datachain 0.6.4__tar.gz → 0.6.6__tar.gz

{datachain-0.6.4 → datachain-0.6.6}/.github/workflows/tests.yml

@@ -152,4 +152,6 @@ jobs:
         run: uv pip install nox --system
 
       - name: Run examples
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
         run: nox -s examples -p ${{ matrix.pyv }} -- -m "${{ matrix.group }}"

{datachain-0.6.4 → datachain-0.6.6}/.pre-commit-config.yaml

@@ -24,7 +24,7 @@ repos:
       - id: trailing-whitespace
         exclude: '^LICENSES/'
   - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: 'v0.7.1'
+    rev: 'v0.7.2'
     hooks:
       - id: ruff
         args: [--fix, --exit-non-zero-on-fix]

{datachain-0.6.4/src/datachain.egg-info → datachain-0.6.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: datachain
-Version: 0.6.4
+Version: 0.6.6
 Summary: Wrangle unstructured AI data at scale
 Author-email: Dmitry Petrov <support@dvc.org>
 License: Apache-2.0

datachain-0.6.6/examples/llm_and_nlp/hf-dataset-llm-eval.py

@@ -0,0 +1,59 @@
+from huggingface_hub import InferenceClient
+
+from datachain import C, DataChain, DataModel
+
+PROMPT = """
+Was this dialog successful? Put result as a single word: Success or Failure.
+Explain the reason in a few words.
+"""
+
+
+class DialogEval(DataModel):
+    result: str
+    reason: str
+
+
+# DataChain function to evaluate dialog.
+# DataChain is using types for inputs, results to automatically infer schema.
+def eval_dialog(user_input: str, bot_response: str) -> DialogEval:
+    client = InferenceClient("meta-llama/Llama-3.1-70B-Instruct")
+
+    completion = client.chat_completion(
+        messages=[
+            {
+                "role": "user",
+                "content": f"{PROMPT}\n\nUser: {user_input}\nBot: {bot_response}",
+            },
+        ],
+        response_format={"type": "json", "value": DialogEval.model_json_schema()},
+    )
+
+    message = completion.choices[0].message
+    try:
+        return DialogEval.model_validate_json(message.content)
+    except ValueError:
+        return DialogEval(result="Error", reason="Failed to parse response.")
+
+
+# Run HF inference in parallel for each example.
+# Get result as Pydantic model that DataChain can understand and serialize it.
+# Save to HF as Parquet. Dataset can be previewed here:
+# https://huggingface.co/datasets/dvcorg/test-datachain-llm-eval/viewer
+(
+    DataChain.from_csv(
+        "hf://datasets/infinite-dataset-hub/MobilePlanAssistant/data.csv"
+    )
+    .settings(parallel=10)
+    .map(response=eval_dialog)
+    .to_parquet("hf://datasets/dvcorg/test-datachain-llm-eval/data.parquet")
+)
+
+# Read it back to filter and show.
+# It restores the Pydantic model from Parquet under the hood.
+(
+    DataChain.from_parquet(
+        "hf://datasets/dvcorg/test-datachain-llm-eval/data.parquet", source=False
+    )
+    .filter(C("response.result") == "Failure")
+    .show(3)
+)

{datachain-0.6.4 → datachain-0.6.6}/overrides/main.html

@@ -4,6 +4,7 @@
 
 {{ super() }}
 
+<script data-domain="docs.datachain.ai" src="https://plausible.io/js/script.outbound-links.js"></script>
 <script type="text/javascript">
   !function () { var e, t, n; e = "14ffd92a6cbf5f2", t = function () { Reo.init({ clientID: "14ffd92a6cbf5f2" }) }, (n = document.createElement("script")).src = "https://static.reo.dev/" + e + "/reo.js", n.async = !0, n.onload = t, document.head.appendChild(n) }();
 </script>

{datachain-0.6.4 → datachain-0.6.6}/src/datachain/client/hf.py

@@ -23,6 +23,7 @@ class HfClient(Client):
 
     def info_to_file(self, v: dict[str, Any], path: str) -> File:
         return File(
+            source=self.uri,
             path=path,
             size=v["size"],
             version=v["last_commit"].oid,

{datachain-0.6.4 → datachain-0.6.6}/src/datachain/lib/arrow.py

@@ -175,7 +175,7 @@ def arrow_type_mapper(col_type: pa.DataType, column: str = "") -> type:  # noqa:
         return dict
     if isinstance(col_type, pa.lib.DictionaryType):
         return arrow_type_mapper(col_type.value_type)  # type: ignore[return-value]
-    raise TypeError(f"{col_type!r} datatypes not supported")
+    raise TypeError(f"{col_type!r} datatypes not supported, column: {column}")
 
 
 def _nrows_file(file: File, nrows: int) -> str:

{datachain-0.6.4 → datachain-0.6.6}/src/datachain/lib/dc.py

@@ -981,10 +981,23 @@ class DataChain:
 
     @resolve_columns
     def order_by(self, *args, descending: bool = False) -> "Self":
-        """Orders by specified set of signals.
+        """Orders by specified set of columns.
 
