Llama.ixx File Reference

LLaMA-style decoder-only transformer network. More...

#include <string>
#include <vector>
#include <memory>
#include <sstream>
#include <iostream>
#include <stdexcept>
#include <cstdint>
#include <format>
#include <optional>
#include <algorithm>
import Serialization.WeightsReader;
import Serialization.Tensor;
import Serialization.ModelArchive;
import Compute.ExecutionContextFactory;
import Compute.ExecutionContext;
import Dnn.RuntimeMode;
import Dnn.LanguageModelNetwork;
import Dnn.ModelType;
import Dnn.ITensor;
import Dnn.TensorDataTypeTraits;
import Dnn.TensorTypes;
import Dnn.ActivationType;
import Compute.DeviceTypeTraits;
import Dnn.ComponentType;
import Dnn.TensorDataType;
import Logging.Logger;
import Dnn.Tensor;
import Dnn.Components.Rope;
import Dnn.Components.LlamaTransformer:Presets;
import Dnn.Quantization.KvCache.Policy;
import Compute.DeviceType;
import Dnn.Components.Linear;
import Dnn.Components.RmsNorm;
import Dnn.Quantization.Weight.Policies;
import Compute.DeviceAllocation;
import Compute.Device;
import Dnn.Component;
import Dnn.Components.TokenEmbedding;
import Compute.DeviceId;
import Compute.GqaWorkspace;
import Compute.GqaState;
import Compute.CpuMemoryResource;

Classes
class	Mila::Dnn::LlamaTransformer< TDeviceType, TPrecision, TWeightQuantization, TKvCachePolicy >
	LLaMA-style transformer (decoder-only) for autoregressive token prediction. More...

Namespaces
namespace	Mila
	Mila main API namespace.

Variables
constexpr int64_t	Mila::Dnn::kLlamaPrefillChunkFloor = 128
constexpr int64_t	Mila::Dnn::kLlamaPrefillChunkRungs [] = { 512, 256, 128 }

Detailed Description

LLaMA-style decoder-only transformer network.

Device-templated network implementing a LLaMA-style autoregressive decoder.