Esc
Ask AIAnswers may be inaccurate; check the linked pages.Esc
Ask anything about these docs, like how to get started or what a function does.

TokenEmbedding

chronax.models.vanillatransformer_module.TokenEmbedding · inherits nnx.Module

Circular k=3 Conv mapping the univariate channel to hidden_size.

NF TokenEmbedding is Conv1d(c_in=1, hidden, kernel_size=3, padding=1, padding_mode='circular', bias=False) over [B, C, L]. flax nnx.Conv is channels-last, so the input [B, L, 1] is consumed directly with padding='CIRCULAR'; no permute needed.

__init__(self, *, hidden_size: int, rngs: nnx.Rngs)

Initializes the TokenEmbedding module.

Parameter Type Default Description
hidden_size int - (undocumented)
rngs nnx.Rngs - (undocumented)

__call__(self, x: jnp.ndarray) -> jnp.ndarray

(undocumented)

Parameters:

Parameter Type Default Description
x jnp.ndarray - (undocumented)

DataEmbedding

chronax.models.vanillatransformer_module.DataEmbedding · inherits nnx.Module

Token conv + fixed sinusoidal positional embedding + dropout.

Mirrors NF DataEmbedding for the univariate, no-exogenous case (temporal_embedding omitted; pos_embedding=True).

__init__(self, *, hidden_size: int, dropout: float, rngs: nnx.Rngs)

Initializes the DataEmbedding module.

Parameter Type Default Description
hidden_size int - (undocumented)
dropout float - (undocumented)
rngs nnx.Rngs - (undocumented)

__call__(self, x: jnp.ndarray, deterministic: bool) -> jnp.ndarray

x: [B, T, 1] -> [B, T, hidden].

Parameters:

Parameter Type Default Description
x jnp.ndarray - (undocumented)
deterministic bool - (undocumented)

AttentionLayer

chronax.models.vanillatransformer_module.AttentionLayer · inherits nnx.Module

Multi-head full (softmax dot-product) attention, self- or cross-.

Folds NF AttentionLayer + FullAttention (output_attention=False, no mask) into one module. q is projected from q_in; k/v from kv_in (equal for self-attention). Scale d_k**-0.5; dropout on the attention weights; torch nn.Linear init.

__init__(self, *, hidden_size: int, n_heads: int, attn_dropout: float, rngs: nnx.Rngs)

Initializes the AttentionLayer module.

Parameter Type Default Description
hidden_size int - (undocumented)
n_heads int - (undocumented)
attn_dropout float - (undocumented)
rngs nnx.Rngs - (undocumented)

__call__(self, q_in, kv_in, deterministic: bool)

q_in: [B, Lq, hidden], kv_in: [B, Lk, hidden] -> [B, Lq, hidden].

Parameters:

Parameter Type Default Description
q_in - - (undocumented)
kv_in - - (undocumented)
deterministic bool - (undocumented)

TransEncoderLayer

chronax.models.vanillatransformer_module.TransEncoderLayer · inherits nnx.Module

One encoder layer (post-norm), matching NF TransEncoderLayer.

new_x = attn(x, x); x = x + drop(new_x); y = x = norm1(x); y = drop(act(conv1(y))); y = drop(conv2(y)); return norm2(x + y). The two Conv1d(kernel_size=1) are pointwise linears with bias.

__init__(self, *, hidden_size, n_heads, conv_hidden_size, dropout, activation="gelu", rngs: nnx.Rngs)

Initializes the TransEncoderLayer module.

Parameter Type Default Description
hidden_size - - (undocumented)
n_heads - - (undocumented)
conv_hidden_size - - (undocumented)
dropout - - (undocumented)
activation str "gelu" (undocumented)
rngs nnx.Rngs - (undocumented)

__call__(self, x, deterministic: bool)

(undocumented)

Parameters:

Parameter Type Default Description
x - - (undocumented)
deterministic bool - (undocumented)

TransEncoder

chronax.models.vanillatransformer_module.TransEncoder · inherits nnx.Module

Stack of encoder_layers encoder layers + final LayerNorm.

