MCPcopy Create free account

hub / github.com/crabml/crabml / functions

Functions598 in github.com/crabml/crabml

⨍ Functions 598◇ Types & classes 96
↓ 248 callersMethodclone
(&self)
crabml-core/src/cpu/buf/api.rs:383
↓ 167 callersMethoditer
(&self)
crabml-core/src/tensor/strider.rs:96
↓ 66 callersMethodlen
(&self)
crabml-core/src/cpu/buf/api.rs:100
↓ 64 callersMethodpush
(&mut self, token: String)
crabml-llama2/src/chat.rs:147
↓ 47 callersMethodshape
(&self)
crabml-core/src/cpu/cpu_tensor.rs:113
↓ 44 callersMethodshape
(&self)
crabml-wgpu/src/wgpu_tensor.rs:57
↓ 43 callersMethodshape
(&self)
crabml-vulkan/src/vulkan_tensor.rs:170
↓ 38 callersMethodstrides
(&self)
crabml-core/src/tensor/strider.rs:57
↓ 34 callersMethodload_tensor
( &self, gf: &'a GGUFFile<'a>, name: &str, device: CpuTensorDeviceRef<'a>,
crabml-llama2/src/model.rs:479
↓ 33 callersMethodto_vec
(&self)
crabml-core/src/cpu/cpu_tensor.rs:100
↓ 25 callersFunctionnearest_i32
(fval: f32)
crabml-core/src/cpu/buf/util.rs:10
↓ 24 callersMethodmetadata
(&self)
crabml-core/src/gguf.rs:773
↓ 22 callersMethodlen
(&self)
crabml-core/src/tensor/strider.rs:208
↓ 21 callersMethodas_f32_ref
(&self)
crabml-core/src/cpu/buf/api.rs:352
↓ 21 callersMethodtrack
(&self)
crabml-core/src/tensor/metrics.rs:179
↓ 21 callersMethodwith_name
(mut self, name: String)
crabml-wgpu/src/wgpu_tensor.rs:162
↓ 20 callersMethoddequantize
dequantize the quantized tensors to f32 or f16. f32 to f16 is not considered as dequantization, but it still will be supported to simplify the convers
crabml-core/src/cpu/buf/api.rs:164
↓ 19 callersMethodexport
(&self, dst: &mut [f32])
crabml-wgpu/src/wgpu_tensor.rs:293
↓ 18 callersMethodmatmul_vec
(m, k) @ (b, k) => (b, m)
crabml-wgpu/src/wgpu_tensor.rs:588
↓ 18 callersMethodreshape
(&self, shape: Vec<usize>)
crabml-core/src/tensor/strider.rs:143
↓ 17 callersMethodadd_inplace
(self, rhs: &Self)
crabml-wgpu/src/wgpu_tensor.rs:523
↓ 16 callersMethodfinish
the reply might ended with <eos>, but not <end_of_turn>, so we need to append the <end_of_turn>
crabml-llama2/src/chat.rs:51
↓ 15 callersMethodstrider
(&self)
crabml-core/src/cpu/cpu_tensor.rs:243
↓ 15 callersFunctionvdotq_s32
(a: int8x16_t, b: int8x16_t)
crabml-core/src/cpu/archutil/aarch64.rs:141
↓ 14 callersMethodcopy_from_iter
( &mut self, iter: impl Iterator<Item = f32>, dst_offset: usize, len: usize,
crabml-core/src/cpu/buf/api.rs:323
↓ 14 callersMethodexport
(&self, dst: &mut [f32])
crabml-vulkan/src/vulkan_tensor.rs:234
↓ 13 callersFunctiongenerate_data
Generate synthetic test data
crabml-core/src/cpu/buf/util.rs:291
↓ 13 callersMethodopen
(&self)
crabml-core/src/gguf.rs:829
↓ 12 callersMethoddims
(&self)
crabml-core/src/tensor/strider.rs:61
↓ 12 callersMethoddispatch_compute
( &self, pipeline_name: &str, buffers: Vec<Subbuffer<[u8]>>, push_constants: P
crabml-vulkan/src/vulkan_device.rs:427
↓ 12 callersMethodencode_pipeline_command
( &self, key: &'static str, entries: &[wgpu::BindGroupEntry], work_group_size:
crabml-wgpu/src/wgpu_device.rs:141
↓ 12 callersMethodload
(self, gf: &'a GGUFFile<'a>)
crabml-llama2/src/model.rs:183
↓ 12 callersMethodmul_inplace
(self, rhs: &Self)
crabml-wgpu/src/wgpu_tensor.rs:493
↓ 12 callersFunctionvld1q_f16
(ptr: *const f16)
