gemma.cpp/gemma/gemma.h

// 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.

#ifndef THIRD_PARTY_GEMMA_CPP_GEMMA_GEMMA_H_
#define THIRD_PARTY_GEMMA_CPP_GEMMA_GEMMA_H_

#include <functional>
#include <memory>
#include <random>
#include <string>
#include <vector>

#include "compression/io.h"  // Path
#include "gemma/common.h"
#include "hwy/aligned_allocator.h"
#include "hwy/base.h"  // hwy::bfloat16_t
#include "hwy/contrib/thread_pool/thread_pool.h"

namespace gcpp {

constexpr size_t kPrefillBatchSize = 16;
constexpr size_t kDecodeBatchSize = 1;
constexpr bool kSystemPrompt = false;

struct KVCache {
  hwy::AlignedFreeUniquePtr<float[]>
      kv_cache;  // kSeqLen * kGemmaLayers * kKVHeads * kQKVDim * 2
  hwy::AlignedFreeUniquePtr<float[]>
      conv1d_cache;  // (kConv1dWidth - 1) * kModelDim * kGriffinLayers
  hwy::AlignedFreeUniquePtr<float[]>
      rglru_cache;  // kModelDim * kGriffinLayers

  static KVCache Create(Model type);
};

// The tokenizer's end of sentence and beginning of sentence token ids.
constexpr int EOS_ID = 1;
constexpr int BOS_ID = 2;

class GemmaTokenizer {
 public:
  GemmaTokenizer();
  explicit GemmaTokenizer(const Path& tokenizer_path);

  // must come after definition of Impl
  ~GemmaTokenizer();
  GemmaTokenizer(GemmaTokenizer&& other);
  GemmaTokenizer& operator=(GemmaTokenizer&& other);

  bool Encode(const std::string& input, std::vector<std::string>* pieces) const;
  bool Encode(const std::string& input, std::vector<int>* ids) const;
  bool Decode(const std::vector<int>& ids, std::string* detokenized) const;

 private:
  class Impl;
  std::unique_ptr<Impl> impl_;
};

// StreamFunc is called with (token, probability). For prompt tokens,
// probability is 0.0f. StreamFunc should return false to stop generation and
// true to continue generation.
using StreamFunc = std::function<bool(int, float)>;
// If not empty, AcceptFunc is called with token. It should return false for
// tokens you don't want to generate and true for tokens you want to generate.
using AcceptFunc = std::function<bool(int, float)>;
// If not empty, SampleFunc is called with the probability distribution for the
// next token, and its return value is used as the next generated token.
using SampleFunc = std::function<int(const float*, size_t)>;
// Will be called for layers output with:
// - position in the tokens sequence
// - name of the data, p.ex. "tokens", "block.1", "final_norm"
// - pointer to the data array
// - size of the data array
using LayersOutputFunc =
    std::function<void(int, const std::string&, const float*, size_t)>;

struct RuntimeConfig {
  size_t max_tokens;
  size_t max_generated_tokens;
  float temperature;
  int verbosity;
  std::mt19937* gen;
  StreamFunc stream_token;
  AcceptFunc accept_token;  // if empty, accepts all tokens.
  SampleFunc sample_func;   // if empty, uses SampleTopK.
  LayersOutputFunc layers_output;  // if not empty, called after each layer.
  int eos_id = EOS_ID;
};

struct TimingInfo {
  double prefill_tok_sec = 0.0;
  double gen_tok_sec = 0.0;
  double time_to_first_token = 0.0;
};

class Gemma {
 public:
  Gemma(const Path& tokenizer_path, const Path& weights, Model model_type,
        Type weight_type, hwy::ThreadPool& pool);

  // Allocates weights, caller is responsible for filling them.
  Gemma(GemmaTokenizer&& tokenizer, Model model_type, Type weight_type,
        hwy::ThreadPool& pool);
  ~Gemma();

  Model ModelType() const { return model_type_; }
  const GemmaTokenizer& Tokenizer() const { return tokenizer_; }
  const ByteStorageT& Weights() const { return weights_u8_; }
  const ByteStorageT& Prefill() const { return prefill_u8_; }
  const ByteStorageT& Decode() const { return decode_u8_; }

  void Generate(const RuntimeConfig& runtime_config,
                const std::vector<int>& prompt, size_t start_pos,
                KVCache& kv_cache, TimingInfo& timing_info);

 private:
  hwy::ThreadPool& pool_;

  GemmaTokenizer tokenizer_;
  // Type-erased so that this can be defined in the header, without requiring
  // forwarding functions.
  ByteStorageT weights_u8_;
  ByteStorageT prefill_u8_;
  ByteStorageT decode_u8_;
  Model model_type_;
  Type weight_type_;
};

// Adds BOS token and possibly 'turn' annotations, which depend on `training`
// and `pos`, the number of tokens decoded so far; returns the corresponding
// tokens. Asserts that tokenization is successful.
std::vector<int> WrapAndTokenize(const GemmaTokenizer& tokenizer,
                                 ModelTraining training, size_t pos,
                                 std::string& prompt);

// DEPRECATED, call Gemma::Generate directly.
HWY_INLINE void GenerateGemma(Gemma& gemma, const RuntimeConfig& runtime_config,
                              const std::vector<int>& prompt, size_t start_pos,
                              KVCache& kv_cache, hwy::ThreadPool& /*pool*/,
                              TimingInfo& timing_info) {
  gemma.Generate(runtime_config, prompt, start_pos, kv_cache, timing_info);
}

}  // namespace gcpp

#endif  // THIRD_PARTY_GEMMA_CPP_GEMMA_GEMMA_H_