add basic gemma2 optimization (#11672)

python/llm/src/ipex_llm/transformers/convert.py

@@ -1508,9 +1508,10 @@ def _optimize_post(model, lightweight_bmm=False):
         modeling_module_name = model.__class__.__module__
         module = importlib.import_module(modeling_module_name)
         from ipex_llm.transformers.models.gemma import gemma_rms_norm_forward
-        convert_forward(model,
-                        module.GemmaRMSNorm,
-                        gemma_rms_norm_forward)
+        from ipex_llm.transformers.models.gemma2 import gemma2_attention_forward
+        from transformers.models.gemma2.modeling_gemma2 import Gemma2RMSNorm, Gemma2Attention
+        convert_forward(model, Gemma2RMSNorm, gemma_rms_norm_forward)
+        convert_forward(model, Gemma2Attention, gemma2_attention_forward)
     elif model.config.model_type == "Yi":
         modeling_module_name = model.__class__.__module__
         module = importlib.import_module(modeling_module_name)

python/llm/src/ipex_llm/transformers/models/gemma2.py

@@ -0,0 +1,116 @@
+#
+# Copyright 2016 The BigDL Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+# Some parts of this file is adapted from
+# https://github.com/huggingface/transformers/blob/main/src/transformers/models/gemma2/modeling_gemma2.py
+# coding=utf-8
+# Copyright 2024 Google Inc. HuggingFace Inc. team. All rights reserved.
+#
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Optional, Tuple
+
+import torch
+from ipex_llm.utils.common import invalidInputError
+from ipex_llm.transformers.models.utils import should_use_fuse_rope
+from transformers.cache_utils import Cache
+from transformers.models.gemma2.modeling_gemma2 import repeat_kv, apply_rotary_pos_emb
+
+
+def gemma2_attention_forward(
+    self,
+    hidden_states: torch.Tensor,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_value: Optional[Cache] = None,
+    output_attentions: bool = False,
+    use_cache: bool = False,
+    cache_position: Optional[torch.LongTensor] = None,
+) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+    bsz, q_len, _ = hidden_states.size()
+
+    query_states = self.q_proj(hidden_states)
+    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)
+    value_states = value_states.view(bsz, q_len, self.num_key_value_heads,
+                                     self.head_dim).transpose(1, 2)
+
+    # IPEX-LLM OPT: fuse rope
+    if should_use_fuse_rope(hidden_states, position_ids, self.training):
+        import xe_addons
+        xe_addons.rotary_half_inplaced(self.rotary_emb.inv_freq, position_ids,
+                                       query_states, key_states)
+        cos, sin = None, None
+    else:
+        cos, sin = self.rotary_emb(value_states, position_ids)
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+    if past_key_value is not None:
+        # sin and cos are specific to RoPE models; cache_position needed for the static cache
+        cache_kwargs = {
+            "sin": sin,
+            "cos": cos,
+            "sliding_window": self.sliding_window,
+            "cache_position": cache_position,
+        }
+        key_states, value_states = past_key_value.update(key_states, value_states,
+                                                         self.layer_idx, cache_kwargs)
+
+    key_states = repeat_kv(key_states, self.num_key_value_groups)
+    value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+    attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) * self.scaling
+
+    if self.config.attn_logit_softcapping is not None:
+        attn_weights = attn_weights / self.config.attn_logit_softcapping
+        attn_weights = torch.tanh(attn_weights)
+        attn_weights = attn_weights * self.config.attn_logit_softcapping
+
+    if attention_mask is not None:  # no matter the length, we just slice it
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    # upcast attention to fp32
+    attn_weights = torch.nn.functional.softmax(attn_weights,
+                                               dim=-1, dtype=torch.float32).to(query_states.dtype)
+    attn_weights = torch.nn.functional.dropout(attn_weights,
+                                               p=self.attention_dropout, training=self.training)
+    attn_output = torch.matmul(attn_weights, value_states)
+
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    attn_output = attn_output.view(bsz, q_len, -1)
+    attn_output = self.o_proj(attn_output)
+
+    if not output_attentions:
+        attn_weights = None
+
+    return attn_output, attn_weights, past_key_value