add phi3 optimization (#10871)

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

@@ -653,6 +653,9 @@ def _optimize_pre(model):
     if model.config.model_type == "phi":
         from ipex_llm.transformers.models.phi import merge_qkv
         model.apply(merge_qkv)
+    if model.config.model_type == "phi3":
+        from ipex_llm.transformers.models.phi3 import split_mlp
+        model.apply(split_mlp)
     if model.config.model_type == "qwen":
         rope_base = model.config.rotary_emb_base
         from accelerate.big_modeling import init_empty_weights
@@ -1426,6 +1429,17 @@ def _optimize_post(model, lightweight_bmm=False):
         from ipex_llm.transformers.models.phi import model_forward
         convert_forward(model, module.PhiAttention, attention_forward)
         convert_forward(model, module.PhiModel, model_forward)
+    elif model.config.model_type == "phi3":
+        # for phi-3
+        modeling_module_name = model.__class__.__module__
+        module = importlib.import_module(modeling_module_name)
+        from ipex_llm.transformers.models.phi3 import attention_forward
+        convert_forward(model, module.Phi3Attention, attention_forward)
+        from ipex_llm.transformers.models.phi3 import mlp_forward
+        convert_forward(model, module.Phi3MLP, mlp_forward)
+        from ipex_llm.transformers.models.phi3 import model_forward_wrapper
+        model_forward = model_forward_wrapper(module.Phi3Model.forward)
+        convert_forward(model, module.Phi3Model, model_forward)
     elif model.config.model_type == 'yuan':
         modeling_module_name = model.__class__.__module__
         module = importlib.import_module(modeling_module_name)

