Support PP for qwen1.5 (#11300)

python/llm/example/GPU/Pipeline-Parallel-Inference/README.md

@@ -6,9 +6,11 @@ This example demonstrates how to run IPEX-LLM optimized low-bit model vertically
 To run this example with IPEX-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information. For this particular example, you will need at least two GPUs on your machine.
 
 ## Verified Models
-- [Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf)
-- [Llama-2-13b-chat-hf](https://huggingface.co/meta-llama/Llama-2-13b-chat-hf)
-- [Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct)
+- [meta-llama/Llama-2-7b-chat-hf](./run_llama_arc_2_card.sh)
+- [meta-llama/Llama-2-13b-chat-hf](./run_llama_arc_2_card.sh)
+- [meta-llama/Meta-Llama-3-8B-Instruct](./run_llama_arc_2_card.sh)
+- [Qwen/Qwen1.5-7B-Chat](./run_qwen1.5_arc_2_card.sh)
+- [Qwen/Qwen1.5-14B-Chat](./run_qwen1.5_arc_2_card.sh)
 
 ## Example: Run pipeline parallel inference on multiple GPUs
 
@@ -28,18 +30,41 @@ pip install oneccl_bind_pt==2.1.100 --extra-index-url https://pytorch-extension.
 
 ### 2. Run pipeline parallel inference on multiple GPUs
 
-For optimal performance, it is recommended to set several environment variables. We provide example usage as following:
+For optimal performance, it is recommended to set several environment variables. We provide example usages as following:
 
-- Run Llama-2-13b-chat-hf on two Intel Arc A770
+</details>
+
+<details>
+  <summary> Show Llama2 and Llama3 example </summary>
+
+#### Run Llama-2-7b-chat-hf / Llama-2-13b-chat-hf / Meta-Llama-3-8B-Instruct on two Intel Arc A770
+
+You could specify `--repo-id-or-model-path` in the test script to be the huggingface repo id for Llama2 / Llama3 to be downloaded, or the path to the huggingface checkpoint folder. Besides, you could change `NUM_GPUS` to the number of GPUs you have on your machine.
 
 ```bash
-bash run_llama2_13b_arc_2_card.sh
+bash run_llama_arc_2_card.sh
 ```
 
-> **Note**: You could change `NUM_GPUS` to the number of GPUs you have on your machine.
+</details>
 
-#### Sample Output
-##### [meta-llama/Llama-2-13b-chat-hf](https://huggingface.co/meta-llama/Llama-2-13b-chat-hf)
+</details>
+
+<details>
+  <summary> Show Qwen1.5 example </summary>
+
+#### Run Qwen1.5-7B-Chat / Qwen1.5-14B-Chat on two Intel Arc A770
+
+You could specify `--repo-id-or-model-path` in the test script to be the huggingface repo id for Qwen1.5 to be downloaded, or the path to the huggingface checkpoint folder. Besides, you could change `NUM_GPUS` to the number of GPUs you have on your machine.
+
+```bash
+pip install transformers==4.37.0
+bash run_qwen1.5_arc_2_card.sh
+```
+
+</details>
+
+### 3. Sample Output
+#### [meta-llama/Llama-2-13b-chat-hf](https://huggingface.co/meta-llama/Llama-2-13b-chat-hf)
 ```log
 Inference time: xxxx s
 First token cost xxxx s and rest tokens cost average xxxx s

python/llm/example/GPU/Pipeline-Parallel-Inference/run_llama_arc_2_card.sh

@@ -0,0 +1,40 @@
+#
+# 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.
+#
+
+source /opt/intel/oneapi/setvars.sh
+export MASTER_ADDR=127.0.0.1
+export MASTER_PORT=9090
+export FI_PROVIDER=tcp
+export USE_XETLA=OFF
+export OMP_NUM_THREADS=6
+if [[ $KERNEL_VERSION != *"6.5"* ]]; then
+    export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+fi
+export TORCH_LLM_ALLREDUCE=0
+
+NUM_GPUS=2 # number of used GPU
+
+# To run Llama-2-7b-chat-hf
+CCL_ZE_IPC_EXCHANGE=sockets torchrun --standalone --nnodes=1 --nproc-per-node $NUM_GPUS \
+    generate.py --repo-id-or-model-path 'meta-llama/Llama-2-7b-chat-hf' --gpu-num $NUM_GPUS
+
+# # To run Llama-2-13b-chat-hf
+# CCL_ZE_IPC_EXCHANGE=sockets torchrun --standalone --nnodes=1 --nproc-per-node $NUM_GPUS \
+#     generate.py --repo-id-or-model-path 'meta-llama/Llama-2-13b-chat-hf' --gpu-num $NUM_GPUS
+
+# # To run Meta-Llama-3-8B-Instruct
+# CCL_ZE_IPC_EXCHANGE=sockets torchrun --standalone --nnodes=1 --nproc-per-node $NUM_GPUS \
+#     generate.py --repo-id-or-model-path 'meta-llama/Meta-Llama-3-8B-Instruct' --gpu-num $NUM_GPUS

python/llm/example/GPU/Pipeline-Parallel-Inference/run_qwen1.5_arc_2_card.sh

@@ -26,5 +26,11 @@ fi
 export TORCH_LLM_ALLREDUCE=0
 
 NUM_GPUS=2 # number of used GPU
+
+# To run Qwen1.5-7B-Chat
 CCL_ZE_IPC_EXCHANGE=sockets torchrun --standalone --nnodes=1 --nproc-per-node $NUM_GPUS \
-    generate.py --repo-id-or-model-path 'meta-llama/Llama-2-13b-chat-hf' --gpu-num $NUM_GPUS
+    generate.py --repo-id-or-model-path 'Qwen/Qwen1.5-7B-Chat' --gpu-num $NUM_GPUS
+
+# # To run Qwen1.5-14B-Chat
+# CCL_ZE_IPC_EXCHANGE=sockets torchrun --standalone --nnodes=1 --nproc-per-node $NUM_GPUS \
+#     generate.py --repo-id-or-model-path 'Qwen/Qwen1.5-14B-Chat' --gpu-num $NUM_GPUS

python/llm/src/ipex_llm/transformers/models/qwen2.py

@@ -73,9 +73,10 @@ def qwen2_model_forward(
     return_dict: Optional[bool] = None,
 ):
     use_cache = use_cache if use_cache is not None else self.config.use_cache
+    input = input_ids if input_ids is not None else inputs_embeds
     use_quantize_kv = (
         self.config.hidden_size != 3584     # disable quantize kv in specific model
-        and use_quantize_kv_cache(self.layers[0].mlp.up_proj, input_ids)
+        and use_quantize_kv_cache(self.layers[0].mlp.up_proj, input)
     )
     if use_cache:
         if use_quantize_kv and not isinstance(past_key_values, DynamicFp8Cache):