         Parameters:
             descending (bool): Whether to sort in descending order or not.
+
+        Example:
+            ```py
+            dc.order_by("similarity_score", descending=True).limit(10)
+            ```
+
+        Note:
+            Order is not guaranteed when steps are added after an `order_by` statement.
+            I.e. when using `from_dataset` an `order_by` statement should be used if
+            the order of the records in the chain is important.
+            Using `order_by` directly before `limit`, `collect` and `collect_flatten`
+            will give expected results.
+            See https://github.com/iterative/datachain/issues/477 for further details.
         """
         if descending:
             args = tuple(sqlalchemy.desc(a) for a in args)
@@ -1179,7 +1192,7 @@ class DataChain:
                           a tuple of row values.
         """
         db_signals = self._effective_signals_schema.db_signals()
-        with self._query.select(*db_signals).as_iterable() as rows:
+        with self._query.ordered_select(*db_signals).as_iterable() as rows:
             if row_factory:
                 rows = (row_factory(db_signals, r) for r in rows)
             yield from rows
@@ -1270,7 +1283,7 @@ class DataChain:
         chain = self.select(*cols) if cols else self
         signals_schema = chain._effective_signals_schema
         db_signals = signals_schema.db_signals()
-        with self._query.select(*db_signals).as_iterable() as rows:
+        with self._query.ordered_select(*db_signals).as_iterable() as rows:
             for row in rows:
                 ret = signals_schema.row_to_features(
                     row, catalog=chain.session.catalog, cache=chain._settings.cache
@@ -1678,7 +1691,7 @@ class DataChain:
 
         model_name = model_name or object_name or ""
         hf_features = next(iter(ds_dict.values())).features
-        output = output | get_output_schema(hf_features, model_name)
+        output = output | get_output_schema(hf_features)
         model = dict_to_data_model(model_name, output)
         if object_name:
             output = {object_name: model}

{datachain-0.6.4 → datachain-0.6.6}/src/datachain/lib/hf.py

@@ -138,17 +138,15 @@ def convert_feature(val: Any, feat: Any, anno: Any) -> Any:  # noqa: PLR0911
         return HFAudio(**val)
 
 
-def get_output_schema(
-    features: Features, model_name: str = "", stream: bool = True
-) -> dict[str, DataType]:
+def get_output_schema(features: Features) -> dict[str, DataType]:
     """Generate UDF output schema from huggingface datasets features."""
     fields_dict = {}
     for name, val in features.items():
-        fields_dict[name] = _feature_to_chain_type(name, val)  # type: ignore[assignment]
-    return fields_dict  # type: ignore[return-value]
+        fields_dict[name] = _feature_to_chain_type(name, val)
+    return fields_dict
 
 
-def _feature_to_chain_type(name: str, val: Any) -> type:  # noqa: PLR0911
+def _feature_to_chain_type(name: str, val: Any) -> DataType:  # noqa: PLR0911
     if isinstance(val, Value):
         return arrow_type_mapper(val.pa_type)
     if isinstance(val, ClassLabel):

{datachain-0.6.4 → datachain-0.6.6}/src/datachain/query/dataset.py

@@ -1276,6 +1276,27 @@ class DatasetQuery:
         query.steps.append(SQLSelect((*args, *named_args)))
         return query
 
+    @detach
+    def ordered_select(self, *args, **kwargs) -> "Self":
+        """
+        Select the given columns or expressions using a subquery whilst
+        maintaining query ordering (only applicable if last step was order_by).
+
+        If used with no arguments, this simply creates a subquery and
+        select all columns from it.
+
+        Example:
+            >>> ds.ordered_select(C.name, C.size * 10)
+            >>> ds.ordered_select(C.name, size10x=C.size * 10)
+        """
+        named_args = [v.label(k) for k, v in kwargs.items()]
+        query = self.clone()
+        order_by = query.last_step if query.is_ordered else None
+        query.steps.append(SQLSelect((*args, *named_args)))
+        if order_by:
+            query.steps.append(order_by)
+        return query
+
     @detach
     def select_except(self, *args) -> "Self":
         """
@@ -1338,7 +1359,7 @@ class DatasetQuery:
         query = self.clone(new_table=False)
         if (
             query.steps
-            and (last_step := query.steps[-1])
+            and (last_step := query.last_step)
             and isinstance(last_step, SQLLimit)
         ):
             query.steps[-1] = SQLLimit(min(n, last_step.n))
@@ -1591,3 +1612,11 @@ class DatasetQuery:
         finally:
             self.cleanup()
         return self.__class__(name=name, version=version, catalog=self.catalog)
+
+    @property
+    def is_ordered(self) -> bool:
+        return isinstance(self.last_step, SQLOrderBy)
+
+    @property
+    def last_step(self) -> Optional[Step]:
+        return self.steps[-1] if self.steps else None

{datachain-0.6.4 → datachain-0.6.6}/src/datachain/sql/types.py

@@ -187,6 +187,22 @@ class Int32(Int):
         return read_converter(dialect).int32(value)
 
 
+class UInt32(Int):
+    def load_dialect_impl(self, dialect):
+        return converter(dialect).uint32()
+
+    @staticmethod
+    def default_value(dialect):
+        return type_defaults(dialect).uint32()
+
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).uint32()
+
+    def on_read_convert(self, value, dialect):
+        return read_converter(dialect).uint32(value)
+
+
 class Int64(Int):
     def load_dialect_impl(self, dialect):
         return converter(dialect).int64()
@@ -395,6 +411,9 @@ class TypeReadConverter:
     def int32(self, value):
         return value
 
+    def uint32(self, value):
+        return value
+
     def int64(self, value):
         return value
 
@@ -446,6 +465,9 @@ class TypeConverter:
     def int32(self):
         return self.int()
 
+    def uint32(self):
+        return self.int()
+
     def int64(self):
         return self.int()
 
@@ -487,6 +509,9 @@ class TypeDefaults:
     def int32(self):
         return None
 
+    def uint32(self):
+        return None
+
     def int64(self):
         return None
 
@@ -528,6 +553,9 @@ class DBDefaults:
     def int32(self):
         return self.int()
 
+    def uint32(self):
+        return self.int()
+
     def int64(self):
         return self.int()
 
@@ -561,6 +589,7 @@ TYPES = [
     Boolean,
     Int,
     Int32,
+    UInt32,
     Int64,
     UInt64,
     Float,

{datachain-0.6.4 → datachain-0.6.6/src/datachain.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: datachain
-Version: 0.6.4
+Version: 0.6.6
 Summary: Wrangle unstructured AI data at scale
 Author-email: Dmitry Petrov <support@dvc.org>
 License: Apache-2.0

{datachain-0.6.4 → datachain-0.6.6}/src/datachain.egg-info/SOURCES.txt

@@ -41,6 +41,7 @@ examples/get_started/udfs/parallel.py
 examples/get_started/udfs/simple.py
 examples/get_started/udfs/stateful.py
 examples/llm_and_nlp/claude-query.py
+examples/llm_and_nlp/hf-dataset-llm-eval.py
 examples/llm_and_nlp/unstructured-embeddings-gen.py
 examples/llm_and_nlp/unstructured-summary-map.py
 examples/multimodal/clip_inference.py

{datachain-0.6.4 → datachain-0.6.6}/tests/examples/test_examples.py

@@ -19,8 +19,12 @@ llm_and_nlp_examples = sorted(
     [
         filename
         for filename in glob.glob("examples/llm_and_nlp/**/*.py", recursive=True)
-        # no anthropic token
+        # no anthropic token, HF runs against actual API - thus run it only once
         if "claude" not in filename
+        and (
+            "hf-" not in filename
+            or (sys.platform == "darwin" and sys.version_info >= (3, 12))
+        )
     ]
 )
 

{datachain-0.6.4 → datachain-0.6.6}/tests/func/test_datachain.py

@@ -448,6 +448,30 @@ def test_show_no_truncate(capsys, test_session):
         assert details[i] in normalized_output
 
 
+@pytest.mark.parametrize("ordered_by", ["letter", "number"])
+def test_show_ordered(capsys, test_session, ordered_by):
+    numbers = [6, 2, 3, 1, 5, 7, 4]
+    letters = ["u", "y", "x", "z", "v", "t", "w"]
+
+    DataChain.from_values(
+        number=numbers, letter=letters, session=test_session
+    ).order_by(ordered_by).show()
+
+    captured = capsys.readouterr()
+    normalized_lines = [
+        re.sub(r"\s+", " ", line).strip() for line in captured.out.strip().split("\n")
+    ]
+
+    ordered_entries = sorted(
+        zip(numbers, letters), key=lambda x: x[0 if ordered_by == "number" else 1]
+    )
+
+    assert normalized_lines[0].strip() == "number letter"
+    for i, line in enumerate(normalized_lines[1:]):
+        number, letter = ordered_entries[i]
+        assert line == f"{i} {number} {letter}"
+
+
 def test_from_storage_dataset_stats(tmp_dir, test_session):
     for i in range(4):
         (tmp_dir / f"file{i}.txt").write_text(f"file{i}")

{datachain-0.6.4 → datachain-0.6.6}/tests/func/test_listing.py

@@ -17,7 +17,7 @@ def test_listing_generator(cloud_test_catalog, cloud_type):
     entries = sorted(
         [e for e in ENTRIES if e.path.startswith("cats/")], key=lambda e: e.path
     )
-    files = sorted(dc.collect("file"), key=lambda f: f.path)
+    files = dc.order_by("file.path").collect("file")
 
     for cat_file, cat_entry in zip(files, entries):
         assert cat_file.source == ctc.src_uri

{datachain-0.6.4 → datachain-0.6.6}/tests/unit/lib/test_datachain.py

@@ -1824,6 +1824,32 @@ def test_order_by_with_nested_columns(test_session, with_function):
     ]
 
 
+def test_order_by_collect(test_session):
+    numbers = [6, 2, 3, 1, 5, 7, 4]
+    letters = ["u", "y", "x", "z", "v", "t", "w"]
+
+    dc = DataChain.from_values(number=numbers, letter=letters, session=test_session)
+    assert list(dc.order_by("number").collect()) == [
+        (1, "z"),
+        (2, "y"),
+        (3, "x"),
+        (4, "w"),
+        (5, "v"),
+        (6, "u"),
+        (7, "t"),
+    ]
+
+    assert list(dc.order_by("letter").collect()) == [
+        (7, "t"),
+        (6, "u"),
+        (5, "v"),
+        (4, "w"),
+        (3, "x"),
+        (2, "y"),
+        (1, "z"),
+    ]
+
+
 @pytest.mark.parametrize("with_function", [True, False])
 def test_order_by_descending(test_session, with_function):
     names = ["a.txt", "c.txt", "d.txt", "a.txt", "b.txt"]
@@ -1852,7 +1878,7 @@ def test_union(test_session):
     chain2 = DataChain.from_values(value=[3, 4], session=test_session)
     chain3 = chain1 | chain2
     assert chain3.count() == 4
-    assert sorted(chain3.collect("value")) == [1, 2, 3, 4]
+    assert list(chain3.order_by("value").collect("value")) == [1, 2, 3, 4]
 
 
 def test_union_different_columns(test_session):
@@ -1887,7 +1913,7 @@ def test_union_different_column_order(test_session):
     chain2 = DataChain.from_values(
         name=["different", "order"], value=[9, 10], session=test_session
     )
-    assert sorted(chain1.union(chain2).collect()) == [
+    assert list(chain1.union(chain2).order_by("value").collect()) == [
         (1, "chain"),
         (2, "more"),
         (9, "different"),

{datachain-0.6.4 → datachain-0.6.6}/tests/unit/lib/test_signal_schema.py

@@ -27,6 +27,7 @@ from datachain.sql.types import (
     Int32,
     Int64,
     String,
+    UInt32,
     UInt64,
 )
 
@@ -721,6 +722,7 @@ def test_mutate_change_type():
         [Boolean, bool],
         [Int, int],
         [Int32, int],
+        [UInt32, int],
         [Int64, int],
         [UInt64, int],
         [Float, float],

{datachain-0.6.4 → datachain-0.6.6}/tests/unit/test_data_storage.py

@@ -18,6 +18,7 @@ from datachain.sql.types import (
     Int32,
     Int64,
     String,
+    UInt32,
     UInt64,
 )
 from tests.utils import (
@@ -173,6 +174,7 @@ def test_convert_type(cloud_test_catalog):
         [Boolean(), False],
         [Int(), 0],
         [Int32(), 0],
+        [UInt32(), 0],
         [Int64(), 0],
         [UInt64(), 0],
         [Float(), lambda val: math.isnan(val)],