Add Qwen pipeline and example (#12292)

* support qwen pipeline

* update error msg

* style

* meet review

* minor

python/llm/example/NPU/HF-Transformers-AutoModels/LLM/Pipeline-Models/README.md

@@ -8,6 +8,7 @@ In this directory, you will find examples on how to directly run HuggingFace `tr
 |------------|----------------------------------------------------------------|
 | Llama2 | [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf) |
 | Llama3 | [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) |
+| Qwen2.5 | [Qwen/Qwen2.5-7b-Instruct](https://huggingface.co/Qwen/Qwen2.5-7b-Instruct) |
 | Baichuan2 | [baichuan-inc/Baichuan2-7B-Chat](https://huggingface.co/baichuan-inc/Baichuan-7B-Chat) |
 | MiniCPM | [openbmb/MiniCPM-1B-sft-bf16](https://huggingface.co/openbmb/MiniCPM-1B-sft-bf16) |
 
@@ -30,7 +31,7 @@ pip install --pre --upgrade ipex-llm[npu]
 
 ## 2. Runtime Configurations
 
-**Following envrionment variables are required**:
+**Following environment variables are required**:
 
 ```cmd
 set BIGDL_USE_NPU=1
@@ -46,6 +47,9 @@ python llama2.py
 :: to run Meta-Llama-3-8B-Instruct
 python llama3.py
 
+:: to run Qwen2.5-7b-Instruct
+python qwen.py
+
 :: to run Baichuan2-7B-Chat
 python baichuan2.py
 

python/llm/example/NPU/HF-Transformers-AutoModels/LLM/Pipeline-Models/qwen.py

@@ -0,0 +1,108 @@
+#
+# 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.
+#
+
+
+import os
+import torch
+import time
+import argparse
+from ipex_llm.transformers.npu_model import AutoModelForCausalLM
+from transformers import AutoTokenizer
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Predict Tokens using `generate()` API for npu model"
+    )
+    parser.add_argument(
+        "--repo-id-or-model-path",
+        type=str,
+        default="Qwen/Qwen2.5-7B-Instruct",  # Or Qwen2-7B-Instruct
+        help="The huggingface repo id for the Baichuan2 model to be downloaded"
+        ", or the path to the huggingface checkpoint folder",
+    )
+    parser.add_argument("--lowbit-path", type=str,
+        default="",
+        help="The path to the lowbit model folder, leave blank if you do not want to save. \
+            If path not exists, lowbit model will be saved there. \
+            Else, lowbit model will be loaded.",
+    )
+    parser.add_argument('--prompt', type=str, default="AI是什么?",
+                        help='Prompt to infer')
+    parser.add_argument("--n-predict", type=int, default=32, help="Max tokens to predict")
+    parser.add_argument("--max-context-len", type=int, default=1024)
+    parser.add_argument("--max-prompt-len", type=int, default=960)
+    parser.add_argument("--disable-transpose-value-cache", action="store_true", default=False)
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+
+    if not args.lowbit_path or not os.path.exists(args.lowbit_path):
+        model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                     optimize_model=True,
+                                                     pipeline=True,
+                                                     max_context_len=args.max_context_len,
+                                                     max_prompt_len=args.max_prompt_len,
+                                                     torch_dtype=torch.float16,
+                                                     attn_implementation="eager",
+                                                     transpose_value_cache=not args.disable_transpose_value_cache,
+                                                     mixed_precision=True,
+                                                     trust_remote_code=True)
+    else:
+        model = AutoModelForCausalLM.load_low_bit(
+            args.lowbit_path,
+            attn_implementation="eager",
+            torch_dtype=torch.float16,
+            max_context_len=args.max_context_len,
+            max_prompt_len=args.max_prompt_len,
+            pipeline=True,
+            transpose_value_cache=not args.disable_transpose_value_cache)
+
+    tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
+
+    if args.lowbit_path and not os.path.exists(args.lowbit_path):
+        model.save_low_bit(args.lowbit_path)
+
+    print("-" * 80)
+    print("done")
