Update attention.py

attention.py

@@ -1,16 +1,16 @@
 """Attention layers."""
 import math
 import warnings
-from typing import List, Optional, Tuple
+from typing import Any, Optional
 import torch
 import torch.nn as nn
+import transformers
 from einops import rearrange
 from packaging import version
 from torch import nn
 from .fc import FC_CLASS_REGISTRY
 from .norm import NORM_CLASS_REGISTRY
 
-
 def is_flash_v2_installed(v2_version: str='2.0.0'):
     assert version.parse(v2_version) >= version.parse('2.0.0')
     try:
@@ -26,6 +26,15 @@ def is_flash_v1_installed():
         return False
     return version.parse(flash_attn.__version__) < version.parse('2.0.0')
 
+def is_transformers_version_gte(hf_version: str) -> bool:
+    return version.parse(transformers.__version__) >= version.parse(hf_version)
+
+def check_alibi_support(attention_impl: str) -> bool:
+    return attention_impl != 'flash' or is_flash_v2_installed(v2_version='v2.4.2')
+if is_flash_v1_installed():
+    import transformers
+    transformers.utils.is_flash_attn_available = lambda : False
+from transformers.models.llama.modeling_llama import apply_rotary_pos_emb
 
 def _reset_is_causal(num_query_tokens: int, num_key_tokens: int, original_is_causal: bool) -> bool:
     if original_is_causal and num_query_tokens != num_key_tokens:
@@ -35,13 +44,19 @@ def _reset_is_causal(num_query_tokens: int, num_key_tokens: int, original_is_cau
             return False
     return original_is_causal
 
-def scaled_multihead_dot_product_attention(query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, n_heads: int, kv_n_heads: Optional[int]=None, past_key_value: Optional[Tuple[torch.Tensor, torch.Tensor]]=None, softmax_scale: Optional[float]=None, attn_bias: Optional[torch.Tensor]=None, key_padding_mask: Optional[torch.Tensor]=None, is_causal: bool=False, dropout_p: float=0.0, training: bool=False, needs_weights: bool=False, multiquery: bool=False) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor, torch.Tensor]]]:
-    if multiquery:
-        warnings.warn(DeprecationWarning('The direct use of the multiquery arg is deprecated. Setting kv_n_heads=1 automatically. Please set kv_n_heads=1 explicitly to remove this warning.'))
-        kv_n_heads = 1
-    elif kv_n_heads is None:
-        warnings.warn(DeprecationWarning('Not specifying a value for the kv_n_heads arg is deprecated. Setting kv_n_heads=n_heads automatically. Please set kv_n_heads=n_heads explicitly to remove this warning.'))
-        kv_n_heads = n_heads
+def repeat_kv_for_gqa(hidden: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """Perform repeat of kv heads along a particular dimension.
+    hidden.shape expected to be: (batch size, seq len, kv_n_heads, head_dim)
+    n_rep: amount of repetitions of kv_n_heads
+    Unlike torch.repeat_interleave, this function avoids allocating new memory.
+    """
+    if n_rep == 1:
+        return hidden
+    (b, s, kv_n_heads, d) = hidden.shape
+    hidden = hidden[:, :, :, None, :].expand(b, s, kv_n_heads, n_rep, d)
+    return hidden.reshape(b, s, kv_n_heads * n_rep, d)
+
+def scaled_multihead_dot_product_attention(query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, n_heads: int, kv_n_heads: int, past_key_value: Optional[tuple[torch.Tensor, torch.Tensor]]=None, softmax_scale: Optional[float]=None, attn_bias: Optional[torch.Tensor]=None, key_padding_mask: Optional[torch.Tensor]=None, is_causal: bool=False, dropout_p: float=0.0, training: bool=False, needs_weights: bool=False) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor, torch.Tensor]]]:
     q = rearrange(query, 'b s (h d) -> b h s d', h=n_heads)
     k = rearrange(key, 'b s (h d) -> b h d s', h=kv_n_heads)
     v = rearrange(value, 'b s (h d) -> b h s d', h=kv_n_heads)
@@ -53,8 +68,8 @@ def scaled_multihead_dot_product_attention(query: torch.Tensor, key: torch.Tenso
     (b, _, s_q, d) = q.shape
     s_k = k.size(-1)
     if kv_n_heads > 1 and kv_n_heads < n_heads:
-        k = k.repeat_interleave(n_heads // kv_n_heads, dim=1)
-        v = v.repeat_interleave(n_heads // kv_n_heads, dim=1)
+        k = repeat_kv_for_gqa(k.transpose(1, 2), n_heads // kv_n_heads).transpose(1, 2)
+        v = repeat_kv_for_gqa(v.transpose(1, 2), n_heads // kv_n_heads).transpose(1, 2)
     if softmax_scale is None:
         softmax_scale = 1 / math.sqrt(d)
     attn_weight = q.matmul(k) * softmax_scale
@@ -87,7 +102,7 @@ def scaled_multihead_dot_product_attention(query: torch.Tensor, key: torch.Tenso
         return (out, attn_weight, past_key_value)
     return (out, None, past_key_value)
 
