Functions — github.com/andrewkchan/deepseek.cpp

CodeHub MCPcopy Create free account

hub / github.com/andrewkchan/deepseek.cpp / functions

Functions195 in github.com/andrewkchan/deepseek.cpp

⨍ Functions 195◇ Types & classes 36

↓ 56 callersFunctioncdiv

src/model.cpp:18

↓ 55 callersFunctioncheck_tensor

src/model.cpp:129

↓ 41 callersFunctionget_tensor

src/model.cpp:138

↓ 23 callersFunctionhalf_to_float

↓ 23 callersFunctionsave_weight_and_scale

(weight_name: str, scale_name: str, weight_and_scale: Tuple[torch.Tensor, torch.Tensor])

↓ 22 callersMethodaqb

activation quantization buffer

src/model.h:140

↓ 21 callersMethodhb

src/model.h:112

↓ 19 callersFunctionmatmul

src/infer.cpp:381

↓ 17 callersFunctionconv

(weight_name: str, scale_name: str)

↓ 17 callersFunctionget_timestamp_ms

src/time_utils.cpp:5

↓ 16 callersMethodxb

src/model.h:107

↓ 15 callersFunctionfloat_to_half

↓ 15 callersMethodx

current activations

src/model.h:106

↓ 11 callersFunctionnearest_int

src/quant.cpp:34

↓ 10 callersFunction_matmul

src/infer.cpp:121

↓ 10 callersMethodkv_a

src/model.h:117

↓ 10 callersMethodxb2

TODO: do we need xb2?

src/model.h:110

↓ 9 callersFunctionerror_usage

src/main.cpp:18

↓ 9 callersMethodhb2

src/model.h:113

↓ 9 callersFunctionmatmul_unscaled

src/infer.cpp:419

↓ 9 callersMethodq_a

src/model.h:114

↓ 8 callersFunctionassertArrayEquals

src/test.cpp:29

↓ 8 callersFunctionget_scale_shuffle_q3k

shuffles to pick the required scales in dot products

src/quant.cpp:55

↓ 7 callersMethodactive_bytes

src/model.cpp:324

↓ 7 callersMethodforward

src/model.cpp:874

↓ 7 callersMethodparse_args

Returns true if args are valid, false otherwise

src/main.cpp:162

↓ 7 callersMethodq

src/model.h:115

↓ 7 callersFunctionrmsnorm

src/infer.cpp:601

↓ 6 callersFunctioncodec_dtype_to_string

src/codec.cpp:70

↓ 6 callersFunctionfloat8e5m2_to_float

↓ 6 callersFunctionrope

src/infer.cpp:648

↓ 6 callersFunctionrope_v3

src/infer.cpp:670

↓ 6 callersMethodropebuf

src/model.h:120

↓ 5 callersMethodatt

src/model.h:125

↓ 5 callersFunctionfloat_to_float8e5m2

↓ 5 callersFunctionhelp_usage_interactive

src/main.cpp:45

↓ 5 callersMethodkv_b

src/model.h:118

↓ 5 callersMethodlogits

src/model.h:137

↓ 4 callersFunctiongelu

src/infer.cpp:636

↓ 4 callersFunctionmatmul_expert

src/infer.cpp:423

↓ 4 callersMethodq_rope

src/model.h:130

↓ 4 callersFunctionquantize

(t: torch.Tensor)

↓ 4 callersFunctionsilu

src/infer.cpp:640

↓ 3 callersFunctionattn

Compute next value in a sequence for a single causal self-attention head.

src/infer.cpp:728

↓ 3 callersFunctionconv_experts

(weight_and_scale_names: List[Tuple[str, str]])

↓ 3 callersMethoddevice

src/model.h:142

↓ 3 callersMethodencode

src/tokenizer.cpp:57

↓ 3 callersFunctionencode_prompt

src/main.cpp:257

↓ 3 callersFunctionfloat_array_to_half

src/test.cpp:54

↓ 3 callersMethodk

src/model.h:121

↓ 3 callersMethodkv_rope_cache

src/model.h:413

↓ 3 callersMethodload

src/debug.cpp:21

↓ 3 callersFunctionload_and_dequantize

(weight_name: str, scale_name: str)

↓ 3 callersFunctionper_tensor_quantize

Quantize a tensor using per-tensor static scaling factor. Args: tensor: The input tensor. dtype: The data type to quantize to.

↓ 3 callersFunctionsoftmax

Compute the softmax of an input vector `x` of length `size` and store it in `o`.

src/infer.cpp:472

↓ 2 callersMethodactive_experts

src/model.h:135

↓ 2 callersMethodactive_experts_weights

src/model.h:134

↓ 2 callersFunctionbits_per_weight

src/codec.cpp:40

↓ 2 callersFunctionblockwise_dequantize

(qweight: torch.Tensor, scale: torch.Tensor, block_size: torch.Tensor)

↓ 2 callersFunctionblockwise_quantize

(weight: torch.Tensor, block_size: torch.Tensor, dtype: torch.dtype)

↓ 2 callersMethoddecode_one

src/tokenizer.cpp:44

↓ 2 callersFunctionhsum_float_8

horizontally add 8 floats

src/quant.cpp:46

↓ 2 callersFunctionis_k_quant

src/codec.cpp:66

↓ 2 callersFunctionk_quantize

Quantize a 2D float32 tensor to Q2_K or Q3_K format. Args: tensor: Input tensor of shape (M, N) where N must be a multiple of 256 Ret

↓ 2 callersMethodkey_cache

src/model.h:321

↓ 2 callersMethodkv_nope_cache

src/model.h:411

↓ 2 callersMethodmoe_weights

mixture of experts

src/model.h:133

↓ 2 callersMethodq_c

src/model.h:128

↓ 2 callersFunctionquant_to_string

src/codec.cpp:13

↓ 2 callersFunctionquantize_row_q8_K_ref

src/quant.cpp:616

↓ 2 callersFunctionrun_completion

src/main.cpp:277

↓ 2 callersFunctionrun_passkey

src/main.cpp:433

↓ 2 callersFunctionrun_perplexity

src/main.cpp:371

↓ 2 callersMethodsample

src/sampler.cpp:41

↓ 2 callersFunctionset_profile_enabled

src/profile.cpp:9

↓ 2 callersMethodv

src/model.h:123

↓ 2 callersMethodvalue_cache

src/model.h:323

↓ 1 callersFunctionallocateAlignedArray

Helper function to allocate aligned memory

src/test.cpp:209

↓ 1 callersFunctionarrayEquals

src/test.cpp:17

↓ 1 callersFunctionattn_mla

Compute next value in a sequence for a single causal self-attention head. MLA variant: uses combined latent-KV cache and PE-KV cache.

src/infer.cpp:766

↓ 1 callersMethodblock

src/model.cpp:290

↓ 1 callersFunctioncodec_dtype_size

src/codec.cpp:109

↓ 1 callersFunctiondequantize_row_q2_K

src/quant.cpp:217

↓ 1 callersFunctiondequantize_row_q3_K

src/quant.cpp:384

↓ 1 callersMethodencoding_to_debug_string

src/tokenizer.cpp:96

↓ 1 callersFunctionfill_random

src/test.cpp:188

↓ 1 callersFunctionfloatEquals

src/test.cpp:13

↓ 1 callersFunctionfloat_array_to_float8e5m2

src/test.cpp:62

↓ 1 callersMethodfrom_json

src/codec.cpp:124

↓ 1 callersMethodfrom_yalm

src/model.cpp:22

↓ 1 callersFunctionget_profile_enabled

src/profile.cpp:13

↓ 1 callersFunctionggml_vec_dot_q2_K_q8_K

src/quant.cpp:666

↓ 1 callersFunctionggml_vec_dot_q3_K_q8_K

src/quant.cpp:434

↓ 1 callersFunctiongpt2_bytes_to_unicode

↓ 1 callersFunctionload_tokens

(tokenizer_path, vocab_size)

↓ 1 callersFunctionload_weights

Generator that yields shards of weights loaded from the model files in huggingface format. Each shard contains a dictionary of tensors, with weig

↓ 1 callersMethodlogit_indices

src/model.h:138

↓ 1 callersFunctionmake_q3_quants

src/quant.cpp:249

↓ 1 callersFunctionmem_bench

src/test.cpp:218

↓ 1 callersFunctionmem_bench2

src/test.cpp:264

next →1–100 of 195, ranked by callers