xpu lora support

Signed-off-by: Wang, Yi A <[email protected]>

Author: sywangyi

server/text_generation_server/adapters/lora.py

@@ -11,7 +11,7 @@
 from peft import LoraConfig as _LoraConfig
 from torch.distributed import ProcessGroup
 from text_generation_server.utils.log import log_master
-
+from text_generation_server.utils.import_utils import SYSTEM
 from text_generation_server.adapters.config import AdapterConfig, ModuleMap
 
 from text_generation_server.adapters.weights import (
@@ -132,12 +132,21 @@ def __init__(
         self._use_cutlass_shrink = use_cutlass_shrink(self.lora_a_r)
         self._is_transposed = False
 
-        # [num_layers, hidden_size, r]
-        weights_a = [orient_for_rank(w, w.size(1)).contiguous() for w in weights_a]
-        self._weights_a = torch.stack(weights_a)
+        if SYSTEM == "ipex":
+            # [num_layers, r, hidden_size]
+            weights_a = [w.transpose(0, 1).contiguous() for w in weights_a]
+            self._weights_a = torch.stack(weights_a)
+
+            # [num_layers, hidden_size, r]
+            weights_b = [w.transpose(0, 1).contiguous() for w in weights_b]
+            self._weights_b = torch.stack(weights_b)
+        else:
+            # [num_layers, hidden_size, r]
+            weights_a = [orient_for_rank(w, w.size(1)).contiguous() for w in weights_a]
+            self._weights_a = torch.stack(weights_a)
 
-        # [num_layers, r, hidden_size]
-        self._weights_b = torch.stack(weights_b)
+            # [num_layers, r, hidden_size]
+            self._weights_b = torch.stack(weights_b)
 
         self.adapter_config = adapter_config
 
@@ -174,7 +183,10 @@ def _transpose_weights(self):
 
     @classmethod
     def get_batch_types(cls) -> List[Type[BatchAdapterWeights]]:
-        return [BatchLoraWeights]
+        if SYSTEM == "ipex":
+            return [IPEXBatchLoraWeights]
+        else:
+            return [BatchLoraWeights]
 
     # prepare pre-loaded lora weights for use in the model.
     #
@@ -243,14 +255,19 @@ def prepare_weights(
             lora_a_list[layer_id] = lora_a.transpose(0, 1)
             lora_b_list[layer_id] = lora_b.transpose(0, 1) * scale
 
-        # pad lora ranks to be compatible with sgmv
-        lora_a_list = [pad_rank(w, dim=1, world_size=world_size) for w in lora_a_list]
-        lora_b_list = [pad_rank(w, dim=0, world_size=world_size) for w in lora_b_list]
+        if SYSTEM != "ipex":
+            # pad lora ranks to be compatible with sgmv
+            lora_a_list = [
+                pad_rank(w, dim=1, world_size=world_size) for w in lora_a_list
+            ]
+            lora_b_list = [
+                pad_rank(w, dim=0, world_size=world_size) for w in lora_b_list
+            ]
 
-        if lora_a_list:
-            # update rank if it was padded
-            padded_rank = lora_a_list[0].size(1)
-            config.r = padded_rank
+            if lora_a_list:
+                # update rank if it was padded
+                padded_rank = lora_a_list[0].size(1)
+                config.r = padded_rank
 
         return LoraWeights(
             *shard_lora_weights(
@@ -466,6 +483,115 @@ def load(
         )
 