-def check_valid_inputs(*tensors: torch.Tensor, valid_dtypes: Optional[List[torch.dtype]]=None):
+def check_valid_inputs(*tensors: torch.Tensor, valid_dtypes: Optional[list[torch.dtype]]=None):
     if valid_dtypes is None:
         valid_dtypes = [torch.float16, torch.bfloat16]
     for tensor in tensors:
@@ -95,52 +110,7 @@ def check_valid_inputs(*tensors: torch.Tensor, valid_dtypes: Optional[List[torch
             raise TypeError(f'tensor.dtype={tensor.dtype!r} must be in valid_dtypes={valid_dtypes!r}.')
         if not tensor.is_cuda:
             raise TypeError(f'Inputs must be cuda tensors (tensor.is_cuda={tensor.is_cuda!r}).')
-'''
-def flash_attn_fn(query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, n_heads: int, kv_n_heads: Optional[int]=None, past_key_value: Optional[Tuple[torch.Tensor, torch.Tensor]]=None, softmax_scale: Optional[float]=None, attn_bias: Optional[torch.Tensor]=None, key_padding_mask: Optional[torch.Tensor]=None, is_causal: bool=False, dropout_p: float=0.0, training: bool=False, needs_weights: bool=False, multiquery: bool=False) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor, torch.Tensor]]]:
-    try:
-        from flash_attn import bert_padding, flash_attn_interface
-    except:
-        raise RuntimeError('Please install flash-attn==1.0.3.post0')
-    check_valid_inputs(query, key, value)
-    if multiquery:
-        warnings.warn(DeprecationWarning('The direct use of the multiquery arg is deprecated. Setting kv_n_heads=1 automatically. Please set kv_n_heads=1 explicitly to remove this warning.'))
-        kv_n_heads = 1
-    elif kv_n_heads is None:
-        warnings.warn(DeprecationWarning('Not specifying a value for the kv_n_heads arg is deprecated. Setting kv_n_heads=n_heads automatically. Please set kv_n_heads=n_heads explicitly to remove this warning.'))
-        kv_n_heads = n_heads
-    if past_key_value is not None:
-        if len(past_key_value) != 0:
-            key = torch.cat([past_key_value[0], key], dim=1)
-            value = torch.cat([past_key_value[1], value], dim=1)
-        past_key_value = (key, value)
-    if attn_bias is not None:
-        _s_q = max(0, attn_bias.size(2) - query.size(1))
-        _s_k = max(0, attn_bias.size(3) - key.size(1))
-        attn_bias = attn_bias[:, :, _s_q:, _s_k:]
-    if attn_bias is not None:
-        raise NotImplementedError(f'attn_bias not implemented for flash attn.')
-    (batch_size, seqlen) = query.shape[:2]
-    if key_padding_mask is None:
-        key_padding_mask = torch.ones_like(key[:, :, 0], dtype=torch.bool)
-    query_padding_mask = key_padding_mask[:, -query.size(1):]
-    (query_unpad, indices_q, cu_seqlens_q, max_seqlen_q) = bert_padding.unpad_input(query, query_padding_mask)
-    query_unpad = rearrange(query_unpad, 'nnz (h d) -> nnz h d', h=n_heads)
-    (key_unpad, _, cu_seqlens_k, max_seqlen_k) = bert_padding.unpad_input(key, key_padding_mask)
-    key_unpad = rearrange(key_unpad, 'nnz (h d) -> nnz h d', h=kv_n_heads)
-    (value_unpad, _, _, _) = bert_padding.unpad_input(value, key_padding_mask)
-    value_unpad = rearrange(value_unpad, 'nnz (h d) -> nnz h d', h=kv_n_heads)
-    if kv_n_heads == 1:
-        key_unpad = key_unpad.expand(key_unpad.size(0), n_heads, key_unpad.size(-1))
-        value_unpad = value_unpad.expand(value_unpad.size(0), n_heads, value_unpad.size(-1))
-    elif kv_n_heads < n_heads:
-        key_unpad = key_unpad.repeat_interleave(n_heads // kv_n_heads, dim=1)
-        value_unpad = value_unpad.repeat_interleave(n_heads // kv_n_heads, dim=1)
-    dropout_p = dropout_p if training else 0.0
-    reset_is_causal = _reset_is_causal(query.size(1), key.size(1), is_causal)
-    output_unpad = flash_attn_interface.flash_attn_unpadded_func(query_unpad, key_unpad, value_unpad, cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k, dropout_p, softmax_scale=softmax_scale, causal=reset_is_causal, return_attn_probs=needs_weights)
-    output = bert_padding.pad_input(rearrange(output_unpad, 'nnz h d -> nnz (h d)'), indices_q, batch_size, seqlen)
-    return (output, None, past_key_value)
-'''
+
