compute Q dot K scores
const size_t start_pos =
pos - std::min(TConfig::kAttentionWindowSizes[layer] - 1, pos);
for (size_t pos2 = start_pos; pos2 <= pos; ++pos2) {
const size_t cache_pos = pos2 % (kSeqLen + kPrefillBatchSize);
const size_t kv_offset =
cache_pos * kCachePosSize + layer * kCacheLayerSize + head_offset;
const float* HWY_RESTRICT k2 = kv_cache.kv_cache.get() + kv_offset;
const float score = Dot(q, k2, kQKVDim);
head_att[pos2 % kSeqLen] = score;
}
const size_t head_att_len = std::min(pos + 1, kSeqLen);
if constexpr (TConfig::kAttCap > 0.0f) {
LogitsSoftCap(TConfig::kAttCap, head_att, head_att_len);
}
Softmax(head_att, head_att_len);
// Weighted summation
float* HWY_RESTRICT att_out = activations.att_out.data() + head * kQKVDim +
batch_and_query_idx * kHeads * kQKVDim;
hwy::ZeroBytes(att_out, kQKVDim * sizeof(*att_out));
for (size_t pos2 = start_pos; pos2 <= pos; ++pos2) {
const size_t cache_pos = pos2 % (kSeqLen + kPrefillBatchSize);
const size_t kv_offset =
cache_pos * kCachePosSize + layer * kCacheLayerSize + head_offset;
float* HWY_RESTRICT v2 = kv_cache.kv_cache.get() + kv_offset + kQKVDim;
MulByConstAndAdd(head_att[pos2 % kSeqLen], v2, att_out, kQKVDim);
}
});
for (size_t batch_and_query_idx = 0;
batch_and_query_idx < num_tokens_and_queries; ++batch_and_query_idx) {
// TODO(szabadka) Use a single MatVecAdd like in GriffinRecurrent() after
// rearranging the weights.
float* HWY_RESTRICT att_out =
activations.att_out.data() + batch_and_query_idx * kHeads * kQKVDim;
float* HWY_RESTRICT layer_out =
activations.att_post2.data() + batch_and_query_idx * kModelDim;
MatVecT</*kAdd=*/TConfig::kSoftmaxAttnOutputBiases, kModelDim, kQKVDim>(
layer_weights->attn_vec_einsum_w, 0, att_out,
layer_weights->attention_output_biases.data(),
activations.even_odd.data(), layer_out, pool);
for (size_t head = 1; head < kHeads; ++head) {
// TODO(patrickms): Check this calculation
float* HWY_RESTRICT head_out =
activations.att_post1.data() +
head * kBatchSize * kQueryBatchSize * kModelDim;
// TODO: requires MatMul support for offsets.
MatVec<kModelDim, kQKVDim>(
layer_weights->attn_vec_einsum_w, head * kModelDim * kQKVDim,
att_out + head * kQKVDim,
activations.even_odd.data(), head_out, pool);
AddFrom(head_out, layer_out, kModelDim);
}
}
}
template <class TConfig, size_t kBatchSize>
HWY_NOINLINE void FFW(Activations<TConfig, kBatchSize>& activations,
size_t num_tokens,
const CompressedLayer<TConfig>* layer_weights,
hwy::ThreadPool& pool) {
HWY_DASSERT(num_tokens <= kBatchSize);
constexpr size_t kModelDim = TConfig::kModelDim;
constexpr size_t kFFHiddenDim = TConfig::kFFHiddenDim;
float* HWY_RESTRICT even_odd = activations.even_odd.data();
// TODO: MatMul does not yet support adding another matrix to the result.
if constexpr (!TConfig::kFFBiases) {
PROFILER_ZONE("Gen.FFW.GatedGELU");
// MatMul expects col-major B, which is what we have: kModelDim consecutive
// elements in memory, repeated kFFHiddenDim times.
const auto b1 = layer_weights->gating_einsum_w.data();
constexpr size_t kColsA = kModelDim;
constexpr size_t kColsB = kFFHiddenDim;
const auto b2 = b1 + kColsA * kColsB;
auto A = activations.bf_pre_ffw_rms_out.data();
// Will go through GELU.
MatMul_4x4_Batch<kColsA, kColsB>(num_tokens, A, b1, activations.C1.data(),
pool);
// What to multiply by.
MatMul_4x4_Batch<kColsA, kColsB>(num_tokens, A, b2, activations.C2.data(),
pool);
// Gelu and multiply by gate.
namespace hn = hwy::HWY_NAMESPACE;
using DF = hn::ScalableTag<float>;
using VF = hn::Vec<DF>;
hn::Transform1(DF(), activations.C1.data(), kFFHiddenDim * num_tokens,
activations.C2.data(), [](DF df, VF v, VF mul) HWY_ATTR {
return hn::Mul(mul, Gelu(df, v));
});
MatMul_4x4_Batch<kFFHiddenDim, kModelDim>(num_tokens, activations.C1.data(),
layer_weights->linear_w.data(),
activations.ffw_out.data(), pool);
} else {
for (size_t batch_idx = 0; batch_idx < num_tokens; ++batch_idx) {
const size_t hidden_offset = batch_idx * kFFHiddenDim * 2;
const hwy::bfloat16_t* HWY_RESTRICT vec =
activations.bf_pre_ffw_rms_out.data() + batch_idx * kModelDim;
float* HWY_RESTRICT out = activations.ffw_hidden.data() + hidden_offset;
float* HWY_RESTRICT out_mul = out + kFFHiddenDim;
PROFILER_ZONE("Gen.FFW.GatedGELU");
// Same matrix, first and second half of rows. Could fuse into one MatVec.
MatVecT<TConfig::kFFBiases, kFFHiddenDim, kModelDim>(
layer_weights->gating_einsum_w, kFFHiddenDim * kModelDim, vec,
TConfig::kFFBiases
? layer_weights->ffw_gating_biases.data() + kFFHiddenDim
: nullptr,
even_odd, out_mul, pool);
// Gate, will go through the nonlinearity.
MatVecT<TConfig::kFFBiases, kFFHiddenDim, kModelDim>(
layer_weights->gating_einsum_w, 0, vec,
layer_weights->ffw_gating_biases.data(), even_odd, out, pool);
namespace hn = hwy::HWY_NAMESPACE;
using DF = hn::ScalableTag<float>;
using VF = hn::Vec<DF>;
hn::Transform1(DF(), out, kFFHiddenDim, out_mul,
[](DF df, VF v, VF mul)
HWY_ATTR { return hn::Mul(mul, Gelu(df, v)); });
MatVecT</*kAdd=*/TConfig::kFFBiases, kModelDim, kFFHiddenDim>(
layer_weights->linear_w, 0,
activations.ffw_hidden.data() + hidden_offset,
layer_weights->ffw_output_biases.data(), even_odd,
activations.ffw_out.data() + batch_idx * kModelDim, pool);
}
}
}
template <class TConfig, size_t kBatchSize>
HWY_NOINLINE void EmbedToken(int token, size_t token_idx, size_t pos,
const CompressedWeights<TConfig>& weights,
Activations<TConfig, kBatchSize>& activations) {
constexpr size_t kModelDim = TConfig::kModelDim;
GEMMA_CONSTEXPR_EMBSCALING const float kEmbScaling =
EmbeddingScaling<TConfig>();
HWY_DASSERT(token >= 0);
HWY_DASSERT(token < TConfig::kVocabSize);
Decompress(weights.embedder_input_embedding, token * kModelDim,
activations.x.data() + token_idx * kModelDim, kModelDim);
MulByConst(kEmbScaling, activations.x.data() + token_idx * kModelDim,
kModelDim);
if constexpr (TConfig::kAbsolutePE) {
AddAbsolutePositionalEmbeddings(
activations.x.data() + token_idx * kModelDim, kModelDim,
pos + token_idx);
};
}
template <class TConfig, size_t kBatchSize, size_t kQueryBatchSize>
HWY_NOINLINE void TransformerLayer(
size_t num_tokens, size_t num_queries, size_t pos, size_t layer,
const CompressedLayer<TConfig>* layer_weights,
Activations<TConfig, kBatchSize * kQueryBatchSize>& activations,
const std::vector<KVCache*>& kv_caches, hwy::ThreadPool& pool) {
constexpr size_t kModelDim = TConfig::kModelDim;
const size_t num_tokens_and_queries = num_tokens * num_queries;
auto type = TConfig::kLayerConfig[layer];
size_t layer_of_type =
NumLayersOfTypeBefore(TConfig::kLayerConfig, type, layer);
RMSNormBatched<kBatchSize * kQueryBatchSize>(
num_tokens_and_queries, activations.x.data(),
layer_weights->pre_attention_norm_scale.data(),
activations.pre_att_rms_out.data(), kModelDim);
if (type == LayerAttentionType::kGemma) {
Attention<TConfig, kBatchSize, kQueryBatchSize>(
pos, num_tokens, num_queries, layer_of_type, activations,
layer_weights, kv_caches, pool);
} else {
// This Griffin layers should never exist unless the model is a Griffin
// model. This conditional prevents the compiler from generating code for
// this branch when building a non-Griffin model, since we have static
// asserts about the query batch size for Griffin layers.