 
+@dataclass
+class IPEXBatchLoraWeights(BatchLoraWeights):
+    @classmethod
+    def load(
+        self,
+        adapter_weights: Dict[int, AdapterWeights],
+        meta: AdapterBatchMetadata,
+        prefill: bool,
+        prefill_head_indices: Optional[torch.Tensor],
+    ) -> Optional["BatchLoraWeights"]:
+        adapter_weights = {k: _convert_lora(v) for k, v in adapter_weights.items()}
+        adapter_weights = {
+            k: v for k, v in adapter_weights.items() if isinstance(v, LoraWeights)
+        }
+        if not adapter_weights:
+            return None
+
+        first_weights = next(iter(adapter_weights.values()))
+        device = first_weights.weights_a.device
+        segment_indices = meta.segment_indices
+
+        lora_a = {
+            idx: adapter_weights[idx].weights_a
+            for idx in segment_indices
+            if idx in adapter_weights
+        }
+        lora_b = {
+            idx: adapter_weights[idx].weights_b
+            for idx in segment_indices
+            if idx in adapter_weights
+        }
+        adapter_index_configs = {
+            idx: adapter_weights[idx].adapter_config
+            for idx in segment_indices
+            if idx in adapter_weights
+        }
+        if len(lora_a) != 0:
+            lora_a_ptr = torch.stack(list(lora_a.values()))
+        if len(lora_b) != 0:
+            lora_b_ptr = torch.stack(list(lora_b.values()))
+
+        use_sgmv = True if prefill else False
+
+        adapter_to_segment = {v: k for k, v in enumerate(segment_indices)}
+
+        rank_indices = defaultdict(list)
+        for segment_idx, adapter_idx in enumerate(segment_indices):
+            if adapter_idx not in adapter_weights:
+                continue
+            rank_indices[adapter_weights[adapter_idx].lora_a_r].append(segment_idx)
+
+        if prefill_head_indices is not None:
+            j, prefill_head_segment_starts, prefill_head_segment_ends = 1, [0], [0]
+            for head_index in prefill_head_indices:
+                # j cannot go out of bounds as that would mean there are tokens without corresponding adapters
+                if head_index < meta.adapter_segments[j]:
+                    prefill_head_segment_ends[-1] += 1
+                else:
+                    prefill_head_segment_starts.append(prefill_head_segment_ends[-1])
+                    prefill_head_segment_ends.append(prefill_head_segment_ends[-1] + 1)
+                    j += 1
+
+        rank_data = {}
+        segment_starts = None
+        segment_ends = None
+        if use_sgmv:
+            segment_starts = meta.adapter_segments[:-1]
+            segment_ends = meta.adapter_segments[1:]
+            if prefill_head_indices is not None:
+                segment_starts = prefill_head_segment_starts[:-1]
+                segment_ends = prefill_head_segment_ends[1:]
+        batch_indices = [
+            adapter_to_segment[idx] for idx in meta.adapter_indices.tolist()
+        ]
+        for rank, indices in rank_indices.items():
+            adapters_indices = []
+            lora_a_keys = list(lora_a.keys())
+            for segment_idx in batch_indices:
+                if segment_idx in indices:
+                    adapters_indices.append(
+                        lora_a_keys.index(segment_indices[segment_idx])
+                    )
+                else:
+                    adapters_indices.append(-1)
+            adapters_indices = torch.tensor(
+                adapters_indices, dtype=torch.int64, device=device
+            )
+            if use_sgmv:
+                adapters_indices = adapters_indices[segment_starts]
+            rank_data[rank] = RankSegments(
+                rank=rank,
+                tmp_shrink=None,
+                tmp_expand=None,
+                lora_a_ptr=lora_a_ptr,
+                lora_b_ptr=lora_b_ptr,
+                segment_starts=segment_starts,
+                segment_ends=segment_ends,
+                indices=adapters_indices,
+            )
+
+        return BatchLoraWeights(
+            lora_a=lora_a,
+            lora_b=lora_b,
+            adapter_index_configs=adapter_index_configs,
+            rank_data=rank_data,
+            use_sgmv=use_sgmv,
+        )
+
+
 def get_scaling_factor(
     lora_alpha: int,
     r: int,

server/text_generation_server/layers/lora.py

@@ -4,6 +4,7 @@
 import torch.distributed
 from torch import nn
 from torch.distributed import ProcessGroup
+from text_generation_server.utils.import_utils import SYSTEM
 
 from text_generation_server.utils.sgmv import (
     add_lora_a_bgmv,
@@ -115,6 +116,91 @@ def forward_layer_type(
                         self.layer_id,
                     )
 
+            if end_idx - start_idx != result.shape[1]:
+                result[:, start_idx:end_idx] += proj
+        elif SYSTEM == "ipex" and data is not None:
+            from intel_extension_for_pytorch.llm.functional import (
+                bgmv_expand,
+                bgmv_shrink,
+                sgmv_expand,
+                sgmv_shrink,
+            )
+
+            # In IPEX, we provide the same API for sgmv and bgmv
+            if end_idx - start_idx != result.shape[1]:
+                proj = torch.zeros_like(result[:, start_idx:end_idx])
+            else:
+                proj = result
+
+            for r, rank_segments in data.rank_data.items():
+                lora_a_ptr = rank_segments.lora_a_ptr[:, self.layer_id, :].contiguous()
+                lora_b_ptr = rank_segments.lora_b_ptr[:, self.layer_id, :].contiguous()
+
+                if lora_a_ptr is None or lora_b_ptr is None:
+                    raise ValueError("LoRA data is missing")
+
+                if data.use_sgmv:
+                    # Use SGMV for prefill
+                    seq_len_tensor = (
+                        rank_segments.segment_ends - rank_segments.segment_starts
+                    ).to(torch.int64)
+                    b_seq_start_loc = rank_segments.segment_starts.to(torch.int64)
+                    total_tokens = seq_len_tensor.sum()
+                    v = torch.zeros(
+                        (total_tokens, r), dtype=input.dtype, device=input.device
+                    )
+                    bs = seq_len_tensor.shape[0]
+                    sgmv_shrink(
+                        input,
+                        lora_a_ptr,
+                        v,
+                        b_seq_start_loc,
+                        seq_len_tensor,
+                        rank_segments.indices,
+                        bs,
+                        seq_len_tensor.max().item(),
+                        1.0,
+                    )
+
+                    if self.process_group.size() > 1:
+                        v = self.collect_lora_a(v)
+
+                    sgmv_expand(
+                        v,
+                        lora_b_ptr,
+                        proj,
+                        b_seq_start_loc,
+                        seq_len_tensor,
+                        rank_segments.indices,
+                        bs,
+                        seq_len_tensor.max().item(),
+                        add_inputs=True,
+                    )
+                else:
+                    # Use BGMV for decode
+                    v = torch.zeros(
+                        (input.size(0), r), dtype=input.dtype, device=input.device
+                    )
+                    # TODO: error with [-1, 0], but not [0, -1]
+                    bgmv_shrink(
+                        input,
+                        lora_a_ptr,
+                        v,
+                        rank_segments.indices,
+                        1.0,
+                    )
+
+                    if self.process_group.size() > 1:
+                        v = self.collect_lora_a(v)
+
+                    bgmv_expand(
+                        v,
+                        lora_b_ptr,
+                        proj,
+                        rank_segments.indices,
+                        add_inputs=True,
+                    )
+
             if end_idx - start_idx != result.shape[1]:
                 result[:, start_idx:end_idx] += proj
         else: