Functions — github.com/Gadersd/llama2-burn

hub / github.com/Gadersd/llama2-burn / functions

Functions66 in github.com/Gadersd/llama2-burn

⨍ Functions 66 ◇ Types & classes 21

↓ 8 callersFunctionload_linear

( path: &str, device: &B::Device, )

src/model.rs:517

↓ 8 callersFunctionsave_linear

(linear, path)

llama-py/dump.py:18

↓ 8 callersFunctionsave_scalar

(s, name, path)

llama-py/dump.py:7

↓ 4 callersMethod__init__

(self, dim: int, eps: float = 1e-6)

llama-py/model.py:34

↓ 4 callersMethoddecode

(&self, tokens: &[i64], skip_special_tokens: bool)

src/token.rs:53

↓ 4 callersMethodencode

(&self, text: &str, include_bos: bool, include_eos: bool)

src/token.rs:23

↓ 4 callersMethodforward

(&self, x: Tensor<B, 2, Int>)

src/model.rs:80

↓ 4 callersFunctionsave_tensor

(tensor, name, path)

llama-py/dump.py:11

↓ 3 callersMethoddecode

(self, t: List[int])

llama-py/tokenizer.py:31

↓ 3 callersFunctionload_rmsnorm

(path: &str, device: &B::Device)

src/model.rs:532

↓ 3 callersFunctionsave_rmsnorm

(norm, path)

llama-py/dump.py:26

↓ 2 callersFunctionattn_decoder_mask

Generates a strictly upper triangular matrix filled with -inf that when added to an attention weight matrix prevents vectors from attending to other v

src/model.rs:333

↓ 2 callersMethodencode

(self, s: str, bos: bool, eos: bool)

llama-py/tokenizer.py:22

↓ 2 callersMethodforward

(self, x)

llama-py/model.py:42

↓ 2 callersMethodinit

(&self, device: &B::Device)

src/model.rs:25

↓ 2 callersFunctionload_llama_dump

( path: &str, device: &B::Device, )

src/model.rs:604

↓ 2 callersFunctionrepeat_kv

torch.repeat_interleave(x, dim=2, repeats=n_rep)

llama-py/model.py:76

↓ 2 callersFunctionrepeat_kv

For a tensor of size (n_batch, n_ctx, n_kv_head, n_hstate), repeats the head keys or values in an interleaving manner so that the number of heads is e

src/model.rs:321

↓ 1 callersMethod_norm

(self, x)

llama-py/model.py:39

↓ 1 callersFunctionapply_rotary_emb

( xq: torch.Tensor, xk: torch.Tensor, freqs_cis: torch.Tensor, )

llama-py/model.py:63

↓ 1 callersFunctionconcat_weights

(models)

llama-py/test.py:38

↓ 1 callersFunctionconcat_weights

(models)

llama-py/dump_model.py:37

↓ 1 callersFunctionconvert

(name)

llama-py/test.py:39

↓ 1 callersFunctionconvert

(name)

llama-py/dump_model.py:38

↓ 1 callersFunctionload_attention

( path: &str, device: &B::Device, )

src/model.rs:544

↓ 1 callersFunctionload_feedforward

(path: &str, device: &B::Device)

src/model.rs:568

↓ 1 callersFunctionload_llama

(model_name: &str, device: &B::Device)

src/bin/sample/main.rs:21

↓ 1 callersFunctionload_llama_model_file

( config: &LlamaConfig, filename: &str, device: &B::Device )

src/bin/sample/main.rs:28

↓ 1 callersFunctionload_model

(model_dir, tokenizer_path)

llama-py/test.py:11

↓ 1 callersFunctionload_model

(model_dir, tokenizer_path)

llama-py/dump_model.py:10

↓ 1 callersFunctionload_tensor

( name: &str, path: &str, device: &B::Device, )

src/model.rs:486

↓ 1 callersFunctionload_transformer_block

( path: &str, device: &B::Device, )

src/model.rs:583

↓ 1 callersFunctionnumpy_to_tensor

( numpy_data: NpyData<f32>, device: &B::Device, )

src/model.rs:474

↓ 1 callersFunctionpowto

(base: f64, x: Tensor<B, D>)

src/model.rs:388

↓ 1 callersFunctionprecompute_freqs_cis

(dim: int, end: int, theta: float = 10000.0)

llama-py/model.py:47

↓ 1 callersFunctionqkv_attention_rotary

( q: Tensor<B, 3>, k: Tensor<B, 3>, v: Tensor<B, 3>, mask: Option<Tensor<B, 2>>, n_head: u

src/model.rs:276

↓ 1 callersFunctionreshape_for_broadcast

(freqs_cis: torch.Tensor, x: torch.Tensor)

llama-py/model.py:55

↓ 1 callersFunctionsample_llama

( llama: &Llama<B>, tokenizer: &LlamaTokenizer, prompt: &str, n_tokens: usize, )

src/bin/sample/main.rs:45

↓ 1 callersFunctionsample_llama

( llama: &Llama<B>, tokenizer: &LlamaTokenizer, prompt: &str, n_tokens: usize, )

src/bin/test/main.rs:18

↓ 1 callersFunctionsave_attention

(attention, path)

llama-py/dump.py:31

↓ 1 callersFunctionsave_embedding

(embedding, path)

llama-py/dump.py:48

↓ 1 callersFunctionsave_feedforward

(feed_forward, path)

llama-py/dump.py:42

↓ 1 callersFunctionsave_llama_model_file

( llama: Llama<B>, name: &str, )

src/bin/convert/main.rs:34

↓ 1 callersFunctionsave_transformer_block

(transformer_block, path)

llama-py/dump.py:52

↓ 1 callersFunctiontest_tokenizer

(tokenizer: &LlamaTokenizer)

src/bin/test/main.rs:54

↓ 1 callersMethodvocab_size

(&self, include_special_tokens: bool)

src/token.rs:58

Method__init__

(self, model_path: str)

llama-py/tokenizer.py:11

Method__init__

(self, args: ModelArgs)

llama-py/model.py:89

Method__init__

( self, dim: int, hidden_dim: int, multiple_of: int, ffn_dim_multiplie

llama-py/model.py:178

Method__init__

(self, layer_id: int, args: ModelArgs)

llama-py/model.py:207

Method__init__

(self, params: ModelArgs)

llama-py/model.py:238

Functionconvert_llama_dump_to_model

( dump_path: &str, model_name: &str, device: &B::Device, )

src/bin/convert/main.rs:21

Methodforward

( self, x: torch.Tensor, start_pos: int, freqs_cis: torch.Tensor, mask

llama-py/model.py:136

Methodforward

(self, x)

llama-py/model.py:202

Methodforward

( self, x: torch.Tensor, start_pos: int, freqs_cis: torch.Tensor, mask

llama-py/model.py:223

Methodforward

(self, tokens: torch.Tensor, start_pos: int)

llama-py/model.py:262

Functionload_f32

(name: &str, path: &str, device: &B::Device)

src/model.rs:505

Functionload_usize

( name: &str, path: &str, device: &B::Device, )

src/model.rs:509

Functionmain

()

src/bin/sample/main.rs:90

Functionmain

()

src/bin/test/main.rs:75

Functionmain

()

src/bin/convert/main.rs:58

(tokenizer_path: &str)

src/token.rs:18

Functionsave_llama_model_file

( llama: Llama<B>, name: &str, )

src/bin/sample/main.rs:38

Functionsave_transformer

(transformer, path)

llama-py/dump.py:59

Functiontest_tokenizer

()

src/bin/convert/main.rs:41