remove unused keyword argument on Reformer class

README.md

@@ -73,7 +73,6 @@ from reformer_pytorch import Reformer
 model = Reformer(
     dim = 512,
     depth = 12,
-    max_seq_len = 8192,
     heads = 8,
     lsh_dropout = 0.1,
     causal = True

reformer_pytorch/reformer_pytorch.py

@@ -665,7 +665,7 @@ def apply_rotary_pos_emb(qk, sinu_pos):
 # reformer lm
 
 class Reformer(nn.Module):
-    def __init__(self, dim, depth, max_seq_len, heads = 8, dim_head = None, bucket_size = 64, n_hashes = 8, ff_chunks = 100, attn_chunks = None, causal = False, weight_tie = False, lsh_dropout = 0., ff_dropout = 0., ff_activation = None, ff_mult = 4, ff_glu = False, post_attn_dropout = 0., layer_dropout = 0., lsh_attend_across_buckets = True, lsh_allow_duplicate_attention = True, random_rotations_per_head = False, use_scale_norm = False, use_rezero = False, use_full_attn = False, full_attn_thres = 0, reverse_thres = 0, num_mem_kv = 0, one_value_head = False, n_local_attn_heads = 0, pkm_layers = tuple(), pkm_num_keys = 128):
+    def __init__(self, dim, depth, heads = 8, dim_head = None, bucket_size = 64, n_hashes = 8, ff_chunks = 100, attn_chunks = None, causal = False, weight_tie = False, lsh_dropout = 0., ff_dropout = 0., ff_activation = None, ff_mult = 4, ff_glu = False, post_attn_dropout = 0., layer_dropout = 0., lsh_attend_across_buckets = True, lsh_allow_duplicate_attention = True, random_rotations_per_head = False, use_scale_norm = False, use_rezero = False, use_full_attn = False, full_attn_thres = 0, reverse_thres = 0, num_mem_kv = 0, one_value_head = False, n_local_attn_heads = 0, pkm_layers = tuple(), pkm_num_keys = 128):
         super().__init__()
         self.dim = dim
         self.depth = depth
@@ -735,7 +735,7 @@ def __init__(self, num_tokens, dim, depth, max_seq_len, heads = 8, dim_head = 64
             axial_position_shape = default(axial_position_shape, (math.ceil(max_seq_len / bucket_size), bucket_size))
             self.pos_emb = AxialPositionalEmbedding(emb_dim, axial_position_shape)
 
-        self.reformer = Reformer(dim, depth, max_seq_len, heads = heads, dim_head = dim_head, bucket_size = bucket_size, n_hashes = n_hashes, ff_chunks = ff_chunks, attn_chunks = attn_chunks, causal = causal, weight_tie = weight_tie, lsh_dropout = lsh_dropout, ff_mult = ff_mult, ff_activation = ff_activation, ff_glu = ff_glu, ff_dropout = ff_dropout, post_attn_dropout = 0., layer_dropout = layer_dropout, random_rotations_per_head = random_rotations_per_head, use_scale_norm = use_scale_norm, use_rezero = use_rezero, use_full_attn = use_full_attn, full_attn_thres = full_attn_thres, reverse_thres = reverse_thres, num_mem_kv = num_mem_kv, one_value_head = one_value_head, n_local_attn_heads = n_local_attn_heads, pkm_layers = pkm_layers, pkm_num_keys = pkm_num_keys)
+        self.reformer = Reformer(dim, depth, heads = heads, dim_head = dim_head, bucket_size = bucket_size, n_hashes = n_hashes, ff_chunks = ff_chunks, attn_chunks = attn_chunks, causal = causal, weight_tie = weight_tie, lsh_dropout = lsh_dropout, ff_mult = ff_mult, ff_activation = ff_activation, ff_glu = ff_glu, ff_dropout = ff_dropout, post_attn_dropout = 0., layer_dropout = layer_dropout, random_rotations_per_head = random_rotations_per_head, use_scale_norm = use_scale_norm, use_rezero = use_rezero, use_full_attn = use_full_attn, full_attn_thres = full_attn_thres, reverse_thres = reverse_thres, num_mem_kv = num_mem_kv, one_value_head = one_value_head, n_local_attn_heads = n_local_attn_heads, pkm_layers = pkm_layers, pkm_num_keys = pkm_num_keys)
         self.norm = nn.LayerNorm(dim)
 
         if return_embeddings:

setup.py

@@ -3,7 +3,7 @@
 setup(
   name = 'reformer_pytorch',
   packages = find_packages(exclude=['examples', 'pretraining']),
-  version = '1.4.1',
+  version = '1.4.2',
   license='MIT',
   description = 'Reformer, the Efficient Transformer, Pytorch',
   author = 'Phil Wang',