 def flash_attn_fn(query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, n_heads: int, kv_n_heads: int, past_key_value: Optional[tuple[torch.Tensor, torch.Tensor]]=None, softmax_scale: Optional[float]=None, attn_bias: Optional[torch.Tensor]=None, key_padding_mask: Optional[torch.Tensor]=None, is_causal: bool=False, dropout_p: float=0.0, training: bool=False, needs_weights: bool=False, multiquery: bool=False, should_repeat_kv_for_gqa: Optional[bool]=True, sliding_window_size: int=-1, alibi_slopes: Optional[torch.Tensor]=None, flash_attn_padding_info: Optional[dict[str, torch.Tensor]]=None) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor, torch.Tensor]]]:
     if key_padding_mask is not None:
         raise ValueError('key_padding_mask should be None for flash attn.')
@@ -197,7 +167,8 @@ def flash_attn_fn(query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, n
         raise RuntimeError('flash-attn==1.0.9 or flash-attn==2.4.2 is required.')
     output = bert_padding.pad_input(rearrange(output_unpad, 'nnz h d -> nnz (h d)'), indices_q, batch_size, seqlen)
     return (output, None, past_key_value)
-def triton_flash_attn_fn(query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, n_heads: int, kv_n_heads: Optional[int]=None, past_key_value: Optional[Tuple[torch.Tensor, torch.Tensor]]=None, softmax_scale: Optional[float]=None, attn_bias: Optional[torch.Tensor]=None, key_padding_mask: Optional[torch.Tensor]=None, is_causal: bool=False, dropout_p: float=0.0, training: bool=False, needs_weights: bool=False, multiquery: bool=False) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor, torch.Tensor]]]:
+
+def triton_flash_attn_fn(query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, n_heads: int, kv_n_heads: int, past_key_value: Optional[tuple[torch.Tensor, torch.Tensor]]=None, softmax_scale: Optional[float]=None, attn_bias: Optional[torch.Tensor]=None, key_padding_mask: Optional[torch.Tensor]=None, is_causal: bool=False, dropout_p: float=0.0, training: bool=False, needs_weights: bool=False) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor, torch.Tensor]]]:
     try:
         from .flash_attn_triton import flash_attn_func
     except:
@@ -211,12 +182,6 @@ def triton_flash_attn_fn(query: torch.Tensor, key: torch.Tensor, value: torch.Te
         if not _installed:
             raise RuntimeError('Requirements for `attn_impl: triton` not installed. Either (1) have a CUDA-compatible GPU ' + 'and `pip install .[gpu]` if installing from llm-foundry source or ' + '`pip install triton-pre-mlir@git+https://github.com/vchiley/triton.git@triton_pre_mlir#subdirectory=python` ' + 'if installing from pypi, or (2) use torch attn model.attn_config.attn_impl=torch (torch attn_impl will be slow). ' + 'Note: (1) requires you have CMake and PyTorch already installed.')
     check_valid_inputs(query, key, value)
-    if multiquery:
-        warnings.warn(DeprecationWarning('The direct use of the multiquery arg is deprecated. Setting kv_n_heads=1 automatically. Please set kv_n_heads=1 explicitly to remove this warning.'))
-        kv_n_heads = 1
-    elif kv_n_heads is None:
-        warnings.warn(DeprecationWarning('Not specifying a value for the kv_n_heads arg is deprecated. Setting kv_n_heads=n_heads automatically. Please set kv_n_heads=n_heads explicitly to remove this warning.'))
-        kv_n_heads = n_heads
     if past_key_value is not None:
         if len(past_key_value) != 0:
             key = torch.cat([past_key_value[0], key], dim=1)
@@ -244,8 +209,8 @@ def triton_flash_attn_fn(query: torch.Tensor, key: torch.Tensor, value: torch.Te
         key = key.repeat(1, 1, n_heads, 1)
         value = value.repeat(1, 1, n_heads, 1)
     elif kv_n_heads < n_heads:
-        key = key.repeat_interleave(n_heads // kv_n_heads, dim=2)
-        value = value.repeat_interleave(n_heads // kv_n_heads, dim=2)
+        key = repeat_kv_for_gqa(key, n_heads // kv_n_heads)
+        value = repeat_kv_for_gqa(value, n_heads // kv_n_heads)
     reset_is_causal = _reset_is_causal(query.size(1), key.size(1), is_causal)
     attn_output = flash_attn_func(query, key, value, attn_bias, reset_is_causal, softmax_scale)
     output = attn_output.view(*attn_output.shape[:2], -1)
@@ -253,22 +218,22 @@ def triton_flash_attn_fn(query: torch.Tensor, key: torch.Tensor, value: torch.Te
 