+    messages = [{"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": args.prompt}]
+    text = tokenizer.apply_chat_template(messages,
+                                         tokenize=False,
+                                         add_generation_prompt=True)
+    with torch.inference_mode():
+        print("finish to load")
+        for i in range(5):
+            _input_ids = tokenizer([text], return_tensors="pt").input_ids
+            print("input length:", len(_input_ids[0]))
+            st = time.time()
+            output = model.generate(
+                _input_ids, max_new_tokens=args.n_predict, do_print=True
+            )
+            end = time.time()
+            print(f"Inference time: {end-st} s")
+            input_str = tokenizer.decode(_input_ids[0], skip_special_tokens=False)
+            print("-" * 20, "Input", "-" * 20)
+            print(input_str)
+            output_str = tokenizer.decode(output[0], skip_special_tokens=False)
+            print("-" * 20, "Output", "-" * 20)
+            print(output_str)
+
+    print("-" * 80)
+    print("done")
+    print("success shut down")

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

@@ -267,6 +267,43 @@ def convert_minicpm(
         convert_forward(model, module.MiniCPMForCausalLM, minicpm_casullm_forward)
 
 
+def convert_qwen(
+        model: torch.nn.Module,
+        max_output_len=1024,
+        max_prompt_len=1024,
+        decoder=False,
+        inter_pp=None,
+        intra_pp=None,
+        transpose_value_cache=True,
+):
+    from ipex_llm.transformers.npu_models.qwen2_mp import gen_qwen2_fused_model_forward
+    from ipex_llm.transformers.npu_models.qwen2_mp import DecodeRunner, PrefillRunner
+    from transformers.models.qwen2.modeling_qwen2 import Qwen2Model
+    if decoder:
+        decode_runner = DecodeRunner(
+            model,
+            max_seq_len=max_output_len,
+            inter_pp=inter_pp,
+            intra_pp=intra_pp,
+            transpose_value_cache=transpose_value_cache,
+        )
+    else:
+        decode_runner = None
+    prefill_runner = PrefillRunner(
+        model,
+        max_output_len=max_output_len,
+        max_prompt_len=max_prompt_len,
+        transpose_value_cache=transpose_value_cache,
+    )
+    qwen2_model_forward = gen_qwen2_fused_model_forward(
+        prefill_runner=prefill_runner, decode_runner=decode_runner
+    )
+    convert_forward(model, Qwen2Model, qwen2_model_forward)
+    from transformers.models.qwen2.modeling_qwen2 import Qwen2ForCausalLM
+    from ipex_llm.transformers.npu_models.qwen2_mp import qwen2_casullm_forward
+    convert_forward(model, Qwen2ForCausalLM, qwen2_casullm_forward)
+
+
 def optimize_llm(
     model: torch.nn.Module,
     max_context_len=1024,
@@ -300,31 +337,13 @@ def optimize_llm(
                     inter_pp = 2
             else:
                 inter_pp = 1
-
+        convert_qwen(model,
-        from ipex_llm.transformers.npu_models.qwen2_mp import gen_qwen2_fused_model_forward
+                     max_output_len=max_context_len,
-        from ipex_llm.transformers.npu_models.qwen2_mp import DecodeRunner, PrefillRunner
+                     max_prompt_len=max_prompt_len,
-        from transformers.models.qwen2.modeling_qwen2 import Qwen2Model
+                     inter_pp=inter_pp,
-
+                     intra_pp=intra_pp,
-        decode_runner = DecodeRunner(
+                     decoder=True,
-            model,
+                     transpose_value_cache=transpose_value_cache)
-            max_seq_len=max_context_len,
-            inter_pp=inter_pp,
-            intra_pp=intra_pp,
-            transpose_value_cache=transpose_value_cache,
-        )
-        prefill_runner = PrefillRunner(
-            model,
-            max_output_len=max_context_len,
-            max_prompt_len=max_prompt_len,
-            transpose_value_cache=transpose_value_cache,
-        )
-        qwen2_model_forward = gen_qwen2_fused_model_forward(
-            prefill_runner=prefill_runner, decode_runner=decode_runner
-        )
-        convert_forward(model, Qwen2Model, qwen2_model_forward)
-        from transformers.models.qwen2.modeling_qwen2 import Qwen2ForCausalLM
-        from ipex_llm.transformers.npu_models.qwen2_mp import qwen2_casullm_forward