if constexpr (TConfig::kGriffinLayers > 0) {
GriffinRecurrent<TConfig, kBatchSize, kQueryBatchSize>(
pos, num_tokens, num_queries, layer_of_type, activations,
layer_weights, kv_caches, pool);
}
}
if (TConfig::kPostNormScale) {
RMSNormInplaceBatched<kBatchSize * kQueryBatchSize>(
num_tokens_and_queries,
layer_weights->post_attention_norm_scale.data(),
activations.att_post2.data(), kModelDim);
}
AddFromBatched<kBatchSize * kQueryBatchSize>(num_tokens_and_queries,
activations.att_post2.data(),
activations.x.data(), kModelDim);
RMSNormBatched<kBatchSize * kQueryBatchSize>(
num_tokens_and_queries, activations.x.data(),
layer_weights->pre_ffw_norm_scale.data(),
activations.bf_pre_ffw_rms_out.data(), kModelDim);
FFW<TConfig, kBatchSize * kQueryBatchSize>(
activations, num_tokens_and_queries, layer_weights, pool);
if (TConfig::kPostNormScale) {
RMSNormInplaceBatched<kBatchSize * kQueryBatchSize>(
num_tokens_and_queries, layer_weights->post_ffw_norm_scale.data(),
activations.ffw_out.data(), kModelDim);
}
AddFromBatched<kBatchSize * kQueryBatchSize>(
num_tokens_and_queries, activations.ffw_out.data(),
activations.x.data(), kModelDim);
}
template <class TConfig, size_t kBatchSize, size_t kQueryBatchSize>
HWY_NOINLINE void Prefill(
const int* tokens, size_t num_tokens, size_t num_queries, size_t pos,
const CompressedWeights<TConfig>& weights,
Activations<TConfig, kBatchSize * kQueryBatchSize>& activations,
const std::vector<KVCache*>& kv_caches, hwy::ThreadPool& pool) {
HWY_DASSERT(num_queries <= kQueryBatchSize);
const size_t minibatch_size = std::min(num_tokens, kBatchSize);
PROFILER_ZONE("Gen.Prefill\\Att\\FFW");
// TODO(patrickms): Try to hoist pool.Run out of the loop.
for (size_t i = 0; i < num_tokens; i += minibatch_size) {
const size_t offset = i * num_queries;
const size_t current_token_count = std::min(
minibatch_size, num_tokens - i);
pool.Run(0, current_token_count * num_queries,
[&](const uint64_t token_idx, size_t /*thread*/) HWY_ATTR {
EmbedToken<TConfig, kBatchSize * kQueryBatchSize>(
tokens[token_idx + offset], token_idx, pos + offset,
weights, activations);
});
for (size_t layer = 0; layer < TConfig::kLayers; ++layer) {
const auto* layer_weights = weights.GetLayer(layer);
TransformerLayer<TConfig, kBatchSize, kQueryBatchSize>(
current_token_count, num_queries, pos + offset , layer, layer_weights,
activations, kv_caches, pool);
}
}
}
// Compute the transformer for a batch of input tokens. During generation,
// we usually have num_tokens == 1 (and also kBatchSize == 1).