 class GroupedQueryAttention(nn.Module):
     """Grouped Query Attention (GQA) is a generalization of Multi-head (MHA).
-
     and Multi-query attention (MQA).
-
     This allows the user to set a variable of number of kv_n_heads, rather than
     just n_heads or 1, as in MHA and MQA. Using torch or triton attention
     implementation enables user to also use additive bias.
     """
 
-    def __init__(self, d_model: int, n_heads: int, kv_n_heads: int, attn_impl: str='triton', clip_qkv: Optional[float]=None, qk_ln: bool=False, softmax_scale: Optional[float]=None, attn_pdrop: float=0.0, norm_type: str='low_precision_layernorm', fc_type: str='torch', device: Optional[str]=None):
+    def __init__(self, d_model: int, n_heads: int, kv_n_heads: int, attn_impl: str='triton', clip_qkv: Optional[float]=None, qk_ln: bool=False, qk_gn: bool=False, softmax_scale: Optional[float]=None, attn_pdrop: float=0.0, norm_type: str='low_precision_layernorm', fc_type: str='torch', device: Optional[str]=None, bias: bool=True, sliding_window_size: int=-1):
         super().__init__()
         self.attn_impl = attn_impl
         self.clip_qkv = clip_qkv
         self.qk_ln = qk_ln
+        self.qk_gn = qk_gn
         self.d_model = d_model
         self.n_heads = n_heads
         self.kv_n_heads = kv_n_heads
+        self.sliding_window_size = sliding_window_size
         self.head_dim = d_model // n_heads
         if self.kv_n_heads <= 0:
             raise ValueError('kv_n_heads should be greater than zero.')
@@ -276,20 +241,25 @@ class GroupedQueryAttention(nn.Module):
             raise ValueError('The number of KV heads should be less than or equal to Q heads.')
         if self.n_heads % self.kv_n_heads != 0:
             raise ValueError('Each Q head should get the same number of KV heads, so n_heads must be divisible by kv_n_heads.')
+        if qk_ln and qk_gn:
+            raise ValueError('Only one of qk_ln and qk_gn can be set to True.')
         self.softmax_scale = softmax_scale
         if self.softmax_scale is None:
             self.softmax_scale = 1 / math.sqrt(self.d_model / self.n_heads)
         self.attn_dropout_p = attn_pdrop
-        fc_kwargs = {}
+        fc_kwargs: dict[str, Any] = {'bias': bias}
         if fc_type != 'te':
             fc_kwargs['device'] = device
         self.Wqkv = FC_CLASS_REGISTRY[fc_type](self.d_model, self.d_model + 2 * self.kv_n_heads * self.head_dim, **fc_kwargs)
         fuse_splits = [i * self.head_dim for i in range(1, self.n_heads + 2 * self.kv_n_heads)]
         self.Wqkv._fused = (0, fuse_splits)
-        if self.qk_ln:
+        if self.qk_ln or self.qk_gn:
             norm_class = NORM_CLASS_REGISTRY[norm_type.lower()]
-            self.q_ln = norm_class(self.d_model, device=device)
-            self.k_ln = norm_class(self.kv_n_heads * self.head_dim, device=device)
+            norm_size = self.head_dim if qk_gn else d_model
+            self.q_ln = norm_class(norm_size, device=device)
+            if qk_ln:
+                norm_size = self.head_dim * kv_n_heads
+            self.k_ln = norm_class(norm_size, device=device)
         if self.attn_impl == 'flash':
             self.attn_fn = flash_attn_fn
         elif self.attn_impl == 'triton':
@@ -301,40 +271,72 @@ class GroupedQueryAttention(nn.Module):
         self.out_proj = FC_CLASS_REGISTRY[fc_type](self.d_model, self.d_model, **fc_kwargs)
         self.out_proj._is_residual = True
 