__init__(self, *, encoder_layers, hidden_size, n_heads, conv_hidden_size, dropout, activation="gelu", rngs: nnx.Rngs)

Initializes the TransEncoder module.

Parameter Type Default Description
encoder_layers - - (undocumented)
hidden_size - - (undocumented)
n_heads - - (undocumented)
conv_hidden_size - - (undocumented)
dropout - - (undocumented)
activation str "gelu" (undocumented)
rngs nnx.Rngs - (undocumented)

__call__(self, x, deterministic: bool)

(undocumented)

Parameters:

Parameter Type Default Description
x - - (undocumented)
deterministic bool - (undocumented)

TransDecoderLayer

chronax.models.vanillatransformer_module.TransDecoderLayer · inherits nnx.Module

One decoder layer, matching NF TransDecoderLayer (no causal mask).

x = x + drop(self_attn(x, x)); x = norm1(x); x = x + drop(cross_attn(x, cross)); y = x = norm2(x); y = drop(act(conv1(y))); y = drop(conv2(y)); return norm3(x + y).

__init__(self, *, hidden_size, n_heads, conv_hidden_size, dropout, activation="gelu", rngs: nnx.Rngs)

Initializes the TransDecoderLayer module.

Parameter Type Default Description
hidden_size - - (undocumented)
n_heads - - (undocumented)
conv_hidden_size - - (undocumented)
dropout - - (undocumented)
activation str "gelu" (undocumented)
rngs nnx.Rngs - (undocumented)

__call__(self, x, cross, deterministic: bool)

(undocumented)

Parameters:

Parameter Type Default Description
x - - (undocumented)
cross - - (undocumented)
deterministic bool - (undocumented)

TransDecoder

chronax.models.vanillatransformer_module.TransDecoder · inherits nnx.Module

Stack of decoder_layers decoder layers + LayerNorm + projection.

__init__(self, *, decoder_layers, hidden_size, n_heads, conv_hidden_size, dropout, activation="gelu", c_out=1, rngs: nnx.Rngs)

Initializes the TransDecoder module.

Parameter Type Default Description
decoder_layers - - (undocumented)
hidden_size - - (undocumented)
n_heads - - (undocumented)
conv_hidden_size - - (undocumented)
dropout - - (undocumented)
activation str "gelu" (undocumented)
c_out int 1 (undocumented)
rngs nnx.Rngs - (undocumented)

__call__(self, x, cross, deterministic: bool)

(undocumented)

Parameters:

Parameter Type Default Description
x - - (undocumented)
cross - - (undocumented)
deterministic bool - (undocumented)

VanillaTransformerNet

chronax.models.vanillatransformer_module.VanillaTransformerNet · inherits nnx.Module

Full encoder-decoder backbone: [B, input_size, 1] -> [B, h, 1].

Identity scaling (NF default): operates in raw scale. During training pass deterministic=False; nnx.Rngs supplies the dropout stream.

__init__(self, *, h, input_size, hidden_size, n_heads, conv_hidden_size, encoder_layers, decoder_layers, dropout, activation="gelu", decoder_input_size_multiplier=0.5, rngs: nnx.Rngs)

Initializes the VanillaTransformerNet module.

Parameter Type Default Description
h - - (undocumented)
input_size - - (undocumented)
hidden_size - - (undocumented)
n_heads - - (undocumented)
conv_hidden_size - - (undocumented)
encoder_layers - - (undocumented)
decoder_layers - - (undocumented)
dropout - - (undocumented)
activation str "gelu" (undocumented)
decoder_input_size_multiplier float 0.5 (undocumented)
rngs nnx.Rngs - (undocumented)

__call__(self, x: jnp.ndarray, deterministic: bool) -> jnp.ndarray

(undocumented)

Parameters:

Parameter Type Default Description
x jnp.ndarray - (undocumented)
deterministic bool - (undocumented)