clarifai 10.1.0__py3-none-any.whl → 10.2.0__py3-none-any.whl

clarifai/datasets/upload/loaders/README.md

@@ -8,15 +8,15 @@ If a dataset module exists in the zoo, uploading the specific dataset can be eas
 
 ```python
 from clarifai.client.app import App
-from clarifai.datasets.upload.loaders.coco_segmentation import COCOSegmentationDataLoader
+from clarifai.datasets.upload.loaders.coco_detection import COCODetectionDataLoader
 
 app = App(app_id="", user_id="")
 # Create a dataset in Clarifai App
 dataset = app.create_dataset(dataset_id="")
 # instantiate dataloader object
-coco_seg_dataloader = COCOSegmentationDataLoader(images_dir="", label_filepath="")
+coco_det_dataloader = COCODetectionDataLoader(images_dir="", label_filepath="")
 # execute data upload to Clarifai app dataset
-dataset.upload_dataset(dataloader=coco_seg_dataloader)
+dataset.upload_dataset(dataloader=coco_det_dataloader)
 ```
 
 ## Dataset Loaders
@@ -24,7 +24,6 @@ dataset.upload_dataset(dataloader=coco_seg_dataloader)
  | dataset name | task | module name (.py)
  | --- | --- | ---
  | [COCO 2017](https://cocodataset.org/#download) | Detection | `coco_detection` |
- |        | Segmentation | `coco_segmentation` |
  |       | Captions | `coco_captions` |
  |[xVIEW](http://xviewdataset.org/)  | Detection | `xview_detection` |
  | [ImageNet](https://www.image-net.org/)  | Classification | `imagenet_classification` |

clarifai/datasets/upload/loaders/xview_detection.py

@@ -6,7 +6,7 @@ from concurrent.futures import ThreadPoolExecutor
 from multiprocessing import cpu_count
 from typing import DefaultDict, Dict, List
 
-import cv2
+from PIL import Image
 from tqdm import tqdm
 
 from clarifai.datasets.upload.base import ClarifaiDataLoader
@@ -54,9 +54,8 @@ class xviewDetectionDataLoader(ClarifaiDataLoader):
   def compress_tiff(self, img_path: str) -> None:
     """Compress tiff image"""
     img_comp_path = os.path.join(self.img_comp_dir, os.path.basename(img_path))
-    img_arr = cv2.imread(img_path)
-    cv2.imwrite(
-        img_comp_path, img_arr, params=(cv2.IMWRITE_TIFF_COMPRESSION, 8))  # 8: Adobe Deflate
+    img_arr = Image.open(img_path)
+    img_arr.save(img_comp_path, 'TIFF', compression='tiff_deflate')
 
   def preprocess(self):
     """Compress the tiff images to comply with clarifai grpc image encoding limit(<20MB) Uses ADOBE_DEFLATE compression algorithm"""
@@ -133,7 +132,8 @@ class xviewDetectionDataLoader(ClarifaiDataLoader):
     _id = os.path.splitext(os.path.basename(self.image_paths[index]))[0]
     image_path = self.image_paths[index]
 
-    image_height, image_width = cv2.imread(image_path).shape[:2]
+    image = Image.open(image_path)
+    image_width, image_height = image.size
     annots = []
     class_names = []
     for bbox, concept in zip(self.all_data[_id]['bboxes'], self.all_data[_id]['concepts']):

clarifai/models/model_serving/cli/_utils.py

@@ -11,7 +11,7 @@ from ..constants import (CLARIFAI_EXAMPLES_REPO, CLARIFAI_EXAMPLES_REPO_PATH,
 def download_examples_repo(forced_download: bool = False):
 
   def _pull():
-    subprocess.run(f"git clone {CLARIFAI_EXAMPLES_REPO} {CLARIFAI_EXAMPLES_REPO_PATH}")
+    subprocess.run(f"git clone {CLARIFAI_EXAMPLES_REPO} {CLARIFAI_EXAMPLES_REPO_PATH}", shell=True)
 
   if not os.path.isdir(CLARIFAI_EXAMPLES_REPO_PATH):
     print(f"Download examples to {CLARIFAI_EXAMPLES_REPO_PATH}")

clarifai/models/model_serving/cli/build.py

@@ -70,7 +70,7 @@ class BuildModelSubCli(BaseClarifaiCli):
     if not self.no_test:
       assert os.path.exists(
           self.test_path), FileNotFoundError(f"Could not find `test.py` in {self.path}")
-      result = subprocess.run(f"pytest -s --log-level=INFO {self.test_path}")
+      result = subprocess.run(f"pytest -s --log-level=INFO {self.test_path}", shell=True)
       assert result.returncode == 0, "Test has failed. Please make sure no error exists in your code."
 
