add slots filtering kernel

server/text_generation_server/models/flash_causal_lm.py

@@ -77,6 +77,7 @@ from text_generation_server.models.metadata_kernels import (
     block_tables_to_ragged,
     block_tables_to_padded,
     prepare_position_slot_ids,
+    slots_filtering,
 )
 
 tracer = trace.get_tracer(__name__)
@@ -500,10 +501,11 @@ class FlashCausalLMBatch(Batch):
         # Used to index into tensors
         indices = []
 
-        # slots to keep after filtering
-        slot_filtering_indices = torch.zeros(
-            self.slots.shape[0], dtype=torch.bool, device=device
-        )
+        if not has_triton():
+            # slots to keep after filtering
+            slot_filtering_indices = torch.zeros(
+                self.slots.shape[0], dtype=torch.bool, device=device
+            )
 
         # Create on CPU to only move to GPU once instead of at every copy
         slot_indices = torch.empty(len(request_ids), dtype=torch.int64)
@@ -531,6 +533,7 @@ class FlashCausalLMBatch(Batch):
 
         num_blocks = 0
         max_blocks = 0
+        max_slots = 0
         cumulative_slot_tokens = 0
 
         for i, request_id in enumerate(request_ids):
@@ -578,8 +581,9 @@ class FlashCausalLMBatch(Batch):
             end_slot = self.cu_slots[idx + 1]
             slot_length = end_slot - start_slot
 
-            # Set slice
-            slot_filtering_indices[start_slot:end_slot] = True
+            if not has_triton():
+                # Set slice
+                slot_filtering_indices[start_slot:end_slot] = True
 
             cu_slots.append(cumulative_slot_tokens + slot_length)
 
@@ -593,6 +597,7 @@ class FlashCausalLMBatch(Batch):
 
             cumulative_slot_tokens += slot_length
             max_blocks = max(max_blocks, len(request_block_table))
+            max_slots = max(max_slots, slot_length)
 
         all_input_ids_tensor = self.all_input_ids_tensor[indices]
         block_tables_tensor = self.block_tables_tensor[indices]
@@ -603,9 +608,18 @@ class FlashCausalLMBatch(Batch):
         )
         prompt_lengths_tensor = self.prompt_lengths_tensor[indices]
 
-        slots = self.slots[slot_filtering_indices]
         cu_slots = torch.tensor(cu_slots, dtype=torch.int64)
 
+        if not has_triton():
+            slots = self.slots[slot_filtering_indices]
+        else:
+            slots = self.slots.new_empty(cumulative_slot_tokens)
+            gpu_cu_slots = cu_slots.to(device)
+            slots_indexing_start = self.cu_slots.to(device)[indices]
+            slots_filtering(
+                max_slots, self.slots, slots, gpu_cu_slots, slots_indexing_start
+            )
+
         if self.prefilling:
             # These values will be set by `FlashCausalLMBatch.prepare_for_prefill`
             position_ids = None

server/text_generation_server/models/metadata_kernels.py

@@ -3,20 +3,28 @@ import triton
 
 import triton.language as tl
 
-from typing import List
+from loguru import logger
+from typing import List, Optional
 from torch.utils._triton import has_triton as has_triton_torch
 
 from text_generation_server.utils.import_utils import (
     SYSTEM,
 )
+from text_generation_server.utils.log import log_master
+
+_HAS_TRITON: Optional[bool] = None
 
 
 def has_triton():
-    # FIXME: it seems that has_triton_torch is bugged on RocM
-    #        For now, only accept cuda
-    if SYSTEM == "cuda":
-        return has_triton_torch()
-    return False
+    global _HAS_TRITON
+    if _HAS_TRITON is None:
+        # FIXME: it seems that has_triton_torch is bugged on RocM
+        #        For now, only accept cuda
+        _HAS_TRITON = has_triton_torch() if SYSTEM == "cuda" else False
+        if _HAS_TRITON:
+            log_master(logger.info, "Using optimized Triton indexing kernels.")
+
+    return _HAS_TRITON
 
 
 def block_tables_to_padded(
@@ -133,6 +141,53 @@ def prepare_position_slot_ids(
     )
 
 
+def slots_filtering(
+    max_slots: int,
+    slots: torch.Tensor,
+    filtered_slots: torch.Tensor,
+    cu_slots: torch.Tensor,
+    slots_start: torch.Tensor,
+):
+    def grid(meta):
+        return (
+            triton.cdiv(max_slots, meta["BLOCK_SIZE"]),
+            len(slots_start),
+        )
+
+    triton_slots_filtering[grid](
+        slots, filtered_slots, slots_start, cu_slots, BLOCK_SIZE=256
+    )
+
+
+@triton.jit
+def triton_slots_filtering(
+    # Inputs
+    slots_ptr,
+    filtered_slots_ptr,
+    slots_start_ptr,
+    cu_slots_ptr,
+    # Const values
+    BLOCK_SIZE: "tl.constexpr",
+):
+    # Position in block_tables_ragged.numel() / BLOCK_SIZE
+    pid = tl.program_id(axis=0)
+    # Position in batch
+    bid = tl.program_id(axis=1)
+
+    block_start = pid * BLOCK_SIZE
+    block_arange = block_start + tl.arange(0, BLOCK_SIZE)
+
+    filter_start = tl.load(slots_start_ptr + bid)
+
+    slot_start = tl.load(cu_slots_ptr + bid)
+    slot_end = tl.load(cu_slots_ptr + bid + 1)
+
+    mask = (slot_start + block_arange) < slot_end
+
+    slots = tl.load(slots_ptr + filter_start + block_arange, mask=mask)
+    tl.store(filtered_slots_ptr + slot_start + block_arange, slots, mask=mask)
+
+
 @triton.jit
 def triton_block_tables_to_padded(
     # Inputs