Types & classes — github.com/andrewkchan/deepseek.cpp

CodeHub MCPcopy Create free account

hub / github.com/andrewkchan/deepseek.cpp / types & classes

Types & classes36 in github.com/andrewkchan/deepseek.cpp

⨍ Functions 195◇ Types & classes 36

↓ 4 callersClassQTensor

Tensor with quantization metadata.

src/codec.h:107

↓ 3 callersClassBlockQuant

↓ 2 callersClassKQuant

↓ 1 callersClassDebugTensor

src/model.h:482

↓ 1 callersClassMetadata

EnumActivationType

ClassBinaryDistribution

ClassBinaryDumper

Transformer Block Base */

src/model.h:182

Transformer Block - Multi-Head Attention */

src/model.h:276

Transformer Block - Multi-Latent Attention */

src/model.h:366

Quant of tensors saved in the file. This corresponds to PyTorch tensor dtypes.

ClassCompletionArgs

src/main.cpp:85

src/model.h:483

EnumInferenceMode

ClassInferenceState

Buffer for all state used during a forward pass. Members are reused across subsequent blocks and passes. This lets us avoid allocations during inferen

src/model.h:101

EnumLayerNormType

src/model.h:455

ClassPasskeyArgs

src/main.cpp:158

ClassPerplexityArgs

src/main.cpp:199

ClassProfileDisabledScope

src/profile.h:50

ClassProfileScope

src/profile.h:42

Internal Quant. This corresponds to the in-memory representation of tensors in the model.

src/sampler.h:7

EnumScoringFunc

src/main.cpp:71

Tensor data as read from the file, which serializes tensors in PyTorch format.

ClassThreadData

64 is the typical cache line size

src/test.cpp:252

src/tokenizer.h:10

src/tokenizer.h:51

src/codec.h:122

src/codec.cpp:342

src/codec.cpp:269