-        convert_forward(model, Qwen2ForCausalLM, qwen2_casullm_forward)
     elif model.config.model_type == "minicpm":
         # for minicpm-1b
         if intra_pp is None:

python/llm/src/ipex_llm/transformers/npu_models/qwen2_mp.py

@@ -140,31 +140,13 @@ class LowBitQwenMultiDecoderlayer(LLMBaseNNFactory):
 
         # Self Attention
         if mode == "decode":
-            attention_mask = self.create_input_op((self.batch_size, 1, 1, self.max_seq_len + 1))
+            attention_mask = self.create_input_op(
+                (self.batch_size, 1, 1, self.max_seq_len + 1), dtype=np.int64)
         else:
-            attention_mask = self.create_input_op((self.batch_size, 1, self.seq_len, self.seq_len))
+            attention_mask = self.create_input_op(
+                (self.batch_size, 1, self.seq_len, self.seq_len), dtype=np.int64)
 
-        position_ids = self.create_input_op((self.batch_size, self.seq_len))
+        position_ids = self.create_input_op((self.batch_size, self.seq_len), dtype=np.int64)
-        past_keys = []
-        past_values = []
-        if mode == "decode":
-            for i in range(num_layers):
-                past_key = self.create_cache_op(
-                    (self.batch_size, self.num_key_value_heads, self.max_seq_len, self.head_dim)
-                )
-                if transpose_value:
-                    past_value = self.create_cache_op(
-                        (self.batch_size, self.num_key_value_heads, self.head_dim, self.max_seq_len)
-                    )
-                else:
-                    past_value = self.create_cache_op(
-                        (self.batch_size, self.num_key_value_heads, self.max_seq_len, self.head_dim)
-                    )
-                past_keys.append(past_key)
-                past_values.append(past_value)
-        else:
-            past_keys = [None] * num_layers
-            past_values = [None] * num_layers
 
         if input_layernorm_weights is None:
             input_layernorm_weights = []
@@ -203,6 +185,27 @@ class LowBitQwenMultiDecoderlayer(LLMBaseNNFactory):
             k_biases = [self.constant(w) for w in k_biases]
             v_biases = [self.constant(w) for w in v_biases]
 
+        past_keys = []
+        past_values = []
+        if mode == "decode":
+            for i in range(num_layers):
+                past_key = self.create_cache_op(
+                    (self.batch_size, self.num_key_value_heads, self.max_seq_len, self.head_dim)
+                )
+                if transpose_value:
+                    past_value = self.create_cache_op(
+                        (self.batch_size, self.num_key_value_heads, self.head_dim, self.max_seq_len)
+                    )
+                else:
+                    past_value = self.create_cache_op(
+                        (self.batch_size, self.num_key_value_heads, self.max_seq_len, self.head_dim)
+                    )
+                past_keys.append(past_key)
+                past_values.append(past_value)
+        else:
+            past_keys = [None] * num_layers
+            past_values = [None] * num_layers
+
         hidden_states = input
 
         curr_key_values = []
@@ -396,8 +399,8 @@ class FusedQwenLowBitMultiDecoderlayer(torch.nn.Module):
 
         inputs = (
             hidden_states.to(torch.float16),
-            attention_mask,
+            attention_mask.to(torch.int64),
-            position_ids.to(torch.float16),
+            position_ids.to(torch.int64),
         )
 
         for i in range(self.intra_stages):
@@ -514,7 +517,9 @@ class FusedQwenLowBitDecoderlayer(torch.nn.Module):
         seq_len = hidden_states.shape[1]
 
         backend_cls = self.backend_cls_prefill
-        inputs = (hidden_states.to(torch.float16), attention_mask, position_ids.to(torch.float16))
+        inputs = (hidden_states.to(torch.float16),
+                  attention_mask.to(torch.int64),
+                  position_ids.to(torch.int64))
         inputs += (self.layer_norm_0, self.layer_norm_1)
         inputs += (self.q_bias, self.k_bias, self.v_bias)
         hidden_states, past_key, past_value = run_model(
@@ -687,9 +692,9 @@ def run_decode(
                 causal_mask[:, :, :, -1] = torch.finfo(torch.float16).min
                 pad_mask = (0, pad_len)
                 padded_causal_mask = F.pad(
-                    causal_mask.to(torch.float16), pad_mask, value=torch.finfo(torch.float16).min
+                    causal_mask.to(torch.int64), pad_mask, value=torch.iinfo(torch.int64).min
                 )
-                padded_causal_mask[:, :, :, -1] = 0.0
+                padded_causal_mask[:, :, :, -1] = 0
                 dist.recv(hidden_states, src=rank - 1)
                 layer_outputs = multi_decoder(
                     hidden_states,
@@ -973,9 +978,9 @@ class PrefillRunner:
         hidden_states = F.pad(hidden_states.to(torch.float16), (0, 0, 0, pad_len), value=0.0)
         position_ids = F.pad(position_ids, (0, pad_len), value=0)
         attention_mask = F.pad(
-            attention_mask.to(torch.float16),
+            attention_mask.to(torch.int64),
             (0, pad_len, 0, pad_len),
-            value=torch.finfo(torch.float16).min,
+            value=torch.iinfo(torch.int64).min,
         )
 
         args = (hidden_states, position_ids, attention_mask, past_key_value)

python/llm/src/ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py

@@ -196,7 +196,7 @@ def convert_llm(model: torch.nn.Module,
                 group_size: int):
     if group_size == 0:
         n_splits_linear = 1
-        n_splits_down_proj = 1
+        n_splits_down_proj = 2 if model.config.intermediate_size == 18944 else 1
     else:
         n_splits_linear = model.config.hidden_size // group_size
         n_splits_down_proj = model.config.intermediate_size // group_size
@@ -318,9 +318,49 @@ def convert_llm(model: torch.nn.Module,
             except:
                 invalidInputError(False,
                                   "False to InitLLMPipeline.")
+    elif model.config.model_type == "qwen2":
+        with tempfile.TemporaryDirectory() as temp_dir:
+            weight_dir = os.path.join(temp_dir, "model_weights")
+            os.mkdir(weight_dir)
+            layer_num = len(model.model.layers)
+            from .qwen import convert_qwen_layer, convert_lm_head_and_embedding
+            first_blob_path, last_blob_path = convert_lm_head_and_embedding(model, n_splits_linear,
+                                                                            temp_dir, weight_dir)
+
+            param_list = []
+            for layer_idx in range(0, layer_num):
+                param_list.append((model, layer_idx, n_splits_linear, n_splits_down_proj,
+                                  temp_dir, weight_dir, transpose_value_cache, kv_len, group_size))
+            with Pool() as pool:
+                result = pool.starmap(convert_qwen_layer, param_list)
+
+            # Prefill Runner
+            from ipex_llm.transformers.npu_models.convert_mp import convert_qwen
+            convert_qwen(model,
+                         max_output_len=kv_len,
+                         max_prompt_len=max_prompt_len,
+                         decoder=False,
+                         transpose_value_cache=transpose_value_cache)
+
+            # patch attrs for generate
+            model.kv_len = kv_len
+            model.num_head = model.model.layers[0].self_attn.num_key_value_heads
+            model.head_dim = model.model.layers[0].self_attn.head_dim
+            model.num_layers = layer_num
+            model.transpose_value_cache = transpose_value_cache
+            model.vocab_size = model.config.vocab_size
+
+            try:
+                res = InitLLMPipeline("qwen", kv_len, model.num_head, model.head_dim, layer_num,
+                                      model.vocab_size, weight_dir, "model",
+                                      first_blob_path, last_blob_path,
+                                      os.path.join(temp_dir, "decoder_layer"))
+            except:
+                invalidInputError(False,
+                                  "False to InitLLMPipeline.")
     else:
-        invalidInputError(False,
+        invalidInputError(False, "Now we only support Llama2 / Llama3 / Baichuan2 / "
-                          "Now we only support Llama2 / Llama3 / Baichuan2 for pipeline running.")
+                                 "Qwen2 / Qwen2.5 / Minicpm for pipeline running.")