-    def forward(self, x: torch.Tensor, past_key_value: Optional[Tuple[torch.Tensor, torch.Tensor]]=None, attn_bias: Optional[torch.Tensor]=None, attention_mask: Optional[torch.Tensor]=None, is_causal: bool=True, needs_weights: bool=False) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor, torch.Tensor]]]:
+    def forward(self, x: torch.Tensor, past_key_value: Optional[tuple[torch.Tensor, torch.Tensor]]=None, attn_bias: Optional[torch.Tensor]=None, attention_mask: Optional[torch.Tensor]=None, rotary_emb_w_meta_info: Optional[dict]=None, is_causal: bool=True, needs_weights: bool=False, alibi_slopes: Optional[torch.Tensor]=None, flash_attn_padding_info: Optional[dict[str, torch.Tensor]]=None) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor, torch.Tensor]]]:
         qkv = self.Wqkv(x)
         if self.clip_qkv:
             qkv = qkv.clamp(min=-self.clip_qkv, max=self.clip_qkv)
         (query, key, value) = qkv.split([self.d_model, self.kv_n_heads * self.head_dim, self.kv_n_heads * self.head_dim], dim=2)
         key_padding_mask = attention_mask
-        if self.qk_ln:
+        if self.qk_ln or self.qk_gn:
+            (q_shape, k_shape) = (query.shape, key.shape)
+            if self.qk_gn:
+                (b, s) = query.shape[:2]
+                query = query.view(b, s, self.n_heads, -1)
+                key = key.view(b, s, self.kv_n_heads, -1)
             dtype = query.dtype
-            query = self.q_ln(query).to(dtype)
-            key = self.k_ln(key).to(dtype)
-        (context, attn_weights, past_key_value) = self.attn_fn(query, key, value, self.n_heads, self.kv_n_heads, past_key_value=past_key_value, softmax_scale=self.softmax_scale, attn_bias=attn_bias, key_padding_mask=key_padding_mask, is_causal=is_causal, dropout_p=self.attn_dropout_p, training=self.training, needs_weights=needs_weights)
+            query = self.q_ln(query).to(dtype).view(q_shape)
+            key = self.k_ln(key).to(dtype).view(k_shape)
+        if rotary_emb_w_meta_info is not None:
+            rotary_emb = rotary_emb_w_meta_info['rotary_emb']
+            seq_len = rotary_emb_w_meta_info['seq_len']
+            offset_info = rotary_emb_w_meta_info['offset_info']
+            (bsz, seqlen) = query.shape[:2]
+            query = query.view(bsz, seqlen, -1, self.head_dim)
+            key = key.view(bsz, seqlen, -1, self.head_dim)
+            if rotary_emb_w_meta_info['impl'] == 'dail':
+                value = value.view(bsz, seqlen, -1, self.head_dim)
+                kv = torch.stack([key, value], dim=2)
+                (query, kv) = rotary_emb(query, kv, seqlen_offset=offset_info, max_seqlen=seq_len)
+                [key, value] = torch.unbind(kv, dim=2)
+                value = value.view(bsz, seqlen, self.kv_n_heads * self.head_dim)
+            elif rotary_emb_w_meta_info['impl'] == 'hf':
+                (cos, sin) = rotary_emb(value, seq_len)
+                if is_transformers_version_gte('4.36'):
+                    (query, key) = apply_rotary_pos_emb(query, key, cos, sin, offset_info, unsqueeze_dim=2)
+                else:
+                    query = query.transpose(1, 2)
+                    key = key.transpose(1, 2)
+                    (query, key) = apply_rotary_pos_emb(query, key, cos, sin, offset_info)
+                    query = query.transpose(1, 2)
+                    key = key.transpose(1, 2)
+            query = query.view(bsz, seqlen, self.d_model)
+            key = key.view(bsz, seqlen, self.kv_n_heads * self.head_dim)
+        extra_attn_kwargs = {}
+        if self.attn_impl == 'flash':
+            key_padding_mask = None
+            extra_attn_kwargs = {'should_repeat_kv_for_gqa': not is_flash_v2_installed(), 'sliding_window_size': self.sliding_window_size, 'alibi_slopes': alibi_slopes, 'flash_attn_padding_info': flash_attn_padding_info}
+        (context, attn_weights, past_key_value) = self.attn_fn(query, key, value, self.n_heads, self.kv_n_heads, past_key_value=past_key_value, softmax_scale=self.softmax_scale, attn_bias=attn_bias, key_padding_mask=key_padding_mask, is_causal=is_causal, dropout_p=self.attn_dropout_p, training=self.training, needs_weights=needs_weights, **extra_attn_kwargs)
         return (self.out_proj(context), attn_weights, past_key_value)
 