python/llm/src/ipex_llm/transformers/models/phi3.py

@@ -0,0 +1,193 @@
+#
+# 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://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/modeling_phi3.py
+# which is licensed under Apache License 2.0:
+#
+# Copyright 2024 Microsoft and the 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.
+
+import math
+import torch
+import warnings
+
+from ipex_llm.transformers.models.utils import (
+    rotate_half, should_use_fuse_rope,
+    apply_rotary_pos_emb_cache_freq_xpu
+)
+from ipex_llm.transformers.models.utils import mlp_fusion_check, SILU
+from ipex_llm.transformers.kv import DynamicNormalCache
+
+from typing import Optional, Tuple, List
+from transformers.models.phi.modeling_phi import repeat_kv
+from transformers.cache_utils import Cache
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+def 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,
+) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+    warnings.warn("You are not running the flash-attention implementation, "
+                  "expect numerical differences.")
+
+    bsz, q_len, _ = hidden_states.size()
+
+    qkv = self.qkv_proj(hidden_states)
+    qkv = qkv.view(bsz, q_len, self.num_heads + 2 * self.num_key_value_heads, self.head_dim)
+    qkv = qkv.transpose(1, 2)
+    query_states, key_states, value_states = qkv.split([self.num_heads,
+                                                        self.num_key_value_heads,
+                                                        self.num_key_value_heads], dim=1)
+
+    kv_seq_len = key_states.shape[-2]
+    if past_key_value is not None:
+        kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
+
+    cos, sin = self.rotary_emb(value_states, position_ids, seq_len=kv_seq_len)
+    # IPEX-LLM OPT: fuse rope
+    if should_use_fuse_rope(hidden_states, position_ids, self.training):
+        query_states, key_states = apply_rotary_pos_emb_cache_freq_xpu(query_states, key_states,
+                                                                       sin, cos, "phi3")
+    else:
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states,
+                                                        cos, sin, position_ids)
+
+    if past_key_value is not None:
+        key_states, value_states = past_key_value.update(key_states, value_states,
+                                                         self.layer_idx, None)
+
+    # repeat k/v heads if n_kv_heads < n_heads
+    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)) / math.sqrt(self.head_dim)
+
+    if attention_mask is not None:
+        attn_weights = attn_weights + attention_mask
+
+    # upcast attention to fp32
+    attn_weights = torch.nn.functional.softmax(attn_weights, dim=-1,
+                                               dtype=torch.float32).to(value_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.reshape(bsz, q_len, self.hidden_size)
+
+    attn_output = self.o_proj(attn_output)
+
+    if not output_attentions:
+        attn_weights = None
+
+    return attn_output, attn_weights, past_key_value
+
+
+def split_mlp(module: torch.nn.Module):
+    if module.__class__.__name__ == "Phi3MLP":
+        gate_weight, up_weight = module.gate_up_proj.weight.data.chunk(2, dim=0)
+
+        gate_proj = torch.nn.Linear(0, 0, bias=False)
+        gate_proj.weight = torch.nn.Parameter(gate_weight, requires_grad=False)
+        gate_proj.in_features = gate_weight.size(1)
+        gate_proj.out_features = gate_weight.size(0)
+
+        up_proj = torch.nn.Linear(0, 0, bias=False)
+        up_proj.weight = torch.nn.Parameter(up_weight, requires_grad=False)
+        up_proj.in_features = up_weight.size(1)
+        up_proj.out_features = up_weight.size(0)
+
+        module.gate_proj = gate_proj
+        module.up_proj = up_proj
+
+        del module.gate_up_proj
+
+
+def mlp_forward(
+    self,
+    hidden_states: torch.FloatTensor
+) -> torch.FloatTensor:
+    x_2d = hidden_states.view(-1, hidden_states.shape[-1])
+    qtype = getattr(self.gate_proj, "qtype", None)
+    if mlp_fusion_check(x_2d, qtype, self.training):
+        x_2d = x_2d.contiguous()
+        import linear_q4_0
+        return self.down_proj(linear_q4_0.mlp_forward_xpu(
+            x_2d, self.gate_proj.weight.data, self.up_proj.weight.data,
+            x_2d.shape[0], x_2d.shape[1], self.gate_proj.out_features,
+            SILU, qtype
+        ))
+    return self.down_proj(
+        self.activation_fn(self.gate_proj(hidden_states)) * self.up_proj(hidden_states)
+    )
+
+
+def model_forward_wrapper(origin_model_forward):
+    def model_forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ):
+        # IPEX-LLM OPT: kv cache but no sdp (its head_dim 96, cannot use sdp)
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        if use_cache:
+            if not isinstance(past_key_values, DynamicNormalCache):
+                past_key_values = DynamicNormalCache.from_legacy_cache(past_key_values)
+        return origin_model_forward(
+            self=self,
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+    return model_forward

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

@@ -167,6 +167,14 @@ def rotate_every_two(x):
     return x.flatten(-2)  # in einsum notation: rearrange(x, '... d j -> ... (d j)')
 
 
+def should_use_fuse_rope(hidden_states, position_ids, training):
+    return (
+        hidden_states.device.type == "xpu"
+        and not training and not hidden_states.requires_grad
+        and position_ids is not None
+    )
+
+
 def apply_rotary_pos_emb(q, k, cos, sin, position_ids, model_family):
     if model_family in ["llama", "baichuan", "internlm", "aquila", "gpt_neox", "mistral",
                         "mixtral", "qwen2", "yuan", "stablelm", "qwen2_moe"]:
@@ -234,7 +242,7 @@ def apply_rotary_pos_emb_cache_freq_xpu(q, k, sin, cos, model_family, position_i
         cos = cos[position_ids].unsqueeze(1)  # [bs, 1, seq_len, dim]
         sin = sin[position_ids].unsqueeze(1)  # [bs, 1, seq_len, dim]
         linear_q4_0.apply_rotary_embedding_half_q_and_k_cache_freq(q, k, sin, cos, q_embed, k_embed)
-    elif model_family in ["gemma"]:
+    elif model_family in ["gemma", "phi3"]:
         cos = cos.unsqueeze(1)
         sin = sin.unsqueeze(1)
         linear_q4_0.apply_rotary_embedding_half_q_and_k_cache_freq(q, k, sin, cos, q_embed, k_embed)