Functions — github.com/Noeda/rllama

hub / github.com/Noeda/rllama / functions

Functions465 in github.com/Noeda/rllama

⨍ Functions 465◇ Types & classes 55

↓ 189 callersMethodunwrap

(self)

src/tensor.rs:187

↓ 133 callersMethodclone

(&self)

src/tensor.rs:118

↓ 89 callersMethodadd

(&self, other: &Tensor)

src/tensor.rs:573

↓ 67 callersFunctionf32x8_zero

()

src/simd_support.rs:58

↓ 48 callersMethodassume_on_cpu

(&self)

src/tensor.rs:238

↓ 47 callersFunctionfma_f32x8

a * b + c

src/simd_support.rs:90

↓ 47 callersMethodget_f32

(&self, row: i64, col: i64)

src/tensor.rs:341

↓ 43 callersMethodto_gpu_inplace

(&mut self, cl: &OpenCL)

src/tensor.rs:2127

↓ 41 callersMethodset_f32

(&mut self, row: i64, col: i64, val: f32)

src/tensor.rs:368

↓ 39 callersMethodto_f16

(&self)

src/tensor.rs:2281

↓ 31 callersMethodto_f32

(&self)

src/tensor.rs:2226

↓ 29 callersFunctionhorizontal_sum_f32x8

(mut ymm: __m256)

src/simd_support.rs:97

↓ 20 callersMethodrows

(&self)

src/tensor.rs:331

↓ 19 callersMethodmatrix_mul_inplace_transposed

Matrix multiplication done in-place, but the second matrix is transposed. With this, you can avoid using .transpose() on the second matrix.

src/tensor.rs:1250

↓ 19 callersMethodmatrix_mul_transposed

(&self, other: &Tensor)

src/tensor.rs:999

↓ 18 callersFunctionload_f32x8

(ptr: *const F32x8)

src/simd_support.rs:26

↓ 17 callersMethodcols

(&self)

src/tensor.rs:335

↓ 16 callersMethodbytes_for_nvalues

(&self, nvalues: usize)

src/tensor.rs:76

↓ 16 callersFunctionhorizontal_sum_and_f32_to_f16

(mut ymm: __m256)

src/simd_support.rs:108

↓ 16 callersMethodto_cpu_inplace

(&mut self)

src/tensor.rs:2181

↓ 15 callersMethoddtype

(&self)

src/tensor.rs:250

↓ 13 callersMethodtranspose

(&self)

src/tensor.rs:911

↓ 10 callersMethodmatrix_mul_naive

Slow, naive matrix multiplication. This is used as a reference to test correctness of other matrix multiplications.

src/tensor.rs:957

↓ 10 callersMethodname

(&self)

src/protomodels/sentencepiece_model.rs:1660

↓ 10 callersMethodread

(&mut self, buf: &mut [u8])

src/data_source.rs:53

↓ 9 callersMethodis_on_gpu

(&self)

src/tensor.rs:1177

↓ 9 callersMethodmatrix_mul

(&self, other: &Tensor)

↓ 8 callersMethodget_int64

(&self)

src/unpickler.rs:124

↓ 8 callersFunctionload_i16x8

(ptr: *const I16x8)

src/simd_support.rs:16

↓ 7 callersMethodget

Gets a value from a dictionary, assuming Value is a dictionary. Returns None if the key is not found, or the value is not a dictionary.

src/unpickler.rs:40

↓ 7 callersFunctioni16x8_as_f16_to_f32x8

(a: I16x8)

src/simd_support.rs:45

↓ 7 callersMethodmatrix_vector_mul_transposed

Same as matrix_vector_mul, but right side is assumed to be transposed.

src/tensor.rs:1699

↓ 6 callersMethodfinish

(&mut self)

src/tensor.rs:2170

↓ 6 callersFunctionmake_program_with_src

(ctx: &Context, src: &str)

src/tensor_opencl_support.rs:415

↓ 6 callersMethodview

(&self, rows: i64, cols: i64)

src/tensor.rs:2063

↓ 4 callersMethodassume_on_gpu

src/tensor_opencl_support.rs:173

↓ 4 callersMethodget_str_key

Same as get() but uses a string as key.

src/unpickler.rs:48

↓ 4 callersMethodis_on_cpu

(&self)

src/tensor.rs:1193

↓ 4 callersMethodmatrix_mul_inplace

Matrix multiplication done in-place

src/tensor.rs:1028

↓ 4 callersMethodopen

( &self, name: P, tensor_name: S, shard: usize, )

src/data_source.rs:97

↓ 4 callersMethodprocess_waiting_for_data_mut

(&mut self)

src/tensor.rs:2150

↓ 4 callersMethodwait

(&self)

src/tensor_opencl_support.rs:409

↓ 4 callersMethodwith_opencl_data

(&self, f: F)

src/tensor.rs:786

↓ 3 callersMethodclip_cols

(&self, cols: usize)

src/tensor.rs:2037

↓ 3 callersMethodget_tuple

(&self)

src/unpickler.rs:138

↓ 3 callersMethodhadamard_product

(&self, other: &Tensor)

src/tensor.rs:679

↓ 3 callersFunctioni16x8_zero

()

src/simd_support.rs:62

↓ 3 callersMethodid_to_str

Gives a string for a token id. Panics if the id is out of range.

src/tokenizer.rs:98

↓ 3 callersMethodinto_same_type

(self, other: &Tensor)

src/tensor.rs:1210

↓ 3 callersMethodmake_caches

(&self)

src/transformer.rs:328

↓ 3 callersMethodread_from_result

(&mut self, buf: &mut [u8])

src/rllama_main.rs:435

↓ 3 callersMethodrepetition_penalty

(self, repetition_penalty: f32)

src/token_sampler.rs:61

↓ 3 callersMethodseek

(&mut self, pos: std::io::SeekFrom)

src/data_source.rs:59

↓ 3 callersMethodsilu

(&self)

src/tensor.rs:768

↓ 3 callersMethodsoftmax

Runs softmax on row dimension.

