Replace deprecated openvino.runtime (#1280)

* remove deprecated openvino.runtime

* remove deprecated openvino.runtime

* remove deprecated openvino.runtime

* remove unused

* fix, test and distribute notebooks

* skip diffusion

* sudo

* skip senetence transformers notebook as well until quantization recipe is fixed

* rm amd

---------

Co-authored-by: IlyasMoutawwakil <[email protected]>

Author: echarlaix

.github/workflows/test_openvino_notebooks.yml

@@ -5,13 +5,12 @@ on:
   schedule:
     - cron: "14 3 * * 1" # run weekly: every Monday at 3:14
   push:
-    paths:
-      - ".github/workflows/test_openvino_notebooks.yml"
-      - "notebooks/openvino/*"
+    branches:
+      - v*-release
   pull_request:
     paths:
-      - ".github/workflows/test_openvino_notebooks.yml"
       - "notebooks/openvino/*"
+      - ".github/workflows/test_openvino_notebooks.yml"
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
@@ -25,7 +24,13 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        python-version: ["3.9", "3.12"]
+        python-version: [3.9]
+        test_file: [
+            "optimum_openvino_inference.ipynb",
+            "question_answering_quantization.ipynb",
+            # "sentence_transformer_quantization.ipynb", TODO: fix and run on any cpu
+            # "stable_diffusion_hybrid_quantization.ipynb", TODO: update and ran on a powerful cpu
+          ]
 
     runs-on: ubuntu-22.04
 
@@ -38,16 +43,20 @@ jobs:
         with:
           python-version: ${{ matrix.python-version }}
 
+      - name: Install packages
+        run: |
+          sudo apt-get install -y ffmpeg
+
       - name: Install dependencies
         run: |
-          # ffmpeg is required for image classification and audio classification pipelines
-          sudo apt-get install ffmpeg
           pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
           pip install -r notebooks/openvino/requirements.txt
           pip install .[tests,openvino] nbval
 
+      - name: Change some variables
+        run: |
+          sed -i 's/NUM_TRAIN_ITEMS = 600/NUM_TRAIN_ITEMS = 2/' notebooks/openvino/question_answering_quantization.ipynb
+
       - name: Test with Pytest
         run: |
-          sed -i 's/NUM_TRAIN_ITEMS = 600/NUM_TRAIN_ITEMS = 10/' notebooks/openvino/question_answering_quantization.ipynb
-          sed -i 's/# %pip install/%pip install/' notebooks/openvino/optimum_openvino_inference.ipynb
-          python -m pytest --nbval-lax notebooks/openvino/optimum_openvino_inference.ipynb  notebooks/openvino/question_answering_quantization.ipynb
+          pytest --nbval-lax -vvvv notebooks/openvino/${{ matrix.test_file }}

notebooks/openvino/quantized_generation_demo.ipynb renamed to notebooks/openvino/demos/quantized_generation_demo.ipynb

@@ -45,6 +45,7 @@
     "import os\n",
     "\n",
     "from transformers import AutoTokenizer\n",
+    "\n",
     "from optimum.intel import OVModelForCausalLM, OVWeightQuantizationConfig"
    ]
   },
@@ -211,6 +212,7 @@
    "source": [
     "from transformers import TextStreamer\n",
     "\n",
+    "\n",
     "# Tokenize the sample\n",
     "inputs = tokenizer([sample], return_tensors='pt')\n",
     "\n",
@@ -294,15 +296,15 @@
     "\n",
     "\n",
     "# Tokenize the sample\n",
-    "inputs = tokenizer([sample], return_tensors='pt')    \n",
+    "inputs = tokenizer([sample], return_tensors='pt')\n",
     "\n",
     "out = stateless_model.generate(\n",
     "    **inputs,\n",
     "    max_new_tokens=128,\n",
     "    streamer=TextStreamer(tokenizer=tokenizer, skip_special_tokens=True),\n",
     "    pad_token_id=tokenizer.eos_token_id,\n",
     "    prompt_lookup_num_tokens=3,\n",
-    ")    "
+    ")"
    ]
   },
   {
@@ -442,6 +444,7 @@
    "outputs": [],
    "source": [
     "from functools import wraps\n",
+    "\n",
     "import numpy as np\n",
     "\n",
     "\n",
@@ -458,15 +461,15 @@
     "        if len(self.seq_lens) > 0 or len(self.win_sizes) > 0:\n",
     "            raise RuntimeError(\"Always use a new instance, don't reuse!\")\n",
     "        self.model_forward = self.model.forward\n",
-    "        \n",
+    "\n",
     "        @wraps(self.model_forward)\n",
     "        def forward_wrapper(**kwargs):\n",
     "            self.seq_lens[-1].append(kwargs.get(\"attention_mask\").shape[-1])\n",
     "            self.win_sizes[-1].append(kwargs.get(\"input_ids\").shape[-1] - 1)\n",
     "            return self.model_forward(**kwargs)\n",
-    "        \n",
+    "\n",
     "        self.model.forward = forward_wrapper\n",
-    "        \n",
+    "\n",
     "        # wrap generate method\n",
     "        self.model_generate = self.model.generate\n",
     "\n",
@@ -494,7 +497,7 @@
     "        self.seq_lens = [sl[1:] for sl in self.seq_lens]\n",
     "        # Add window size for output to ease calculation later\n",
     "        for ws, sl in zip(self.win_sizes, self.seq_lens):\n",
-    "            ws.append(0)    \n",
+    "            ws.append(0)\n",
     "\n",
     "    def acceptance_rate(self, return_mean=True, normalize=False):\n",
     "        # ar_per_win = ((cur_seq_len - cur_win_size) - (prev_seq_len - prev_win_size) - 1) / prev_win_size\n",
@@ -533,8 +536,9 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "from tqdm import tqdm\n",
     "from datasets import load_dataset\n",
+    "from tqdm import tqdm\n",
+    "\n",
     "\n",
     "dataset_name = \"openai_humaneval\"\n",
     "dataset_subset_name = None\n",
@@ -590,10 +594,10 @@
     "from threading import Thread\n",
     "\n",
     "from transformers import (\n",
-    "    TextIteratorStreamer,\n",
+    "    GenerationConfig,\n",
     "    StoppingCriteria,\n",
     "    StoppingCriteriaList,\n",
-    "    GenerationConfig,\n",
+    "    TextIteratorStreamer,\n",
     ")\n",
     "\n",
     "\n",
@@ -690,7 +694,7 @@
     "    prompt_char = \"▌\"\n",
     "    history[-1][1] = prompt_char\n",
     "    yield history, \"Status: Generating...\", *([gr.update(interactive=False)] * 4)\n",
-    "    \n",
+    "\n",
     "    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)\n",
     "\n",
     "    # Create a stopping criteria to prevent the model from playing the role of the user aswell.\n",
@@ -770,6 +774,7 @@
    "source": [
     "import gradio as gr\n",
     "\n",
+    "\n",
     "try:\n",
     "    demo.close()\n",
     "except:\n",
@@ -808,7 +813,7 @@
     "      history: conversation history\n",
     "    Returns:\n",
     "      updated history\n",
-    "    \"\"\" \n",
+    "    \"\"\"\n",
     "    history[-1][1] = None\n",
     "    return history\n",
     "\n",