template <class TConfig, size_t kBatchSize, size_t kQueryBatchSize>
HWY_NOINLINE void Transformer(
const int* tokens, size_t num_tokens, size_t num_queries, size_t pos,
const CompressedWeights<TConfig>& weights,
Activations<TConfig, kBatchSize * kQueryBatchSize>& activations,
const std::vector<KVCache*>& kv_caches,
hwy::ThreadPool& pool,
const LayersOutputFunc& layers_output) {
HWY_ASSERT(num_tokens <= kBatchSize);
const size_t num_tokens_and_queries = num_tokens * num_queries;
if (layers_output) {
for (size_t token_idx = 0; token_idx < num_tokens_and_queries;
++token_idx) {
float token_f = tokens[token_idx];
layers_output(pos + token_idx, "Tokens", &token_f, 1);
}
}
constexpr size_t kModelDim = TConfig::kModelDim;
for (size_t token_idx = 0; token_idx < num_tokens_and_queries; ++token_idx) {
EmbedToken<TConfig, kBatchSize * kQueryBatchSize>(
tokens[token_idx], token_idx, pos, weights, activations);
}
for (size_t layer = 0; layer < TConfig::kLayers; ++layer) {
const CompressedLayer<TConfig>* layer_weights = weights.GetLayer(layer);
TransformerLayer<TConfig, kBatchSize, kQueryBatchSize>(
num_tokens, num_queries, pos, layer, layer_weights,
activations, kv_caches, pool);
if (layers_output) {
const std::string block_name = "blocks." + std::to_string(layer);
for (size_t token_idx = 0; token_idx < num_tokens_and_queries;
++token_idx) {
layers_output(pos + token_idx, block_name,
activations.x.data() + token_idx * kModelDim, kModelDim);
}
}
}
RMSNormInplaceBatched<kBatchSize * kQueryBatchSize>(
num_tokens * num_queries, weights.final_norm_scale.data(),
activations.x.data(), kModelDim);
if (layers_output) {
for (size_t token_idx = 0; token_idx < num_tokens_and_queries;
++token_idx) {
layers_output(pos + token_idx, "final_norm",
activations.x.data() + token_idx * kModelDim, kModelDim);
}
}
}
template <class TConfig>
void RangeChecks(size_t& max_tokens, size_t& max_generated_tokens,
size_t& prompt_size) {
if (!TConfig::kUseLocalAttention) {
if (max_tokens > TConfig::kSeqLen) {
fprintf(stderr, "WARNING: max_tokens %zu > kSeqLen %d, truncating.\n",
max_tokens, TConfig::kSeqLen);
max_tokens = static_cast<size_t>(TConfig::kSeqLen);
}
}
if (max_generated_tokens > max_tokens) {
fprintf(stderr,
"WARNING: max_generated_tokens %zu > max_tokens %zu, truncating.\n",
max_generated_tokens, max_tokens);
max_generated_tokens = max_tokens - 1;
}
if (!TConfig::kUseLocalAttention) {
if (prompt_size + max_generated_tokens > max_tokens) {
fprintf(stderr,
"WARNING: prompt_size %zu + max_generated_tokens %zu > "
"max_tokens %zu, truncating to ",
prompt_size, max_generated_tokens, max_tokens);
prompt_size = std::min(prompt_size, max_tokens - max_generated_tokens);
fprintf(stderr, "%zu\n", prompt_size);
}
}
HWY_ASSERT(prompt_size > 0);
}
} // namespace
// TODO(janwas): move into RuntimeConfig
bool StreamToken(size_t query_idx, size_t pos, int token, float prob,
const RuntimeConfig& runtime_config) {
if (runtime_config.batch_stream_token) {
return runtime_config.batch_stream_token(query_idx, pos, token, prob);
}
return runtime_config.stream_token(token, prob);
}
// Placeholder for internal test3, do not remove
template <class TConfig, size_t kQueryBatchSize>
void GenerateT(const ByteStorageT& weights_u8, const ByteStorageT& prefill_u8,
const ByteStorageT& decode_u8,
const RuntimeConfig& runtime_config,
const hwy::Span<const hwy::Span<int>>& prompts, size_t pos,
const size_t query_index_offset,
const std::vector<KVCache*>& kv_caches, hwy::ThreadPool& pool,
TimingInfo& timing_info) {
constexpr size_t kAdjustedPrefillBatchSize =
std::max((size_t)1, kPrefillBatchSize / kQueryBatchSize);
static_assert(kAdjustedPrefillBatchSize >= kMinAdjustedPrefillBatchSize);
const size_t num_queries = prompts.size();
HWY_DASSERT(num_queries <= kQueryBatchSize);
pos *= num_queries; // position in (num_queries) interleaved token sequence.
