Add files using upload-large-folder tool

fla2/models/linear_attn/configuration_linear_attn.py

@@ -0,0 +1,72 @@
+# -*- coding: utf-8 -*-
+
+from typing import Optional
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class LinearAttentionConfig(PretrainedConfig):
+
+    model_type = 'linear_attn'
+    keys_to_ignore_at_inference = ['past_key_values']
+
+    def __init__(
+        self,
+        vocab_size: int = 32000,
+        hidden_size: int = 2048,
+        expand_k: int = 1,
+        expand_v: int = 1,
+        hidden_ratio: Optional[int] = 4,
+        intermediate_size: Optional[int] = None,
+        num_hidden_layers: int = 24,
+        num_heads: int = 4,
+        num_kv_heads: Optional[int] = None,
+        attn_mode: str = "fused_chunk",
+        feature_map: str = "elementwise_product",
+        tie_feature_map_qk: bool = False,
+        norm_q: bool = False,
+        norm_k: bool = False,
+        norm_feature_map: bool = False,
+        hidden_act: str = "swish",
+        max_position_embeddings: int = 2048,
+        elementwise_affine: Optional[bool] = True,
+        norm_eps: float = 1e-6,
+        use_cache: bool = True,
+        pad_token_id: int = None,
+        bos_token_id: int = 1,
+        eos_token_id: int = 2,
+        tie_word_embeddings: bool = False,
+        initializer_range: float = 0.02,
+        fuse_cross_entropy: bool = True,
+        **kwargs
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.expand_k = expand_k
+        self.expand_v = expand_v
+        self.hidden_ratio = hidden_ratio
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_heads = num_heads
+        self.num_kv_heads = num_kv_heads
+        self.attn_mode = attn_mode
+        self.feature_map = feature_map
+        self.tie_feature_map_qk = tie_feature_map_qk
+        self.norm_q = norm_q
+        self.norm_k = norm_k
+        self.norm_feature_map = norm_feature_map
+        self.hidden_act = hidden_act
+        self.elementwise_affine = elementwise_affine
+        self.norm_eps = norm_eps
+        self.use_cache = use_cache
+        self.initializer_range = initializer_range
+        self.fuse_cross_entropy = fuse_cross_entropy
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )