Merge pull request #169 from roboflow/deps/transformers/qwen

deps: 📦 update transformers version constraints and better qwen2.5 VL support for installation
roboflow · Feb 18, 2025 · d3daf16 · d3daf16
2 parents 26835d8 + 4c0c9ae
commit d3daf16
Show file tree

Hide file tree

Showing 5 changed files with 107 additions and 78 deletions.
diff --git a/cookbooks/maestro_florence_2_object_detection.ipynb b/cookbooks/maestro_florence_2_object_detection.ipynb
diff --git a/cookbooks/maestro_paligemma_2_json_extraction.ipynb b/cookbooks/maestro_paligemma_2_json_extraction.ipynb
@@ -6,13 +6,13 @@
     "id": "ckbZSXr_nXrD"
    },
    "source": [
-        "# Fine-tuning PaliGemma 2 for JSON Data Extraction using Maestro\n",
-        "\n",
-        "---\n",
-        "\n",
-        "[![colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/roboflow/maestro/blob/develop/cookbooks/maestro_paligemma_2_json_extraction.ipynb)\n",
-        "[![dataset](https://app.roboflow.com/images/download-dataset-badge.svg)](https://universe.roboflow.com/roboflow-jvuqo/pallet-load-manifest)"
-      ]
+    "# Fine-tuning PaliGemma 2 for JSON Data Extraction using Maestro\n",
+    "\n",
+    "---\n",
+    "\n",
+    "[![colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/roboflow/maestro/blob/develop/cookbooks/maestro_paligemma_2_json_extraction.ipynb)\n",
+    "[![dataset](https://app.roboflow.com/images/download-dataset-badge.svg)](https://universe.roboflow.com/roboflow-jvuqo/pallet-load-manifest)"
+   ]
   },
   {
    "cell_type": "markdown",
@@ -110,7 +110,7 @@
    },
    "outputs": [],
    "source": [
-    "!pip install \"maestro[paligemma_2]\"\n",
+    "!pip install \"maestro[paligemma_2]==1.1.0rc2\"\n",
     "!pip install roboflow"
    ]
   },
@@ -444,17 +444,17 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": null,
    "metadata": {
     "id": "bOyiFToVjGEK"
    },
    "outputs": [],
    "source": [
-    "from maestro.trainer.common.datasets import RoboflowJSONLDataset\n",
+    "from maestro.trainer.common.datasets.jsonl import JSONLDataset\n",
     "\n",
-    "ds = RoboflowJSONLDataset(\n",
-    "    jsonl_file_path=f\"{dataset.location}/test/annotations.jsonl\",\n",
-    "    image_directory_path=f\"{dataset.location}/test\",\n",
+    "ds = JSONLDataset(\n",
+    "    annotations_path=f\"{dataset.location}/test/annotations.jsonl\",\n",
+    "    images_directory_path=f\"{dataset.location}/test\",\n",
     ")\n",
     "\n",
     "image, entry = ds[0]"

diff --git a/cookbooks/maestro_qwen2_5_vl_json_extraction.ipynb b/cookbooks/maestro_qwen2_5_vl_json_extraction.ipynb
diff --git a/docs/index.md b/docs/index.md
@@ -67,13 +67,8 @@ we recommend creating a dedicated Python environment for each model.
 
     ```bash
     pip install "maestro[qwen_2_5_vl]"
-    pip install git+https://github.com/huggingface/transformers
     ```
 
-    !!! warning
-        Support for Qwen2.5-VL in transformers is experimental.
-        For now, please install transformers from source to ensure compatibility.
-
 ### CLI
 
 Kick off fine-tuning with our command-line interface, which leverages the configuration

diff --git a/pyproject.toml b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "maestro"
-version = "1.1.0rc1"
+version = "1.1.0rc2"
 description = "Streamline the fine-tuning process for vision-language models like PaliGemma 2, Florence-2, and Qwen2.5-VL."
 authors = [
     {name = "Piotr Skalski", email = "[email protected]"}
@@ -79,16 +79,15 @@ florence_2 = [
 paligemma_2 = [
     "peft>=0.12",
     "torch>=2.4.0",
-    "transformers<4.49.0",
+    "transformers>=4.48.3",
     "bitsandbytes>=0.45.0"
 ]
 qwen_2_5_vl = [
     "accelerate>=1.2.1",
     "peft>=0.12",
     "torch>=2.4.0",
     "torchvision>=0.20.0",
-    # PyPi doesn't allow git repo packages; uncomment when transformers release support for Qwen2.5-VL
-    # "transformers @ git+https://github.com/huggingface/transformers",
+    "transformers>=4.49.0",
     "bitsandbytes>=0.45.0",
     "qwen-vl-utils>=0.0.8"
 ]