const CompressedWeights<TConfig>& weights =
*reinterpret_cast<const CompressedWeights<TConfig>*>(weights_u8.get());
auto& prefill_activations =
GetActivations<TConfig,
kAdjustedPrefillBatchSize * kQueryBatchSize>(prefill_u8);
auto& activations = GetActivations<TConfig, kQueryBatchSize>(decode_u8);
size_t min_prompt_size = (size_t)-1;
size_t max_prompt_size = 0;
for (int i=0; i < prompts.size(); ++i) {
min_prompt_size = std::min(min_prompt_size, prompts[i].size());
max_prompt_size = std::max(max_prompt_size, prompts[i].size());
}
std::vector<int> prompt;
prompt.reserve(max_prompt_size * prompts.size());
for (int i = 0; i < max_prompt_size; ++i) {
for (int j=0; j < prompts.size(); ++j) {
if (i < prompts[j].size()) {
prompt.push_back(prompts[j][i]);
} else {
prompt.push_back(0);
}
}
}
constexpr size_t kVocabSize = TConfig::kVocabSize;
size_t max_tokens = runtime_config.max_tokens;
size_t max_generated_tokens = runtime_config.max_generated_tokens;
RangeChecks<TConfig>(max_tokens, max_generated_tokens, max_prompt_size);
if (pos >= max_tokens) {
fprintf(stderr, "Warning: pos %zu >= max_tokens %zu, aborting.\n", pos,
max_tokens);
return;
}
// If no sample_func is provided, we use top-k sampling.
const SampleFunc sample_token =
runtime_config.sample_func
? runtime_config.sample_func
: [&](const float* logits, size_t vocab_size) -> int {
return SampleTopK<TConfig::kTopK>(logits, vocab_size, *runtime_config.gen,
runtime_config.temperature,
runtime_config.accept_token);
};
std::vector<bool> reached_eos(num_queries);
std::fill(reached_eos.begin(), reached_eos.end(), false);
// pos indexes the KV cache. In the first turn of a chat, pos = 0.
//
// After the first turn, pos gets passed in with > 0 corresponding to the
// current token position in the KV cache.
//
// pos_offset keeps track of the relative position within the turn, starting
// at 0 each turn. During prefill, pos_offset corresponds to the index into
// the prompt vector.
//
// In single-turn (non-chat) usage, pos and pos_offset start at 0 and are
// always equal.
size_t pos_offset = 0; // offset relative to pos
// Used to keep track of how many tokens are processed per prompt,
// so that we know when to start generating tokens.
size_t single_prompt_pos_offset = 0;
const double prefill_start = hwy::platform::Now();
// Prefill stops before prompt_size - 1 since the last prompt token is the
// first input token for generation.
while (single_prompt_pos_offset < min_prompt_size - 1) {
const size_t batch_size = std::min(
kPrefillBatchSize, min_prompt_size - 1 - single_prompt_pos_offset);
const size_t batch_and_query_size = batch_size * num_queries;
HWY_DASSERT(batch_size <= kPrefillBatchSize);
HWY_DASSERT(single_prompt_pos_offset + batch_size <= min_prompt_size - 1);
HWY_DASSERT(pos_offset + batch_size <= (min_prompt_size - 1) * num_queries);
const int* batch_tokens = prompt.data() + pos_offset;
Prefill<TConfig, kAdjustedPrefillBatchSize, kQueryBatchSize>(
batch_tokens, batch_size, num_queries, pos, weights,
prefill_activations, kv_caches, pool);
for (size_t idx = 0; idx < batch_size; ++idx) {
bool all_tokens_eos = true;
for (size_t query_idx = 0; query_idx < num_queries; ++query_idx) {
if (reached_eos[query_idx]) continue;
if (StreamToken(
query_idx + query_index_offset, single_prompt_pos_offset,
batch_tokens[idx * num_queries + query_idx], 0.0f,
runtime_config)) {
all_tokens_eos = false;
} else {
reached_eos[query_idx] = true;
}
}
if (all_tokens_eos) {
return;
}
}
pos += batch_and_query_size;
pos_offset += batch_and_query_size;
single_prompt_pos_offset += batch_size;
}
timing_info.prefill_tok_sec =
static_cast<double>(pos_offset) / (hwy::platform::Now() - prefill_start);
// Start generation.
const double gen_start = hwy::platform::Now();
HWY_DASSERT(single_prompt_pos_offset == min_prompt_size - 1);
size_t pos_gen_start = pos_offset;
int token = prompt.at(pos_offset);
std::vector<int>::const_iterator first = prompt.begin() + pos_offset;
std::vector<int>::const_iterator last = first + num_queries;
std::vector<int> gen_tokens(first, last);
// The loop below is not yet prepared for decode batch size > 1.
