Optimize stablelm on NPU (#11512)

* stablelm_optimize

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

@@ -118,3 +118,18 @@ def optimize_llm(model: torch.nn.Module):
         convert_forward(model, module.MiniCPMForCausalLM, minicpm_model_causal_lm_forward)
         convert_forward(model, module.MiniCPMAttention, minicpm_attention_forward)
         convert_forward(model, module.MiniCPMMLP, minicpm_mlp_forward)
+    elif model.config.model_type == "stablelm":
+        from ipex_llm.transformers.npu_models.stablelm import merge_qkv
+        from ipex_llm.transformers.npu_models.stablelm import merge_mlp
+        model.apply(merge_qkv)
+        model.apply(merge_mlp)
+
+        from ipex_llm.transformers.npu_models.stablelm import stablelm_model_forward
+        from ipex_llm.transformers.npu_models.stablelm import stablelm_attention_forward
+        from ipex_llm.transformers.npu_models.stablelm import stablelm_mlp_forward
+        from transformers.models.stablelm.modeling_stablelm import StableLmModel
+        from transformers.models.stablelm.modeling_stablelm import StableLmAttention
+        from transformers.models.stablelm.modeling_stablelm import StableLmMLP
+        convert_forward(model, StableLmModel, stablelm_model_forward)
+        convert_forward(model, StableLmAttention, stablelm_attention_forward)
+        convert_forward(model, StableLmMLP, stablelm_mlp_forward)

python/llm/src/ipex_llm/transformers/npu_models/stablelm.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://github.com/huggingface/transformers/blob/v4.40.0/src/transformers/models/llama/modeling_llama.py
+# which is licensed under Apache License 2.0:
+#
+# Copyright 2021 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.
+
+
+from typing import Optional, Tuple, List
+
+import math
+import torch
+from transformers.cache_utils import Cache
+from transformers.models.stablelm.modeling_stablelm import repeat_kv, apply_rotary_pos_emb
+from transformers.models.stablelm.modeling_stablelm import StableLmAttention, StableLmMLP, \
+    StableLmModel
+
+from ipex_llm.transformers.npu_models.common import merge_linear
+
+
+def merge_qkv(module: torch.nn.Module):
+    if isinstance(module, StableLmAttention):
+        new_weight = torch.cat([
+            module.q_proj.weight.data,
+            module.k_proj.weight.data,
+            module.v_proj.weight.data,
+        ], dim=0)
+
+        if module.q_proj.bias is not None:
+            qkv_proj = torch.nn.Linear(0, 0, bias=True)
+            new_bias = torch.cat([
+                module.q_proj.bias.data,
+                module.k_proj.bias.data,
+                module.v_proj.bias.data,
+            ], dim=0)
+            qkv_proj.bias = torch.nn.Parameter(new_bias, requires_grad=False)
+        else:
+            qkv_proj = torch.nn.Linear(0, 0, bias=False)
+        qkv_proj.weight = torch.nn.Parameter(new_weight, requires_grad=False)
+        qkv_proj.in_features = new_weight.size(1)
+        qkv_proj.out_features = new_weight.size(0)
+        module.qkv_proj = qkv_proj
+
+        del module.q_proj, module.k_proj, module.v_proj
+
+
+def merge_mlp(module: torch.nn.Module):
+    if isinstance(module, StableLmMLP):
+        gate_up_proj = merge_linear([
+            module.gate_proj,
+            module.up_proj,
+        ])
+        module.gate_up_proj = gate_up_proj
+        del module.gate_proj, module.up_proj
+
+
+def stablelm_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 changes start
+    from ipex_llm.transformers.kv import DynamicNormalCache
+    # IPEX-LLM OPT: kv cache and quantize kv cache
+    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 StableLmModel.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,
+    )
+
+
+def stablelm_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]]]:
+    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)
+    # For stablelm-2-12b's qk per-head norm
+    if getattr(self, "qk_layernorm", False):
+        query_states = self.q_layernorm(query_states)
+        key_states = self.k_layernorm(key_states)
+
+    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)
+
+    # Partial rotary embedding
+    # [batch_size, num_heads, seq_length, head_dim * config.partial_rotary_factor]
+    rot_dim = self.rotary_emb.dim
+
+    query_rot, query_pass = query_states[..., :rot_dim], query_states[..., rot_dim:]
+    key_rot, key_pass = key_states[..., :rot_dim], key_states[..., rot_dim:]
+    cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+    query_rot, key_rot = apply_rotary_pos_emb(query_rot,
+                                              key_rot,
+                                              cos,
+                                              sin,
+                                              position_ids)
+    query_states = torch.cat((query_rot, query_pass), dim=-1)
+    key_states = torch.cat((key_rot, key_pass), dim=-1)
+
+    if past_key_value is not None:
+        # Specific to RoPE models with partial rotation
+        cache_kwargs = {"sin": sin, "cos": cos, "partial_rotation_size": self.rotary_emb.dim}
+        key_states, value_states = past_key_value.update(key_states, value_states,
+                                                         self.layer_idx, cache_kwargs)
+
+    # 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 = self.attention_dropout(attn_weights)
+    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 stablelm_mlp_forward(self, x):
+    gate_up_proj = self.gate_up_proj(x)
+    gate_proj, up_proj = gate_up_proj.chunk(2, dim=-1)
+    down_proj = self.down_proj(self.act_fn(gate_proj) * up_proj)
+    return down_proj