src/tensor.rs:479

↓ 3 callersMethodtemperature

(self, temperature: f32)

src/token_sampler.rs:46

↓ 3 callersMethodto_same_type

Casts data type to whatever the other tensors data type is.

src/tensor.rs:1198

↓ 3 callersMethodto_tensor_builder

Assume that the value represents a tensor in PyTorch and return instructions how to actually load the values.

src/unpickler.rs:147

↓ 3 callersMethodtokenize_to_ids

(&self, s: S)

src/tokenizer.rs:164

↓ 3 callersMethodtop_k

(self, top_k: usize)

src/token_sampler.rs:57

↓ 3 callersMethodtop_p

(self, top_p: f32)

src/token_sampler.rs:53

↓ 3 callersFunctionunpickle

(bytes: &[u8])

src/unpickler.rs:317

↓ 3 callersMethodunpickled

(&self)

src/data_source.rs:90

↓ 3 callersMethodwith_opencl_data_mut

(&mut self, f: F)

src/tensor.rs:796

↓ 2 callersFunctioncompute_capacity_cols

(dtype: TensorDType, cols: i64)

src/tensor.rs:192

↓ 2 callersMethodcopy_inplace

Copies all values from another tensor

src/tensor_opencl_support.rs:215

↓ 2 callersFunctionf32x8_singleton

(value: f32)

src/simd_support.rs:66

↓ 2 callersFunctionf32x8_to_i16x8_as_f16

(a: F32x8)

src/simd_support.rs:50

↓ 2 callersMethodflush

(&self)

src/tensor_opencl_support.rs:130

↓ 2 callersMethodforce_f16

(mut self)

src/transformer.rs:105

↓ 2 callersMethodforward

( &self, tokens: &[TokenId], start_pos: usize, caches: &mut TransformerCaches,

src/transformer.rs:343

↓ 2 callersMethodget_str

(&self)

src/unpickler.rs:117

↓ 2 callersMethodhadamard_product_cpu

(&self, other: &Tensor)

src/tensor.rs:722

↓ 2 callersMethodhuggingface_llama_model_antitranspose

This is a special operation that undoes a transpose operation done by transformers/models/llama/convert_llama_weights_to_hf.py transpose operation whi

src/tensor.rs:841

↓ 2 callersMethodinto_dtype

(self, dtype: TensorDType)

src/tensor.rs:1221

↓ 2 callersMethodmatrix_vector_mul

Computes matrix multiplication assuming that the number of rows on the latter matrix is 1. AxB @ Cx1 = Ax1

src/tensor.rs:1673

↓ 2 callersMethodprocess_waiting_for_data

(&self)

src/tensor.rs:2160

↓ 2 callersMethodput

(&mut self, tokens: Vec<TokenId>, caches: TransformerCaches, prev_pos: usize)

src/rllama_main.rs:614

↓ 2 callersFunctionremove_first_directory

(path: P)

src/huggingface_loader.rs:125

↓ 2 callersMethodsample

( &self, logits: &Tensor, _tokenizer: &Tokenizer, existing_tokens: &[TokenId],

src/token_sampler.rs:83

↓ 2 callersMethodscalar_multiply_f32

(&self, scalar: f32)

src/tensor.rs:607

↓ 2 callersMethodsilu_cpu

(&self)

src/tensor.rs:822

↓ 2 callersMethodsqrt

(&self)

src/tensor.rs:549

↓ 2 callersFunctionstore_f32x8

(ptr: *mut F32x8, a: F32x8)

src/simd_support.rs:31

↓ 2 callersFunctionstore_i16x8

(ptr: *mut I16x8, a: I16x8)

src/simd_support.rs:21

↓ 2 callersFunctiontest_with_type

(dtype: TensorDType)

src/tensor.rs:2641

↓ 2 callersMethodtranspose_cpu

(&self)

src/tensor.rs:942

↓ 2 callersMethodtrue_clone

Cloning AttentionCache normally just makes new references to the same cache. This creates a true clone with copied tensors.

src/transformer.rs:164

↓ 2 callersMethodvalue

(&self)

src/protomodels/sentencepiece_model.rs:1595

↓ 2 callersMethodvector_matrix_mul

(&self, other: &Tensor)

src/tensor.rs:1923

↓ 2 callersMethodwait_until_ready

(&mut self)

src/tensor_opencl_support.rs:177

↓ 1 callersMethodacquire

(&self)

src/semaphore.rs:30

↓ 1 callersMethodadd_scalar

(&self, scalar: f32)

src/tensor.rs:595

↓ 1 callersFunctionapply_rotary_emb

( xq: &Tensor, xk: &Tensor, freqs_cis: &FreqsCis, seq_idx: usize, start_pos: usize, )

src/transformer.rs:823

↓ 1 callersFunctioncommand_line_inference

( cli: Cli, tr: Arc<Transformer>, tok: Arc<Tokenizer>, prompt: String, interactive_stop: V

src/rllama_main.rs:714

↓ 1 callersFunctioncompute_capacity_cols_f16

(cols: i64)

src/tensor.rs:216

↓ 1 callersFunctioncompute_capacity_cols_f32

(cols: i64)

src/tensor.rs:208

↓ 1 callersFunctioncompute_capacity_cols_k4

(cols: i64)

src/tensor.rs:200

next →1–100 of 465, ranked by callers