 
     if isinstance(model.lm_head, SlicedLMHead):
         model.lm_head.get_fused_lm_head()

python/llm/src/ipex_llm/transformers/npu_pipeline_model/qwen.py

@@ -0,0 +1,187 @@
+#
+# 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.
+#
+
+
+import torch
+import numpy as np
+import os
+from .common import update_names_of_IR_and_export_blob, LLMEmbedding, LowBitLLMLMHead
+
+
+def convert_lm_head_and_embedding(model, n_splits_linear, temp_dir, weight_dir):
+    num_heads = model.model.layers[0].self_attn.num_heads
+    head_dim = model.model.layers[0].self_attn.head_dim
+    rms_norm_eps = model.config.rms_norm_eps
+    vocab_size = model.config.vocab_size
+    model_norm = model.model.norm
+    lm_heads = model.lm_head.lm_heads  # Qwen2 is always SlicedLMHead
+    if n_splits_linear == 1:
+        weights = [(lm_heads[0].weight, lm_heads[0].scale)]
+    else:
+        lm_head_weights = []
+        scales = []
+        for i in range(n_splits_linear):
+            lm_head_weights.append(lm_heads[i].weight)
+            scales.append(lm_heads[i].scale)
+        weights = [(torch.stack(lm_head_weights, axis=0),
+                    torch.stack(scales, axis=0))]
+    if isinstance(weights[0], tuple):
+        np_dtype = np.int8 if weights[0][0].dtype == torch.int8 else np.uint8
+    else:  # FP16 Linear
+        np_dtype = np.float16
+
+    new_lm_head = LowBitLLMLMHead(
+        [1, 1, num_heads * head_dim],
+        num_heads=num_heads,
+        max_seq_len=1,  # seems doesn't matter
+        rms_norm_eps=rms_norm_eps,
+        mode="decode",
+        transpose_value=False,  # seems doesn't matter
+        dtype=np_dtype,
+        model_norm_weight=model_norm.weight.to(torch.float16),
+        vocab_size=vocab_size,
+        n_splits=n_splits_linear
+    )
+    last_blob_path = update_names_of_IR_and_export_blob(new_lm_head, "lm_head", temp_dir)
+
+    # save weights bins files
+    if n_splits_linear == 1:
+        weight_numpy = [
+            lm_heads[0].weight.data.numpy(), lm_heads[0].scale.data.numpy(),
+        ]
+    else:
+        weight_numpy = [v.numpy() for v in weights[0]]
+
+    for idx, weight in enumerate(weight_numpy):
+        bin_file = os.path.join(weight_dir, f"model_lm_head_input_{1+idx}.bin")
+        weight.tofile(bin_file)
+
+    embedding_layer = model.model.embed_tokens
+    new_embedding = LLMEmbedding(
+        vocab_size=model.config.vocab_size,
+        embedding_dim=model.config.hidden_size,
+        embedding_weight=embedding_layer.weight.to(torch.float16).detach().numpy(),
+        padding_idx=model.config.pad_token_id,
+        dtype=np.float16,
+    )
+    first_blob_path = update_names_of_IR_and_export_blob(new_embedding, "embedding",
+                                                         temp_dir)
+    return first_blob_path, last_blob_path
+
+
+def convert_qwen_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
+                       temp_dir, weight_dir, transpose_value_cache, kv_len, group_size):
+    num_heads = model.model.layers[0].self_attn.num_heads
+    num_key_value_heads = model.model.layers[0].self_attn.num_key_value_heads
+    head_dim = model.model.layers[0].self_attn.head_dim
+    intermediate_size = model.config.intermediate_size
+    rms_norm_eps = model.config.rms_norm_eps
+
+    from ipex_llm.transformers.npu_models.qwen2_mp import LowBitQwenMultiDecoderlayer
+    curr_layer = model.model.layers[layer_idx]
+    attn_layer = curr_layer.self_attn
+    mlp_layer = curr_layer.mlp
+
+    weights = []
+    if n_splits_linear == 1:
+        for q, k, v, o, g, u in zip(attn_layer.q_proj_dq_list,
+                                    attn_layer.k_proj_dq_list,
+                                    attn_layer.v_proj_dq_list,
+                                    attn_layer.o_proj_dq_list,
