Refactor Cohere Model (#30027)

* changes

* addressing comments

* smol fix

Author: saurabhdash2512

src/transformers/models/cohere/configuration_cohere.py

@@ -85,6 +85,8 @@ class CohereConfig(PretrainedConfig):
             Whether to use a bias in the query, key, value and output projection layers during self-attention.
         attention_dropout (`float`, *optional*, defaults to 0.0):
             The dropout ratio for the attention probabilities.
+        use_qk_norm (`bool`, *optional*, defaults to `False`):
+            Whether to use query-key normalization in the attention
 
     ```python
     >>> from transformers import CohereModel, CohereConfig
@@ -123,6 +125,7 @@ def __init__(
         rope_theta=10000.0,
         attention_bias=False,
         attention_dropout=0.0,
+        use_qk_norm=False,
         **kwargs,
     ):
         self.vocab_size = vocab_size
@@ -145,6 +148,7 @@ def __init__(
         self.rope_theta = rope_theta
         self.attention_bias = attention_bias
         self.attention_dropout = attention_dropout
+        self.use_qk_norm = use_qk_norm
 
         super().__init__(
             pad_token_id=pad_token_id,

src/transformers/models/cohere/modeling_cohere.py

@@ -76,10 +76,10 @@ def _get_unpad_data(attention_mask):
 
 
 class CohereLayerNorm(nn.Module):
-    def __init__(self, hidden_size, eps=1e-5, bias=False):
+    def __init__(self, hidden_size=None, eps=1e-5, bias=False):
+        """The hidden size can be a tuple or an int. The tuple is used for QKNorm to normalize across head_dim"""
         super().__init__()
         self.weight = nn.Parameter(torch.ones(hidden_size))
-        self.bias = nn.Parameter(torch.zeros(hidden_size)) if bias else None
         self.variance_epsilon = eps
 
     def forward(self, hidden_states):
@@ -89,8 +89,6 @@ def forward(self, hidden_states):
         variance = (hidden_states - mean).pow(2).mean(-1, keepdim=True)
         hidden_states = (hidden_states - mean) * torch.rsqrt(variance + self.variance_epsilon)
         hidden_states = self.weight.to(torch.float32) * hidden_states
-        if self.bias is not None:
-            hidden_states = hidden_states + self.bias.to(torch.float32)
         return hidden_states.to(input_dtype)
 
 
@@ -122,7 +120,7 @@ def forward(self, x, position_ids):
             emb = torch.repeat_interleave(freqs, 2, dim=-1)
             cos = emb.cos()
             sin = emb.sin()
-        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+        return cos, sin
 
 
 def rotate_half(x):
@@ -133,7 +131,6 @@ def rotate_half(x):
     return rot_x
 
 
-# Copied from transformers.models.llama.modeling_llama.apply_rotary_pos_emb
 def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """Applies Rotary Position Embedding to the query and key tensors.
 
@@ -154,11 +151,14 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     Returns:
         `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
     """
+    dtype = q.dtype
+    q = q.float()
+    k = k.float()
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
-    return q_embed, k_embed
+    return q_embed.to(dtype=dtype), k_embed.to(dtype=dtype)
 
 
 # Copied from transformers.models.llama.modeling_llama.LlamaMLP Llama->Cohere
@@ -192,7 +192,6 @@ def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
     return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
 
 
-# Copied from transformers.models.llama.modeling_llama.LlamaAttention Llama->Cohere
 class CohereAttention(nn.Module):
     """Multi-headed attention from 'Attention Is All You Need' paper"""
 
@@ -216,13 +215,21 @@ def __init__(self, config: CohereConfig, layer_idx: Optional[int] = None):
         self.max_position_embeddings = config.max_position_embeddings
         self.rope_theta = config.rope_theta
         self.is_causal = True
+        self.use_qk_norm = config.use_qk_norm
 
         if (self.head_dim * self.num_heads) != self.hidden_size:
             raise ValueError(
                 f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
                 f" and `num_heads`: {self.num_heads})."
             )
 
+        if self.use_qk_norm:
+            # When sharding the model using Tensor Parallelism, need to be careful to use n_local_heads
+            self.q_norm = CohereLayerNorm(hidden_size=(self.num_heads, self.head_dim), eps=config.layer_norm_eps)
+            self.k_norm = CohereLayerNorm(
+                hidden_size=(self.num_key_value_heads, self.head_dim), eps=config.layer_norm_eps
+            )
+
         self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias)
         self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias)
         self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias)
@@ -255,8 +262,14 @@ def forward(
         key_states = self.k_proj(hidden_states)
         value_states = self.v_proj(hidden_states)
 
-        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
-        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim)
+        if self.use_qk_norm:
+            query_states = self.q_norm(query_states)
+            key_states = self.k_norm(key_states)
+
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
         value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
 
         past_key_value = getattr(self, "past_key_value", past_key_value)
@@ -335,11 +348,14 @@ def forward(
         key_states = self.k_proj(hidden_states)
         value_states = self.v_proj(hidden_states)
 
-        # Flash attention requires the input to have the shape
-        # batch_size x seq_length x head_dim x hidden_dim
-        # therefore we just need to keep the original shape
-        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
-        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim)
+        if self.use_qk_norm:
+            query_states = self.q_norm(query_states)
+            key_states = self.k_norm(key_states)
+
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
         value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
 
         cos, sin = self.rotary_emb(value_states, position_ids)
@@ -505,7 +521,7 @@ class CohereSdpaAttention(CohereAttention):
     SDPA API.
     """
 
-    # Adapted from CohereAttention.forward
+    # Ignore copy
     def forward(
         self,
         hidden_states: torch.Tensor,
@@ -538,8 +554,14 @@ def forward(
         key_states = self.k_proj(hidden_states)
         value_states = self.v_proj(hidden_states)
 
-        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
-        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim)
+        if self.use_qk_norm:
+            query_states = self.q_norm(query_states)
+            key_states = self.k_norm(key_states)
+
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
         value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
 
         cos, sin = self.rotary_emb(value_states, position_ids)
@@ -599,7 +621,7 @@ def __init__(self, config: CohereConfig, layer_idx: int):
         self.self_attn = COHERE_ATTENTION_CLASSES[config._attn_implementation](config=config, layer_idx=layer_idx)
 
         self.mlp = CohereMLP(config)
-        self.input_layernorm = CohereLayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.input_layernorm = CohereLayerNorm(hidden_size=(config.hidden_size), eps=config.layer_norm_eps)
 
     def forward(
         self,
@@ -822,7 +844,7 @@ def __init__(self, config: CohereConfig):
         self.layers = nn.ModuleList(
             [CohereDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
         )
-        self.norm = CohereLayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.norm = CohereLayerNorm(hidden_size=(config.hidden_size), eps=config.layer_norm_eps)
         self.gradient_checkpointing = False
 
         # Initialize weights and apply final processing