add ipex readme

README.md

@@ -44,6 +44,7 @@ where `extras` can be one or more of `ipex`, `neural-compressor`, `openvino`, `n
 
 # Quick tour
 
+
 ## Neural Compressor
 
 Dynamic quantization can be used through the Optimum command-line interface:
@@ -202,6 +203,28 @@ Quantization aware training (QAT) is applied in order to simulate the effects of
 You can find more examples in the [documentation](https://huggingface.co/docs/optimum/intel/index).
 
 
+## IPEX
+With `export=True`, IPEX model will replace torch linear to ipex linear which prepacks the weights. It will also apply linear fusion and [IAKV](https://intel.github.io/intel-extension-for-pytorch/cpu/latest/tutorials/llm.html#indirect-access-kv-cache) for generation. Finally, jit.trace will be applied to change the model to graph mode.
+Here is the example of how to use IPEX optimized model to generate texts.
+### generate
+```diff
+  from transformers import AutoTokenizer, pipeline
+- from transformers import AutoModelForCausalLM
++ from optimum.intel.ipex import IPEXModelForCausalLM
+
+
+  model_id = "gpt2"
+- model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16)
++ model = IPEXModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, export=True)
+  tokenizer = AutoTokenizer.from_pretrained(model_id)
+  pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
+  results = pipe("He's a dreadful magician and")
+
+```
+
+For more details, please refer to the [documentation](https://intel.github.io/intel-extension-for-pytorch/#introduction).
+
+
 ## Running the examples
 
 Check out the [`examples`](https://github.com/huggingface/optimum-intel/tree/main/examples) directory to see how 🤗 Optimum Intel can be used to optimize models and accelerate inference.