diff --git a/src/banks/env.py b/src/banks/env.py index 247dd34..7d495a8 100644 --- a/src/banks/env.py +++ b/src/banks/env.py @@ -4,7 +4,7 @@ from jinja2 import Environment, PackageLoader, select_autoescape from .config import config -from .filters import cache_control, lemmatize, tool +from .filters import cache_control, image, lemmatize, tool def _add_extensions(_env): @@ -38,7 +38,9 @@ def _add_extensions(_env): # Setup custom filters and defaults -env.filters["lemmatize"] = lemmatize env.filters["cache_control"] = cache_control +env.filters["image"] = image +env.filters["lemmatize"] = lemmatize env.filters["tool"] = tool + _add_extensions(env) diff --git a/src/banks/extensions/chat.py b/src/banks/extensions/chat.py index ee75fa2..db6cb15 100644 --- a/src/banks/extensions/chat.py +++ b/src/banks/extensions/chat.py @@ -33,12 +33,12 @@ def content(self) -> ChatMessageContent: return self._content_blocks - def handle_starttag(self, tag, _): - if tag == "content_block_txt": + def handle_starttag(self, tag, attrs): # noqa + if tag == "content_block": self._parse_block_content = True def handle_endtag(self, tag): - if tag == "content_block_txt": + if tag == "content_block": self._parse_block_content = False def handle_data(self, data): diff --git a/src/banks/filters/__init__.py b/src/banks/filters/__init__.py index 27645ad..60ad899 100644 --- a/src/banks/filters/__init__.py +++ b/src/banks/filters/__init__.py @@ -2,11 +2,13 @@ # # SPDX-License-Identifier: MIT from .cache_control import cache_control +from .image import image from .lemmatize import lemmatize from .tool import tool __all__ = ( "cache_control", + "image", "lemmatize", "tool", ) diff --git a/src/banks/filters/cache_control.py b/src/banks/filters/cache_control.py index 65f4640..0e0284c 100644 --- a/src/banks/filters/cache_control.py +++ b/src/banks/filters/cache_control.py @@ -17,8 +17,8 @@ def cache_control(value: str, cache_type: str = "ephemeral") -> str: ``` Important: - this filter marks the content to cache by surrounding it with `` and - ``, so it's only useful when used within a `{% chat %}` block. + this filter marks the content to cache by surrounding it with `` and + ``, so it's only useful when used within a `{% chat %}` block. """ block = ContentBlock.model_validate({"type": "text", "text": value, "cache_control": {"type": cache_type}}) - return f"{block.model_dump_json()}" + return f"{block.model_dump_json()}" diff --git a/src/banks/filters/image.py b/src/banks/filters/image.py new file mode 100644 index 0000000..81ea7cd --- /dev/null +++ b/src/banks/filters/image.py @@ -0,0 +1,40 @@ +# SPDX-FileCopyrightText: 2023-present Massimiliano Pippi +# +# SPDX-License-Identifier: MIT +from pathlib import Path +from urllib.parse import urlparse + +from banks.types import ContentBlock, ImageUrl + + +def _is_url(string: str) -> bool: + try: + result = urlparse(string) + return all([result.scheme, result.netloc]) + except ValueError: + return False + + +def image(value: str) -> str: + """Wrap the filtered value into a ContentBlock with the proper cache_control field set. + + The resulting ChatMessage will have the field `content` populated with a list of ContentBlock objects. + + Example: + ```jinja + Describe what you see + + {{ "path/to/image/file" | image }} + ``` + + Important: + this filter marks the content to cache by surrounding it with `` and + ``, so it's only useful when used within a `{% chat %}` block. + """ + if _is_url(value): + image_url = ImageUrl(url=value) + else: + image_url = ImageUrl.from_path(Path(value)) + + block = ContentBlock.model_validate({"type": "image_url", "image_url": image_url}) + return f"{block.model_dump_json()}" diff --git a/src/banks/types.py b/src/banks/types.py index fb011b9..7650034 100644 --- a/src/banks/types.py +++ b/src/banks/types.py @@ -1,8 +1,10 @@ # SPDX-FileCopyrightText: 2023-present Massimiliano Pippi # # SPDX-License-Identifier: MIT +import base64 from enum import Enum from inspect import Parameter, getdoc, signature +from pathlib import Path from typing import Callable from pydantic import BaseModel @@ -26,9 +28,14 @@ class ImageUrl(BaseModel): url: str @classmethod - def from_base64(cls, media_type: str, base64_str: str): + def from_base64(cls, media_type: str, base64_str: str) -> Self: return cls(url=f"data:{media_type};base64,{base64_str}") + @classmethod + def from_path(cls, file_path: Path) -> Self: + with open(file_path, "rb") as image_file: + return cls.from_base64("image/jpeg", base64.b64encode(image_file.read()).decode("utf-8")) + class ContentBlock(BaseModel): type: ContentBlockType diff --git a/tests/test_cache_control.py b/tests/test_cache_control.py index 100153b..7be8b3b 100644 --- a/tests/test_cache_control.py +++ b/tests/test_cache_control.py @@ -3,6 +3,6 @@ def test_cache_control(): res = cache_control("foo", "ephemeral") - res = res.replace("", "") - res = res.replace("", "") + res = res.replace("", "") + res = res.replace("", "") assert res == '{"type":"text","cache_control":{"type":"ephemeral"},"text":"foo","image_url":null}' diff --git a/tests/test_chat.py b/tests/test_chat.py index 71e3be1..38a90c0 100644 --- a/tests/test_chat.py +++ b/tests/test_chat.py @@ -30,7 +30,7 @@ def test_content_block_parser_init(): def test_content_block_parser_single_with_cache_control(): p = _ContentBlockParser() p.feed( - '{"type":"text","cache_control":{"type":"ephemeral"},"text":"foo","source":null}' + '{"type":"text","cache_control":{"type":"ephemeral"},"text":"foo","source":null}' ) assert p.content == [ ContentBlock(type=ContentBlockType.text, cache_control=CacheControl(type="ephemeral"), text="foo", source=None) @@ -39,15 +39,15 @@ def test_content_block_parser_single_with_cache_control(): def test_content_block_parser_single_no_cache_control(): p = _ContentBlockParser() - p.feed('{"type":"text","cache_control":null,"text":"foo","source":null}') + p.feed('{"type":"text","cache_control":null,"text":"foo","source":null}') assert p.content == "foo" def test_content_block_parser_multiple(): p = _ContentBlockParser() p.feed( - '{"type":"text","cache_control":null,"text":"foo","source":null}' - '{"type":"text","cache_control":null,"text":"bar","source":null}' + '{"type":"text","cache_control":null,"text":"foo","source":null}' + '{"type":"text","cache_control":null,"text":"bar","source":null}' ) assert p.content == [ ContentBlock(type=ContentBlockType.text, cache_control=None, text="foo", source=None),