Add OpenVINO weights compression to docs

docs/source/optimization_ov.mdx

@@ -62,6 +62,27 @@ tokenizer.save_pretrained(save_dir)
 
 The `quantize()` method applies post-training static quantization and export the resulting quantized model to the OpenVINO Intermediate Representation (IR). The resulting graph is represented with two files: an XML file describing the network topology and a binary file describing the weights. The resulting model can be run on any target Intel device.
 
+### Weights compression
+
+For large language models (LLMs), it is often beneficial to only quantize weights, and keep activations in floating point precision. This method does not require a calibration dataset. To enable weights compression, set the `weights_only` parameter of `OVQuantizer`:
+
+```python
+from optimum.intel.openvino import OVQuantizer, OVModelForCausalLM
+from transformers import AutoModelForCausalLM
+
+save_dir = "int8_weights_compressed_model"
+model = AutoModelForCausalLM.from_pretrained("databricks/dolly-v2-3b")
+quantizer = OVQuantizer.from_pretrained(model, task="text-generation")
+quantizer.quantize(save_directory=save_dir, weights_only=True)
+```
+
+To load the optimized model for inference:
+
+```python
+optimized_model = OVModelForCausalLM.from_pretrained(save_dir)
+```
+
+Weights compression is enabled for PyTorch and OpenVINO models: the starting model can be an `AutoModelForCausalLM` or `OVModelForCausalLM` instance.
 
 ## Training-time optimization
 
@@ -221,4 +242,4 @@ text = "He's a dreadful magician."
 outputs = cls_pipe(text)
 
 [{'label': 'NEGATIVE', 'score': 0.9840195178985596}]
-```
+```