Remove OneDNN env setint for BF16 inference (#509)

Signed-off-by: lvliang-intel <liang1.lv@intel.com> Co-authored-by: VincyZhang <wenxin.zhang@intel.com>

Author: lvliang-intel

intel_extension_for_transformers/neural_chat/models/model_utils.py

@@ -254,11 +254,6 @@ def init_deepspeed_inference(model, model_name_or_path, use_hpu_graphs, is_meta)
  model = deepspeed.init_inference(model, **ds_inference_kwargs)
  return model.module
 
-
-def set_cpu_running_env():
- os.environ["ONEDNN_MAX_CPU_ISA"] = "AVX512_CORE_BF16"
-
-
 def load_model(
  model_name,
  tokenizer_name,
@@ -298,8 +293,6 @@ def load_model(
  )
 
  adapt_transformers_to_gaudi()
- elif device == "cpu" and not ipex_int8:
- set_cpu_running_env()
 
  if isinstance(optimization_config, AMPConfig):
  dtype = optimization_config.dtype

intel_extension_for_transformers/neural_chat/pipeline/plugins/ner/ner.py

@@ -28,7 +28,6 @@
 )
 import intel_extension_for_pytorch as intel_ipex
 from .utils.utils import (
- set_cpu_running_env, 
  enforce_stop_tokens,
  get_current_time
 )
@@ -43,9 +42,6 @@ class NamedEntityRecognition():
  """
 
  def __init__(self, model_path="./Llama-2-7b-chat-hf/", spacy_model="en_core_web_lg", bf16: bool=False) -> None:
- # set up cpu running environment
- if bf16:
- set_cpu_running_env()
  # initialize tokenizer and models
  self.nlp = spacy.load(spacy_model)
  config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)

intel_extension_for_transformers/neural_chat/pipeline/plugins/ner/utils/utils.py

@@ -35,7 +35,3 @@ def get_current_time() -> str:
  utc_now = datetime.datetime.utcnow().replace(tzinfo=timezone.utc)
  cur_time = utc_now.astimezone(SHA_TZ).strftime("%Y/%m/%d")
  return cur_time
-
-
-def set_cpu_running_env():
- os.environ["ONEDNN_MAX_CPU_ISA"] = "AVX512_CORE_BF16"