[NPU] Fix MTL and ARL support (#12580)

intel-analytics · Dec 19, 2024 · 4e7e988 · 4e7e988
1 parent 80f2fdc
commit 4e7e988
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 3 deletions.
diff --git a/docs/mddocs/Quickstart/npu_quickstart.md b/docs/mddocs/Quickstart/npu_quickstart.md
@@ -88,6 +88,7 @@ For `ipex-llm` NPU support, please set the following environment variable with a
 - For **Intel Core™ Ultra Processors (Series 2) with processor number 2xxV (code name Lunar Lake)**:
 
   - For Intel Core™ Ultra 7 Processor 258V:
+
       No runtime configuration required.
 
   - For Intel Core™ Ultra 5 Processor 228V & 226V:
@@ -97,7 +98,7 @@ For `ipex-llm` NPU support, please set the following environment variable with a
 
 - For **Intel Core™ Ultra Processors (Series 2) with processor number 2xxK (code name Arrow Lake)**:
    ```cmd
-   set IPEX_LLM_NPU_DISABLE_COMPILE_OPT=1
+   set IPEX_LLM_NPU_ARL=1
    ```
 
 - For **Intel Core™ Ultra Processors (Series 1) with processor number 1xxH (code name Meteor Lake)**:

diff --git a/python/llm/src/ipex_llm/transformers/npu_models/convert_mp.py b/python/llm/src/ipex_llm/transformers/npu_models/convert_mp.py
@@ -37,6 +37,10 @@ def optimize_llm_pre(model: torch.nn.Module, qtype, mixed_precision,
         os.environ["IPEX_LLM_NPU_USE_LEVEL0"] = "0"
         os.environ["IPEX_LLM_NPU_DISABLE_COMPILE_OPT"] = "1"
 
+    if os.environ.get("IPEX_LLM_NPU_ARL", "0") == "1":
+        # For ARL support
+        os.environ["IPEX_LLM_NPU_DISABLE_COMPILE_OPT"] = "1"
+
     if model.config.model_type == "baichuan":
         # process NormHead module in Baichuan2 7B
         if hasattr(model, 'lm_head') and model.lm_head is not None:
@@ -144,7 +148,9 @@ def optimize_llm_pre(model: torch.nn.Module, qtype, mixed_precision,
                 # do not split mlp down_proj for Qwen2-7B & sym_int8
                 n_splits_down_proj = 1
             else:
-                n_splits_down_proj = 2 if model.config.intermediate_size == 18944 else 1
+                n_splits_down_proj = 2 if (model.config.intermediate_size == 18944 or
+                                           os.environ.get("IPEX_LLM_NPU_MTL", "0") == "1" or
+                                           os.environ.get("IPEX_LLM_NPU_ARL", "0") == "1") else 1
         else:
             invalidInputError(
                 model.config.hidden_size % quantization_group_size == 0 and

diff --git a/python/llm/src/ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py b/python/llm/src/ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py
@@ -205,7 +205,9 @@ def convert_llm(model: torch.nn.Module,
             # do not split mlp down_proj for Qwen2-7B & sym_int8
             n_splits_down_proj = 1
         else:
-            n_splits_down_proj = 2 if model.config.intermediate_size == 18944 else 1
+            n_splits_down_proj = 2 if (model.config.intermediate_size == 18944 or
+                                       os.environ.get("IPEX_LLM_NPU_MTL", "0") == "1" or
+                                       os.environ.get("IPEX_LLM_NPU_ARL", "0") == "1") else 1
     else:
         n_splits_linear = model.config.hidden_size // group_size
         n_splits_down_proj = model.config.intermediate_size // group_size