Refactor and annotate types for attention

cacheflow/models/attention.py

@@ -1,4 +1,4 @@
-from typing import Optional
+from typing import List, Optional
 
 import torch
 import torch.nn as nn
@@ -30,24 +30,34 @@ class OPTCacheFlowAttention(nn.Module):
 
     def multi_query_kv_attention(
         self,
-        output: torch.Tensor,
+        output: torch.Tensor,       # [num_prompt_tokens, num_heads, head_size]
-        query: torch.Tensor,
+        query: torch.Tensor,        # [num_prompt_tokens, num_heads, head_size]
-        key: torch.Tensor,
+        key: torch.Tensor,          # [num_prompt_tokens, num_heads, head_size]
-        value: torch.Tensor,
+        value: torch.Tensor,        # [num_prompt_tokens, num_heads, head_size]
+        prompt_lens: List[int],
     ) -> None:
-        # FIXME(woosuk): Replace this with a custom op call.
+        # FIXME(woosuk): Replace the following with a custom op.
-        attention_mask = torch.triu(
+        start_idx = 0
-            torch.ones(query.shape[0], key.shape[0]), diagonal=1) * -1e5
+        for prompt_len in prompt_lens:
-        attention_mask = attention_mask.to(dtype=query.dtype, device=query.device)
+            out = output[start_idx:start_idx + prompt_len]
-        out = self._masked_attention(query, key, value, attention_mask)
+            q = query[start_idx:start_idx + prompt_len]
-        output.copy_(out, non_blocking=True)
+            k = key[start_idx:start_idx + prompt_len]
+            v = value[start_idx:start_idx + prompt_len]
+
+            attention_mask = torch.triu(
+                torch.ones(q.shape[0], k.shape[0]), diagonal=1) * -1e5
+            attention_mask = attention_mask.to(dtype=q.dtype, device=q.device)
+            attention_out = self._masked_attention(q, k, v, attention_mask)
+            out.copy_(attention_out, non_blocking=True)
+
+            start_idx += prompt_len
 
     def single_query_cached_kv_attention(
         self,
-        output: torch.Tensor,
+        output: torch.Tensor,           # [num_generation_tokens, num_heads, head_size]
-        query: torch.Tensor,
+        query: torch.Tensor,            # [num_generation_tokens, num_heads, head_size]
-        key_cache: torch.Tensor,
+        key_cache: torch.Tensor,        # [num_blocks, num_heads, head_size/x, block_size, x]
-        value_cache: torch.Tensor,
+        value_cache: torch.Tensor,      # [num_blocks, num_heads, block_size, head_size]
         input_metadata: InputMetadata,
     ) -> None:
         num_heads = value_cache.shape[1]
@@ -82,15 +92,18 @@ class OPTCacheFlowAttention(nn.Module):
 
     def forward(
         self,
-        query: torch.Tensor,
+        query: torch.Tensor,                    # [num_tokens, num_heads * head_size]
-        key: torch.Tensor,
+        key: torch.Tensor,                      # [num_tokens, num_heads * head_size]
-        value: torch.Tensor,
+        value: torch.Tensor,                    # [num_tokens, num_heads * head_size]
-        key_cache: torch.Tensor,
+        key_cache: torch.Tensor,                # [num_blocks, num_heads, head_size/x, block_size, x]
-        value_cache: torch.Tensor,
+        value_cache: torch.Tensor,              # [num_blocks, num_heads, block_size, head_size]
         input_metadata: InputMetadata,
         cache_event: Optional[torch.cuda.Event],
-    ) -> torch.Tensor:
+    ) -> torch.Tensor:                          # [num_tokens, num_heads * head_size]
-        # Prune out invalid tokens.
+        # Pre-allocate the output tensor.
+        output = torch.empty_like(query)
+
+        # Prune out paddings if any.
         query = query[:input_metadata.num_valid_tokens]
         key = key[:input_metadata.num_valid_tokens]
         value = value[:input_metadata.num_valid_tokens]
@@ -101,18 +114,11 @@ class OPTCacheFlowAttention(nn.Module):
         query = query.view(-1, num_heads, head_size)
         key = key.view(-1, num_heads, head_size)
         value = value.view(-1, num_heads, head_size)
+        output = output.view(-1, num_heads, head_size)
 
         # Compute the attention op for prompts.
-        output = torch.empty_like(query)
+        self.multi_query_kv_attention(
-        start_idx = 0
+            output, query, key, value, input_metadata.prompt_lens)
-        for i in range(input_metadata.num_prompts):
-            prompt_len = input_metadata.prompt_lens[i]
-            out = output[start_idx:start_idx + prompt_len]
-            q = query[start_idx:start_idx + prompt_len]
-            k = key[start_idx:start_idx + prompt_len]
-            v = value[start_idx:start_idx + prompt_len]
-            self.multi_query_kv_attention(out, q, k, v)
-            start_idx += prompt_len
 
         # Wait until the cache op is done.
         if cache_event is not None:
@@ -124,6 +130,7 @@ class OPTCacheFlowAttention(nn.Module):
 
         if input_metadata.num_generation_tokens > 0:
             # Compute the attention op for generation tokens.
+            start_idx = sum(input_metadata.prompt_lens)
             self.single_query_cached_kv_attention(
                 output[start_idx:],
                 query[start_idx:],
@@ -132,4 +139,5 @@ class OPTCacheFlowAttention(nn.Module):
                 input_metadata)
 
         # Reshape the output tensor.
+        # NOTE(woosuk): The output tensor may include paddings.
         return output.view(-1, num_heads * head_size)