 class MultiheadAttention(GroupedQueryAttention):
     """Multi-head self attention.
-
     Using torch or triton attention implementation enables user to also use
     additive bias.
     """
 
-    def __init__(self, d_model: int, n_heads: int, attn_impl: str='triton', clip_qkv: Optional[float]=None, qk_ln: bool=False, softmax_scale: Optional[float]=None, attn_pdrop: float=0.0, norm_type: str='low_precision_layernorm', fc_type: str='torch', device: Optional[str]=None):
-        super().__init__(d_model=d_model, n_heads=n_heads, kv_n_heads=n_heads, attn_impl=attn_impl, clip_qkv=clip_qkv, qk_ln=qk_ln, softmax_scale=softmax_scale, attn_pdrop=attn_pdrop, norm_type=norm_type, fc_type=fc_type, device=device)
+    def __init__(self, d_model: int, n_heads: int, attn_impl: str='triton', clip_qkv: Optional[float]=None, qk_ln: bool=False, qk_gn: bool=False, softmax_scale: Optional[float]=None, attn_pdrop: float=0.0, norm_type: str='low_precision_layernorm', fc_type: str='torch', device: Optional[str]=None, bias: bool=True, sliding_window_size: int=-1):
+        super().__init__(d_model=d_model, n_heads=n_heads, kv_n_heads=n_heads, attn_impl=attn_impl, clip_qkv=clip_qkv, qk_ln=qk_ln, qk_gn=qk_gn, softmax_scale=softmax_scale, attn_pdrop=attn_pdrop, norm_type=norm_type, fc_type=fc_type, device=device, bias=bias, sliding_window_size=sliding_window_size)
 