     # build

clarifai/models/model_serving/cli/upload.py

@@ -126,7 +126,7 @@ class UploadModelSubCli(BaseClarifaiCli):
     # Run test before uploading
     if not self.no_test:
       assert os.path.exists(self.test_path), FileNotFoundError(f"Not found {self.test_path}")
-      result = subprocess.run(f"pytest -s --log-level=INFO {self.test_path}")
+      result = subprocess.run(f"pytest -s --log-level=INFO {self.test_path}", shell=True)
       assert result.returncode == 0, "Test has failed. Please make sure no error exists in your code."
 
     deploy(

clarifai/models/model_serving/utils.py

@@ -18,4 +18,6 @@ def _read_pat():
 
 def login(pat=None):
   """ if pat provided, set pat to CLARIFAI_PAT otherwise read pat from file"""
-  os.environ["CLARIFAI_PAT"] = pat or _read_pat()
+  pat = pat or _read_pat()
+  assert pat, Exception("PAT is not found, please run `clarifai login` to persist your PAT")
+  os.environ["CLARIFAI_PAT"] = pat

clarifai/rag/rag.py

@@ -76,16 +76,17 @@ class RAG:
         >>> rag_agent = RAG.setup(app_url=YOUR_APP_URL)
         >>> rag_agent.chat(messages=[{"role":"human", "content":"What is Clarifai"}])
     """
-
+    now_ts = str(int(datetime.now().timestamp()))
     if user_id and not app_url:
       user = User(user_id=user_id, base_url=base_url, pat=pat)
       ## Create an App
-      now_ts = str(int(datetime.now().timestamp()))
       app_id = f"rag_app_{now_ts}"
       app = user.create_app(app_id=app_id, base_workflow=base_workflow)
 
     if not user_id and app_url:
       app = App(url=app_url, pat=pat)
+      uid = app_url.split(".com/")[1].split("/")[0]
+      user = User(user_id=uid, base_url=base_url, pat=pat)
 
     if user_id and app_url:
       raise UserError("Must provide one of user_id or app_url, not both.")
@@ -95,7 +96,7 @@ class RAG:
           "user_id or app_url must be provided. The user_id can be found at https://clarifai.com/settings."
       )
 
-    llm = Model(llm_url)
+    llm = Model(url=llm_url, pat=pat)
 
     min_score = kwargs.get("min_score", 0.95)
     max_results = kwargs.get("max_results", 5)
@@ -109,8 +110,8 @@ class RAG:
     prompter_model_params = {"params": params}
 
     ## Create rag-prompter model and version
-    prompter_model = app.create_model(
-        model_id=f"rag_prompter_{now_ts}", model_type_id="rag-prompter")
+    model_id = f"prompter-{workflow_id}" if workflow_id is not None else f"rag-prompter-{now_ts}"
+    prompter_model = app.create_model(model_id=model_id, model_type_id="rag-prompter")
     prompter_model = prompter_model.create_version(output_info=prompter_model_params)
 
     ## Generate a tmp yaml file for workflow creation
@@ -153,6 +154,8 @@ class RAG:
              batch_size: int = 128,
              chunk_size: int = 1024,
              chunk_overlap: int = 200,
+             dataset_id: str = None,
+             metadata: dict = None,
              **kwargs) -> None:
     """Uploads documents to the app.
         - Read from a local directory or public url or local filename.
@@ -192,14 +195,15 @@ class RAG:
 
     #splitting documents into chunks
     text_chunks = []
-    metadata = []
+    metadata_list = []
 
     #iterate through documents
     for doc in documents:
+      doc_i = 0
       cur_text_chunks = split_document(
           text=doc.text, chunk_size=chunk_size, chunk_overlap=chunk_overlap, **kwargs)
       text_chunks.extend(cur_text_chunks)
-      metadata.extend([doc.metadata for _ in range(len(cur_text_chunks))])
+      metadata_list.extend([doc.metadata for _ in range(len(cur_text_chunks))])
       #if batch size is reached, upload the batch
       if len(text_chunks) > batch_size:
         for idx in range(0, len(text_chunks), batch_size):
@@ -208,18 +212,23 @@ class RAG:
           batch_texts = text_chunks[0:batch_size]
           batch_ids = [uuid.uuid4().hex for _ in range(batch_size)]
           #metadata
-          batch_metadatas = metadata[0:batch_size]
+          batch_metadatas = metadata_list[0:batch_size]
           meta_list = []
           for meta in batch_metadatas:
             meta_struct = Struct()
             meta_struct.update(meta)
+            meta_struct.update({"doc_chunk_no": doc_i})
+            if metadata and isinstance(metadata, dict):
+              meta_struct.update(metadata)
             meta_list.append(meta_struct)
+            doc_i += 1
           del batch_metadatas
           #creating input proto
           input_batch = [
               self._app.inputs().get_text_input(
                   input_id=batch_ids[i],
                   raw_text=text,
+                  dataset_id=dataset_id,
                   metadata=meta_list[i],
               ) for i, text in enumerate(batch_texts)
           ]
@@ -227,32 +236,37 @@ class RAG:
           self._app.inputs().upload_inputs(inputs=input_batch)
           #delete uploaded chunks
           del text_chunks[0:batch_size]
-          del metadata[0:batch_size]
+          del metadata_list[0:batch_size]
 
     #uploading the remaining chunks
     if len(text_chunks) > 0:
       batch_size = len(text_chunks)
       batch_ids = [uuid.uuid4().hex for _ in range(batch_size)]
       #metadata
-      batch_metadatas = metadata[0:batch_size]
+      batch_metadatas = metadata_list[0:batch_size]
       meta_list = []
       for meta in batch_metadatas:
         meta_struct = Struct()
         meta_struct.update(meta)
+        meta_struct.update({"doc_chunk_no": doc_i})
+        if metadata and isinstance(metadata, dict):
+          meta_struct.update(metadata)
         meta_list.append(meta_struct)
+        doc_i += 1
       del batch_metadatas
       #creating input proto
       input_batch = [
           self._app.inputs().get_text_input(
               input_id=batch_ids[i],
               raw_text=text,
+              dataset_id=dataset_id,
               metadata=meta_list[i],
           ) for i, text in enumerate(text_chunks)
       ]
       #uploading input with metadata
       self._app.inputs().upload_inputs(inputs=input_batch)
       del text_chunks
-      del metadata
+      del metadata_list
 
   def chat(self, messages: List[dict], client_manage_state: bool = False) -> List[dict]:
     """Chat interface in OpenAI API format.

clarifai/rag/utils.py

@@ -3,10 +3,6 @@ from pathlib import Path
 from typing import List
 
 import requests
-from llama_index.core import Document, SimpleDirectoryReader
-from llama_index.core.node_parser.text import SentenceSplitter
-from llama_index.core.readers.download import download_loader
-from pypdf import PdfReader
 
 
 ## TODO: Make this token-aware.
@@ -36,8 +32,7 @@ def format_assistant_message(raw_text: str) -> dict:
   return {"role": "assistant", "content": raw_text}
 
 
-def load_documents(file_path: str = None, folder_path: str = None,
-                   url: str = None) -> List[Document]:
+def load_documents(file_path: str = None, folder_path: str = None, url: str = None) -> List[any]:
   """Loads documents from a local directory or public url or local filename.
 
   Args:
@@ -45,6 +40,13 @@ def load_documents(file_path: str = None, folder_path: str = None,
       folder_path (str): The path to the folder.
       url (str): The url to the file.
   """
+  #check import packages
+  try:
+    from llama_index.core import Document, SimpleDirectoryReader
+    from llama_index.core.readers.download import download_loader
+  except ImportError:
+    raise ImportError("Could not import llama index package. "
+                      "Please install it with `pip install llama-index-core==0.10.1`.")
   #document loaders for filepath
   if file_path:
     if file_path.endswith(".pdf"):
@@ -77,6 +79,12 @@ def load_documents(file_path: str = None, folder_path: str = None,
       documents = [Document(text=response.content)]
     #for pdf files
     except Exception:
+      #check import packages
+      try:
+        from pypdf import PdfReader
+      except ImportError:
+        raise ImportError("Could not import pypdf package. "
+                          "Please install it with `pip install pypdf==3.17.4`.")
       documents = []
       pdf_file = PdfReader(io.BytesIO(response.content))
       num_pages = len(pdf_file.pages)
@@ -98,6 +106,13 @@ def split_document(text: str, chunk_size: int, chunk_overlap: int, **kwargs) ->
       chunk_overlap (int): The amount of overlap between each chunk.
       **kwargs: Additional keyword arguments for the SentenceSplitter.
   """
+  #check import packages
+  try:
+    from llama_index.core.node_parser.text import SentenceSplitter
+  except ImportError:
+    raise ImportError("Could not import llama index package. "
+                      "Please install it with `pip install llama-index-core==0.10.1`.")
+  #document
   text_parser = SentenceSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap, **kwargs)
   text_chunks = text_parser.split_text(text)
   return text_chunks