+                                    mlp_layer.gate_proj_dq_list,
+                                    mlp_layer.up_proj_dq_list):
+            weights.append((q.weight, q.scale))
+            weights.append((k.weight, k.scale))
+            weights.append((v.weight, v.scale))
+            weights.append((o.weight, o.scale))
+            weights.append((g.weight, g.scale))
+            weights.append((u.weight, u.scale))
+    else:
+        for layer_list in [attn_layer.q_proj_dq_list, attn_layer.k_proj_dq_list,
+                           attn_layer.v_proj_dq_list, attn_layer.o_proj_dq_list,
+                           mlp_layer.gate_proj_dq_list, mlp_layer.up_proj_dq_list]:
+            l_weights = []
+            scales = []
+            for l in layer_list:
+                l_weights.append(l.weight)
+                scales.append(l.scale)
+            weights.append((torch.stack(l_weights, axis=0), torch.stack(scales, axis=0)))
+
+    if n_splits_down_proj == 1:
+        for l in mlp_layer.down_proj_dq_list:
+            weights.append((l.weight, l.scale))
+    else:
+        l_weights = []
+        scales = []
+        for l in mlp_layer.down_proj_dq_list:
+            l_weights.append(l.weight)
+            scales.append(l.scale)
+        weights.append((torch.stack(l_weights, axis=0), torch.stack(scales, axis=0)))
+
+    q_bias = attn_layer.q_proj_dq_list.q_proj_dq_0.bias.to(torch.float16)
+    k_bias = attn_layer.k_proj_dq_list.k_proj_dq_0.bias.to(torch.float16)
+    v_bias = attn_layer.v_proj_dq_list.v_proj_dq_0.bias.to(torch.float16)
+    cached_cos = curr_layer.self_attn.rotary_emb.cos_cached.to(torch.float16)
+    cached_sin = curr_layer.self_attn.rotary_emb.sin_cached.to(torch.float16)
+    layer_norm_0 = curr_layer.input_layernorm.weight.to(torch.float16)
+    layer_norm_1 = curr_layer.post_attention_layernorm.weight.to(torch.float16)
+
+    if isinstance(weights[0], tuple):
+        np_dtype = np.int8 if weights[0][0].dtype == torch.int8 else np.uint8
+    else:  # FP16 Linear
+        np_dtype = np.float16
+
+    single_decoder = LowBitQwenMultiDecoderlayer(
+        [1, 1, num_heads * head_dim],
+        input_layernorm_weights=[layer_norm_0],
+        post_attn_layernorm_weights=[layer_norm_1],
+        q_biases=None,
+        k_biases=None,
+        v_biases=None,
+        cached_cos=cached_cos,
+        cached_sin=cached_sin,
+        num_heads=num_heads,
+        num_key_value_heads=num_key_value_heads,
+        num_layers=1,
+        max_seq_len=kv_len,
+        rms_norm_eps=rms_norm_eps,
+        intermediate_size=intermediate_size,
+        mode="decode",
+        transpose_value=transpose_value_cache,
+        dtype=np_dtype,
+        n_splits_linear=n_splits_linear,
+        n_splits_down_proj=n_splits_down_proj,
+        group_size=group_size
+    )
+    rest_blob_path = update_names_of_IR_and_export_blob(single_decoder,
+                                                        f"decoder_layer_{layer_idx}",
+                                                        temp_dir)
+
+    # 0, 1, 2 are input_embed/attention_mask/position_id
+    q_bias_bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_3.bin")
+    k_bias_bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_4.bin")
+    v_bias_bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_5.bin")
+    q_bias.data.numpy().tofile(q_bias_bin_file)
+    k_bias.data.numpy().tofile(k_bias_bin_file)
+    v_bias.data.numpy().tofile(v_bias_bin_file)
+    # 6, 7 are past k/v
+    for idx, (weight, scale) in enumerate(weights):
+        bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_{8+idx*2}.bin")
+        weight.numpy().tofile(bin_file)
+        bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_{8+idx*2+1}.bin")
+        scale.numpy().tofile(bin_file)
+
+    del single_decoder