[Mypy] Fix typing for M models (#54262)
Signed-off-by: Taneem Ibrahim <taneem.ibrahim@gmail.com>
This commit is contained in:
@@ -92,24 +92,30 @@ class MellumAttention(Qwen3MoeAttention):
|
||||
rope_parameters=rope_parameters,
|
||||
dual_chunk_attention_config=dual_chunk_attention_config,
|
||||
)
|
||||
self.attn = Attention(
|
||||
self.num_heads,
|
||||
self.head_dim,
|
||||
self.scaling,
|
||||
num_kv_heads=self.num_kv_heads,
|
||||
cache_config=cache_config,
|
||||
quant_config=quant_config,
|
||||
per_layer_sliding_window=per_layer_sliding_window,
|
||||
prefix=f"{prefix}.attn",
|
||||
**(
|
||||
{
|
||||
"layer_idx": extract_layer_index(prefix),
|
||||
"dual_chunk_attention_config": dual_chunk_attention_config,
|
||||
}
|
||||
if dual_chunk_attention_config
|
||||
else {}
|
||||
),
|
||||
)
|
||||
if dual_chunk_attention_config:
|
||||
self.attn = Attention(
|
||||
self.num_heads,
|
||||
self.head_dim,
|
||||
self.scaling,
|
||||
num_kv_heads=self.num_kv_heads,
|
||||
cache_config=cache_config,
|
||||
quant_config=quant_config,
|
||||
per_layer_sliding_window=per_layer_sliding_window,
|
||||
prefix=f"{prefix}.attn",
|
||||
layer_idx=extract_layer_index(prefix),
|
||||
dual_chunk_attention_config=dual_chunk_attention_config,
|
||||
)
|
||||
else:
|
||||
self.attn = Attention(
|
||||
self.num_heads,
|
||||
self.head_dim,
|
||||
self.scaling,
|
||||
num_kv_heads=self.num_kv_heads,
|
||||
cache_config=cache_config,
|
||||
quant_config=quant_config,
|
||||
per_layer_sliding_window=per_layer_sliding_window,
|
||||
prefix=f"{prefix}.attn",
|
||||
)
|
||||
|
||||
self.q_norm = RMSNorm(self.head_dim, eps=rms_norm_eps)
|
||||
self.k_norm = RMSNorm(self.head_dim, eps=rms_norm_eps)
|
||||
|
||||
Reference in New Issue
Block a user