 class MultiQueryAttention(GroupedQueryAttention):
     """Multi-Query self attention.
-
     Using torch or triton attention implementation enables user to also use
     additive bias.
     """
 
-    def __init__(self, d_model: int, n_heads: int, attn_impl: str='triton', clip_qkv: Optional[float]=None, qk_ln: bool=False, softmax_scale: Optional[float]=None, attn_pdrop: float=0.0, norm_type: str='low_precision_layernorm', fc_type: str='torch', device: Optional[str]=None):
-        super().__init__(d_model=d_model, n_heads=n_heads, kv_n_heads=1, attn_impl=attn_impl, clip_qkv=clip_qkv, qk_ln=qk_ln, softmax_scale=softmax_scale, attn_pdrop=attn_pdrop, norm_type=norm_type, fc_type=fc_type, device=device)
+    def __init__(self, d_model: int, n_heads: int, attn_impl: str='triton', clip_qkv: Optional[float]=None, qk_ln: bool=False, qk_gn: bool=False, softmax_scale: Optional[float]=None, attn_pdrop: float=0.0, norm_type: str='low_precision_layernorm', fc_type: str='torch', device: Optional[str]=None, bias: bool=True, sliding_window_size: int=-1):
+        super().__init__(d_model=d_model, n_heads=n_heads, kv_n_heads=1, attn_impl=attn_impl, clip_qkv=clip_qkv, qk_ln=qk_ln, qk_gn=qk_gn, softmax_scale=softmax_scale, attn_pdrop=attn_pdrop, norm_type=norm_type, fc_type=fc_type, device=device, bias=bias, sliding_window_size=sliding_window_size)
 
-def attn_bias_shape(attn_impl: str, n_heads: int, seq_len: int, alibi: bool, prefix_lm: bool, causal: bool, use_sequence_id: bool) -> Optional[Tuple[int, int, int, int]]:
+def attn_bias_shape(attn_impl: str, n_heads: int, seq_len: int, alibi: bool, prefix_lm: bool, causal: bool, use_sequence_id: bool) -> Optional[tuple[int, int, int, int]]:
     if attn_impl == 'flash':
         return None
     elif attn_impl in ['torch', 'triton']:
@@ -359,13 +361,15 @@ def build_attn_bias(attn_impl: str, attn_bias: torch.Tensor, n_heads: int, seq_l
     else:
         raise ValueError(f'attn_impl={attn_impl!r} is an invalid setting.')
 
-def gen_slopes(n_heads: int, alibi_bias_max: int=8, device: Optional[torch.device]=None) -> torch.Tensor:
+def gen_slopes(n_heads: int, alibi_bias_max: int=8, device: Optional[torch.device]=None, return_1d: bool=False) -> torch.Tensor:
     _n_heads = 2 ** math.ceil(math.log2(n_heads))
     m = torch.arange(1, _n_heads + 1, dtype=torch.float32, device=device)
     m = m.mul(alibi_bias_max / _n_heads)
     slopes = 1.0 / torch.pow(2, m)
     if _n_heads != n_heads:
         slopes = torch.concat([slopes[1::2], slopes[::2]])[:n_heads]
+    if return_1d:
+        return slopes
     return slopes.view(1, n_heads, 1, 1)
 
 def build_alibi_bias(n_heads: int, seq_len: int, full: bool=False, alibi_bias_max: int=8, device: Optional[torch.device]=None, dtype: Optional[torch.dtype]=None) -> torch.Tensor: