small refactor

launcher/src/main.rs

@@ -173,8 +173,9 @@ struct Args {
     /// for end users.
     #[clap(default_value = "20", long, env)]
     max_waiting_tokens: usize,
-    #[clap(default_value = "3000", long, short, env)]
+
     /// The port to listen on.
+    #[clap(default_value = "3000", long, short, env)]
     port: u16,
 
     /// The name of the socket for gRPC communication between the webserver

router/client/src/client.rs

@@ -113,7 +113,7 @@ impl Client {
             requests.push(Request {
                 id: 0,
                 // We truncate the input on the server side to be sure that it has the correct size
-                inputs: "test".to_string().repeat(max_input_length as usize),
+                inputs: "_test ".to_string().repeat(max_input_length as usize),
                 truncate: min(max_input_length, max_prefill_tokens - n_tokens),
                 // Set sampling parameters to also take these ops into account in the max memory
                 parameters: Some(NextTokenChooserParameters {

server/Makefile-vllm

@@ -5,7 +5,7 @@ vllm:
 	git clone https://github.com/OlivierDehaene/vllm.git
 
 build-vllm: vllm
-	cd vllm && git fetch && git checkout $(flash_att_commit)
+	cd vllm && git fetch && git checkout $(vllm_commit)
 	cd vllm && python setup.py build
 
 install-vllm: build-vllm

server/text_generation_server/cache.py

@@ -19,7 +19,6 @@ class Cache:
     def delete(self, batch_id: int):
         batch = self.pop(batch_id)
         if batch is not None:
-            batch.free()
             del batch
 
     def clear(self):

server/text_generation_server/models/flash_causal_lm.py

@@ -63,24 +63,48 @@ class CacheManager:
             0, num_blocks * self.block_size, dtype=torch.int32
         ).view(num_blocks, self.block_size)
 
-    def allocate(self, num_blocks: int) -> Tuple[torch.Tensor, torch.Tensor]:
+    def allocate(self, batch: "FlashCausalLMBatch"):
         # Get free blocks indices by finding values in mask that are not set to 0
         free_block_indices = self.free_block_mask.nonzero()
         assert (
-            len(free_block_indices) >= num_blocks
-        ), f"Out of available cache blocks: asked {num_blocks}, only {len(free_block_indices)} free blocks"
+            len(free_block_indices) >= batch.blocks
+        ), f"Out of available cache blocks: asked {batch.blocks}, only {len(free_block_indices)} free blocks"
+
+        # Slice by the number of required blocks
+        block_indices = free_block_indices[: batch.blocks]
+        block_indices = block_indices.flatten()
+
+        # Padded block tables
+        block_tables_tensor = torch.zeros(
+            (len(batch), batch.max_blocks), dtype=torch.int32
+        )
+
+        # Allocate paged attention blocks
+        cumulative_blocks = 0
+        slots = []
+        block_tables = []
+        for i, (needed_blocks, needed_slots) in enumerate(batch.needed_blocks_slots):
+            # Get allocated blocks for this sequence
+            allocated_blocks = block_indices[
+                cumulative_blocks : cumulative_blocks + needed_blocks
+            ]
+            # Get slots for the allocated blocks
+            allocated_slots = self.slots[allocated_blocks].flatten()[:needed_slots]
+
+            slots.append(allocated_slots)
+            block_tables.append(allocated_blocks.tolist())
+            block_tables_tensor[i, :needed_blocks] = allocated_blocks
+
+        batch.needed_blocks_slots = None
+        batch.block_tables = block_tables
+        batch.block_tables_tensor = block_tables_tensor.to(batch.input_ids.device)
+        batch.slots = torch.concat(slots).to(batch.input_ids.device)
 
         # Allocate the required number of blocks by setting the mask to 0
-        block_indices = free_block_indices[:num_blocks]
         self.free_block_mask[block_indices] = 0
 
-        # Get slots for the allocated blocks
-        slots = self.slots[block_indices].flatten()
-
-        return block_indices.flatten(), slots
-
     def free(self, block_indices: Optional[List[int]]):
-        if block_indices is not None:
+        if block_indices is not None and block_indices:
             # Reset mask
             self.free_block_mask[block_indices] = 1
 