HWY_ASSERT(kDecodeBatchSize == 1);
bool all_tokens_eos = true;
for (size_t i=0; i < num_queries; ++i) {
if (reached_eos[i]) continue;
if (StreamToken(i + query_index_offset,
single_prompt_pos_offset, gen_tokens[i], 0.0f,
runtime_config)) {
all_tokens_eos = false;
} else {
reached_eos[i] = true;
}
}
if (all_tokens_eos) {
return;
}
for (size_t generate_pos = 0;
generate_pos < max_tokens && generate_pos < max_generated_tokens;
++single_prompt_pos_offset, ++generate_pos) {
Transformer<TConfig, kDecodeBatchSize, kQueryBatchSize>(
gen_tokens.data(), kDecodeBatchSize, num_queries, pos, weights,
activations, kv_caches, pool, runtime_config.layers_output);
float token_logit = 0.0f;
// The condition below is always true if we are doing Prefill above.
// We keep it here for clarity so that the code is correct even if Prefill
// is disabled.
bool all_tokens_eos = true;
float* x = activations.x.data();
float* logits = activations.logits.data();
for (size_t i = 0; i < num_queries; ++i, ++pos, ++pos_offset,
x += TConfig::kModelDim, logits += kVocabSize) {
const size_t prompt_size = prompts[i].size();
const bool is_generating_phase =
(single_prompt_pos_offset >= prompt_size - 1);
if (is_generating_phase) {
PROFILER_ZONE("Gen.Embedding");
// Compute logits from last layer activations.
MatVec<kVocabSize, TConfig::kModelDim>(
weights.embedder_input_embedding, 0, x, activations.even_odd.data(),
logits, pool);
if constexpr (TConfig::kFinalCap > 0.0f) {
LogitsSoftCap(TConfig::kFinalCap, activations.logits.data(),
kVocabSize);
}
// Barrier: must have all logits so we can subtract max.
Softmax(logits, kVocabSize);
token = sample_token(logits, kVocabSize);
token_logit = logits[token];
if (generate_pos == 0) {
timing_info.time_to_first_token = hwy::platform::Now() - gen_start;
}
} else {
// We would take this branch if we were not doing Prefill but would
// process the tokens of the prompt one at a time.
token = prompt.at(pos_offset);
token_logit = 0.0f;
}
if (!reached_eos[i]) {
if (!StreamToken(i + query_index_offset, single_prompt_pos_offset+1,
token, token_logit, runtime_config)) {
token = runtime_config.eos_id;
}
if (token != runtime_config.eos_id) {
all_tokens_eos = false;
} else {
reached_eos[i] = true;
}
}
gen_tokens[i] = token;
}
if (all_tokens_eos) {
break;
}
}
timing_info.gen_tok_sec = static_cast<double>(pos_offset - pos_gen_start) /
(hwy::platform::Now() - gen_start);
}
template <class TConfig>
void GenerateOneQueryT(const ByteStorageT& weights_u8,
const ByteStorageT& prefill_u8,
const ByteStorageT& decode_u8,
const RuntimeConfig& runtime_config,
const std::vector<int>& prompt, size_t pos,
KVCache& kv_cache, hwy::ThreadPool& pool,
TimingInfo& timing_info) {
std::vector<hwy::Span<int>> prompt_vector = {
hwy::Span<int>(const_cast<int*>(prompt.data()), prompt.size())};
const hwy::Span<const hwy::Span<int>> prompts(
prompt_vector.data(), prompt_vector.size());
std::vector<KVCache*> kv_caches = {&kv_cache};
GenerateT<TConfig, 1>(weights_u8, prefill_u8, decode_u8,
runtime_config, prompts, pos, 0,
kv_caches, pool, timing_info);
}
template <class TConfig>
void GenerateBatchT(const ByteStorageT& weights_u8,
const ByteStorageT& prefill_u8,
const ByteStorageT& decode_u8,
const RuntimeConfig& runtime_config,
const hwy::Span<const hwy::Span<int>>& prompts,
size_t pos, const std::vector<KVCache*>& kv_caches,
hwy::ThreadPool& pool,
TimingInfo& timing_info) {
// Disable query batching for Griffin models.
constexpr size_t kQueryBatchSize =
(TConfig::kGriffinLayers > 0) ? 1 : kBatchedQueryBatchSize;
for (size_t i = 0; i < prompts.size(); i += kQueryBatchSize) {
const size_t num_queries = std::min(prompts.size() - i, kQueryBatchSize);
const hwy::Span<const hwy::Span<int>> current_prompts(
prompts.data() + i, num_queries);
GenerateT<TConfig, kQueryBatchSize>(weights_u8, prefill_u8, decode_u8,
runtime_config, current_prompts,
pos, i, kv_caches, pool, timing_info);
}
}
} // namespace HWY_NAMESPACE
} // namespace gcpp
HWY_AFTER_NAMESPACE();
#if HWY_ONCE
namespace gcpp {
Gemma::Gemma(const Path& tokenizer_path, const Path& weights,
const ModelInfo& info, hwy::ThreadPool& pool)
: pool_(pool), tokenizer_(tokenizer_path), info_(info) {
weights_u8_ = LoadCompressedWeights(weights, info.model, info.weight, pool);
CallForModelAndWeight<AllocateState>(info.model, info.weight, prefill_u8_,
decode_u8_);
}
Gemma::Gemma(GemmaTokenizer&& tokenizer, const ModelInfo& info,
hwy::ThreadPool& pool)
: pool_(pool), tokenizer_(std::move(tokenizer)), info_(info) {
HWY_ASSERT(info.weight == Type::kF32);
weights_u8_ =
CallForModel<float, AllocateCompressedWeights>(info.model, pool);
CallForModelAndWeight<AllocateState>(info.model, info.weight, prefill_u8_,
decode_u8_);
}
Gemma::~Gemma() {
CallForModelAndWeight<DeleteCompressedWeights>(info_.model, info_.weight,
weights_u8_);
}
void Gemma::Generate(const RuntimeConfig& runtime_config,
const std::vector<int>& prompt, size_t start_pos,
KVCache& kv_cache, TimingInfo& timing_info) {
pool_.SetWaitMode(hwy::PoolWaitMode::kSpin);
GEMMA_EXPORT_AND_DISPATCH(
info_.model, info_.weight, GenerateOneQueryT,
(weights_u8_, prefill_u8_, decode_u8_, runtime_config, prompt, start_pos,
kv_cache, pool_, timing_info));
pool_.SetWaitMode(hwy::PoolWaitMode::kBlock);
}
void Gemma::GenerateBatch(const RuntimeConfig& runtime_config,
const hwy::Span<const hwy::Span<int>>& prompts,
size_t start_pos,
const std::vector<KVCache*>& kv_caches,
TimingInfo& timing_info) {
pool_.SetWaitMode(hwy::PoolWaitMode::kSpin);
GEMMA_EXPORT_AND_DISPATCH(
info_.model, info_.weight, GenerateBatchT,
(weights_u8_, prefill_u8_, decode_u8_, runtime_config, prompts, start_pos,
kv_caches, pool_, timing_info));
pool_.SetWaitMode(hwy::PoolWaitMode::kBlock);
}
// TODO(janwas): move to common.h.
void Wrap(const ModelInfo& info, size_t pos, std::string& prompt) {
// Instruction-tuned models are trained to expect control tokens.
if (info.training == ModelTraining::GEMMA_IT) {
// Prepend "<end_of_turn>" if this is a multi-turn dialogue continuation.
const std::string start = (pos == 0)
? "<start_of_turn>user\n"
: "<end_of_turn>\n<start_of_turn>user\n";
prompt = start + prompt + "<end_of_turn>\n<start_of_turn>model\n";
}
}
std::vector<int> WrapAndTokenize(const GemmaTokenizer& tokenizer,
const ModelInfo& info, size_t pos,
std::string& prompt) {
Wrap(info, pos, prompt);
std::vector<int> tokens;
HWY_ASSERT(tokenizer.Encode(prompt, &tokens));
// Both pre-trained and instruction-tuned require BOS as first token.
if (pos == 0) {
tokens.insert(tokens.begin(), BOS_ID);
}
return tokens;
}
} // namespace gcpp
#endif // HWY_ONCE
Reference in New Issue View Git Blame Copy Permalink