crabml-core/src/cpu/archutil/aarch64.rs:103
↓ 11 callersMethodbuf
(&self)
crabml-core/src/cpu/cpu_tensor.rs:117
↓ 11 callersMethodbuf_mut
(&mut self)
crabml-core/src/cpu/cpu_tensor.rs:121
↓ 11 callersMethodmake_storage_buffer
(&self, name: &'static str, content: &[u8])
crabml-wgpu/src/wgpu_device.rs:105
↓ 10 callersMethodrms_norm_inplace
(self, eps: f32)
crabml-wgpu/src/wgpu_tensor.rs:393
↓ 9 callersMethodas_f32_mut
(&mut self)
crabml-core/src/cpu/buf/api.rs:359
↓ 9 callersMethodcopy_rows_from
(&mut self, src: &Self, src_rows: &[usize])
crabml-wgpu/src/wgpu_tensor.rs:260
↓ 9 callersMethodrope_inplace
(self, mode: RopeMode, pos: usize, rope_dims: usize)
crabml-wgpu/src/wgpu_tensor.rs:347
↓ 8 callersMethoddevice
(&self)
crabml-core/src/cpu/cpu_tensor.rs:82
↓ 8 callersMethodlen
(&self)
crabml-core/src/cpu/buf/buf_q8_0.rs:60
↓ 7 callersMethoddup
(&self)
crabml-wgpu/src/wgpu_tensor.rs:335
↓ 7 callersFunctionget_scale_min_k4
(j: usize, q: &[u8], d: &mut u8, m: &mut u8)
crabml-core/src/cpu/buf/util.rs:19
↓ 7 callersMethoditer_f32
the quantized tensor can not be iterated directly. to iterate the quantized tensor, use `dequantize` to convert it to f32/f16 tensor first.
crabml-core/src/cpu/buf/api.rs:372
↓ 7 callersMethodnext
(&mut self)
crabml-llama2/src/chat.rs:98
↓ 7 callersMethodprefill_and_generate
simplify the test cases
crabml-llama2/src/llama2.rs:175
↓ 7 callersMethodtranspose
(&self, dims: &[usize])
crabml-core/src/tensor/strider.rs:162
↓ 6 callersMethodconcatenate
(&mut self, rhs: &Self, axis: usize)
crabml-wgpu/src/wgpu_tensor.rs:189
↓ 6 callersMethodcontiguous
(self)
crabml-wgpu/src/wgpu_tensor.rs:695
↓ 6 callersFunctionconvert_u32_vec4
(v: &[usize])
crabml-vulkan/src/vulkan_tensor.rs:489
↓ 6 callersMethoddtype
(&self)
crabml-core/src/cpu/cpu_tensor.rs:199
↓ 6 callersMethodextend
(&mut self, iter: impl Iterator<Item = f32>)
crabml-core/src/cpu/buf/api.rs:251
↓ 6 callersMethodget_string_array
(&self, key: &str)
crabml-core/src/gguf.rs:479
↓ 6 callersMethodhead_size
(&self)
crabml-llama2/src/model.rs:50
↓ 6 callersMethodlen
(&self)
crabml-core/src/cpu/cpu_tensor.rs:86
↓ 6 callersFunctionvfmaq_f16
(mut a: float16x8_t, b: float16x8_t, c: float16x8_t)
crabml-core/src/cpu/archutil/aarch64.rs:115
↓ 5 callersFunctionarray_rmse
Calculate RMSE between two f32 slices as the difference
crabml-core/src/cpu/buf/util.rs:300
↓ 5 callersMethodget_string
(&self, key: &str)
crabml-core/src/gguf.rs:471
↓ 5 callersMethoditer_f32_mut
(&mut self)
crabml-core/src/cpu/buf/api.rs:377
↓ 5 callersMethodlen
(&self)
crabml-core/src/cpu/buf/buf_q5_0.rs:73
↓ 5 callersMethodlen
(&self)
crabml-core/src/cpu/buf/buf_q5_1.rs:72
↓ 5 callersMethodlen
(&self)
crabml-core/src/cpu/buf/buf_q4_0.rs:63
↓ 5 callersMethodlen
(&self)
crabml-core/src/cpu/buf/buf_q4_1.rs:67
↓ 5 callersFunctionvdupq_n_f16
(a: u16)
crabml-core/src/cpu/archutil/aarch64.rs:127
↓ 4 callersMethodas_bytes
(&'a self)
crabml-core/src/cpu/buf/api.rs:70
↓ 4 callersMethodas_hashmap
(&self)
crabml-core/src/gguf.rs:456
↓ 4 callersMethodat_unchecked
(&self, idx: &[usize])
crabml-core/src/tensor/strider.rs:88
↓ 4 callersMethodconf
(&self)
crabml-llama2/src/llama2.rs:102
↓ 4 callersFunctiondot_product
Calculate the dot product of original inputs for test reference
crabml-core/src/cpu/buf/util.rs:319
↓ 4 callersMethodforward_multi_query_attention
( &mut self, q: T, k: T, v: T, l: usize, _pos: usize,
crabml-llama2/src/llama2.rs:527
↓ 4 callersFunctionhsum_float_8
(x: __m256)
crabml-core/src/cpu/archutil/x86_64.rs:37
↓ 4 callersMethodis_contiguous
(&self)
crabml-core/src/cpu/cpu_tensor.rs:109
↓ 4 callersFunctionmul_sum_i8_pairs_float
(x: __m256i, y: __m256i)
crabml-core/src/cpu/archutil/x86_64.rs:6
↓ 4 callersMethodprefill
prefill the model with the prompt, return the next position and the first generated token
crabml-llama2/src/llama2.rs:111
↓ 4 callersMethodread_len
Read the length for string & array. It would be an 32 bit unsigned integer on spec v1, but 64 bit on spec v2. For more information: https://github.com
crabml-core/src/gguf.rs:399
↓ 4 callersMethodread_string
(&mut self)
crabml-core/src/gguf.rs:386
↓ 4 callersMethodresize
resize is useful on pre-allocated tensors, such as kv caches, which is the only place where we use this function: 1. preallocate a tensor with a larg
crabml-core/src/tensor/strider.rs:36
↓ 4 callersMethodstrider
(&self)
crabml-wgpu/src/wgpu_tensor.rs:181
↓ 4 callersFunctionvaddvq_f16
(a: float16x8_t)
crabml-core/src/cpu/archutil/aarch64.rs:132
↓ 4 callersMethodwith_strider
(self, strider: TensorStrider)
crabml-wgpu/src/wgpu_tensor.rs:151
↓ 3 callersMethodbatch_matmul
(b, m, k) @ (b, k, n) => (b, m, n) the A matrix is dense and the B matrix is allowed to be strided
crabml-wgpu/src/wgpu_tensor.rs:638
↓ 3 callersFunctionconcatenate_2d
( buf1: &mut [A], buf2: &[B], shape1: &[usize], shape2: &[usize], strides1: &[usize],
crabml-core/src/cpu/primitives/concatenate.rs:117
↓ 3 callersFunctionconcatenate_inner
( buf1: &mut [A], buf2: &[B], shape1: &[usize], shape2: &[usize], strides1: &[usize],
crabml-core/src/cpu/primitives/concatenate.rs:80
↓ 3 callersFunctionexp_f32_cached
(x: f32, cache: &[f16])
crabml-core/src/cpu/buf/buf_f32.rs:29
↓ 3 callersMethodforward_ffn
(&self, mut x: T, l: usize, _pos: usize, activation: Activation)
crabml-llama2/src/llama2.rs:605
↓ 3 callersMethodgelu_inplace
(self)
crabml-wgpu/src/wgpu_tensor.rs:476
↓ 3 callersMethodgenerate
( &mut self, pos: usize, token: usize, steps: Option<usize>, )
crabml-llama2/src/llama2.rs:141
↓ 3 callersMethodis_contiguous
(&self)
crabml-wgpu/src/wgpu_tensor.rs:53
↓ 3 callersMethoditer_axis_inner
( &self, pos: &[usize], axis: usize, )
crabml-core/src/tensor/strider.rs:128
↓ 3 callersMethodkv_cache_len
(&self)
crabml-llama2/src/llama2.rs:106
↓ 3 callersMethodlen
(&self)
crabml-core/src/cpu/buf/buf_q5_k.rs:96
↓ 3 callersMethodlen
(&self)
crabml-core/src/cpu/buf/buf_q4_k.rs:84
↓ 3 callersFunctionmake_qkx1_quants
( n: usize, nmax: i32, data: &[f32], l: &mut [u8], the_min: &mut f32, ntry: i32, )
crabml-core/src/cpu/buf/util.rs:154
↓ 3 callersMethodread
(&mut self, n: usize)
crabml-core/src/gguf.rs:260
↓ 3 callersMethodresize
(self, axis: usize, n: usize)
crabml-wgpu/src/wgpu_tensor.rs:113
↓ 3 callersMethodscale_inplace
(self, rhs: f32)
crabml-wgpu/src/wgpu_tensor.rs:553
↓ 3 callersMethodtyp
(&self)
crabml-core/src/gguf.rs:211
↓ 3 callersFunctionvec_dot_q4_0_q8_0_fallback
(abs: &[BlockQ4_0], bbs: &[BlockQ8_0])
crabml-core/src/cpu/buf/buf_q4_0.rs:240
next →1–100 of 598, ranked by callers