@@ -448,12 +472,14 @@ class FlashCausalLMBatch(Batch):
             max_blocks = max(max_blocks, len(request_block_table))
 
         global CACHE_MANAGER
+        block_indices_to_free = []
         # Iterate on all requests
         for i, r in enumerate(self.requests):
             # Filter requests that are not part of the new batch
             if r.id not in requests_idx_mapping.keys():
+                block_indices_to_free.extend(self.block_tables[i])
         # Free blocks
-                CACHE_MANAGER.free(self.block_tables[i])
+        CACHE_MANAGER.free(block_indices_to_free)
 
         # Index into tensors
         input_ids = self.input_ids[indices]
@@ -643,7 +669,7 @@ class FlashCausalLMBatch(Batch):
             max_blocks=max_blocks,
         )
 
-    def free(self):
+    def __del__(self):
         if self.block_tables is not None:
             global CACHE_MANAGER
             # Free blocks
@@ -703,10 +729,10 @@ class FlashCausalLM(Model):
             logger.exception(
                 f"Not enough memory to handle {max_total_tokens} total tokens with {len(batch.input_ids)} "
                 f"prefill tokens. "
-                f"You need to decrease `--max-batch-total-tokens` and `--max-batch-prefill-tokens`"
+                f"You need to decrease `--max-batch-total-tokens` or `--max-batch-prefill-tokens`"
             )
             raise e
-        batch.free()
+        del batch
 
     def decode(self, generated_ids: Union[torch.Tensor, List[int]]) -> str:
         return self.tokenizer.decode(
@@ -749,33 +775,8 @@ class FlashCausalLM(Model):
         prefill_logprobs = batch.prefill_next_token_indices is not None
 
         if batch.needed_blocks_slots:
-            # Padded block tables
-            block_tables_tensor = torch.zeros(
-                (len(batch), batch.max_blocks), dtype=torch.int32
-            )
-
-            # Allocate paged attention blocks
-            slots = []
-            block_tables = []
-            try:
-                for i, (needed_blocks, needed_slots) in enumerate(
-                    batch.needed_blocks_slots
-                ):
-                    allocated_blocks, allocated_slots = CACHE_MANAGER.allocate(
-                        needed_blocks
-                    )
-                    slots.append(allocated_slots[:needed_slots])
-                    block_tables.append(allocated_blocks.tolist())
-                    block_tables_tensor[i, :needed_blocks] = allocated_blocks
-            except Exception as e:
-                for blocks in block_tables:
-                    CACHE_MANAGER.free(blocks)
-                raise e
-
-            batch.needed_blocks_slots = None
-            batch.block_tables = block_tables
-            batch.block_tables_tensor = block_tables_tensor.to(self.device)
-            batch.slots = torch.concat(slots).to(self.device)
+            # Allocate blocks to this batch
+            CACHE_MANAGER.allocate(batch)
 
         out = self.forward(
             batch.input_ids,
@@ -990,7 +991,7 @@ class FlashCausalLM(Model):
             batch.all_input_ids[i] = all_input_ids
 
         if stopped:
-            batch.free()
+            del batch
             # No need to return a batch if we know that all requests stopped
             return generations, None
 

server/text_generation_server/models/types.py

@@ -35,9 +35,6 @@ class Batch(ABC):
     def concatenate(cls, batches: List["Batch"]) -> "Batch":
         raise NotImplementedError
 
-    def free(self):
-        pass
-
     @abstractmethod
     def __len__(self):
         raise NotImplementedError

server/text_generation_server/server.py

@@ -65,12 +65,7 @@ class TextGenerationService(generate_pb2_grpc.TextGenerationServiceServicer):
             request.batch, self.model.tokenizer, self.model.dtype, self.model.device
         )
 
-        try:
         generations, next_batch = self.model.generate_token(batch)
-        except Exception as e:
-            batch.free()
-            raise e
-
         self.cache.set(next_batch)
 
         return generate_pb2.PrefillResponse(
@@ -93,20 +88,11 @@ class TextGenerationService(generate_pb2_grpc.TextGenerationServiceServicer):
             raise ValueError("All batches are empty")
 
         if len(batches) > 1:
-            try:
             batch = self.model.batch_type.concatenate(batches)
-            except Exception as e:
-                [batch.free() for batch in batches]
-                raise e
         else:
             batch = batches[0]
 
-        try:
         generations, next_batch = self.model.generate_token(batch)
-        except Exception as e:
-            batch.free()
-            raise e
-
         self.cache.set(next_batch)
 
         return generate_pb2.DecodeResponse(