mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-06 02:48:13 +00:00
feat(bedrock): S3 URL pass-through for multimodal content blocks
This commit is contained in:
Ryan H
parent
d8e4fc4dd0
commit
b8aa8dea1a
9 changed files with 918 additions and 92 deletions
|
|
@ -3,6 +3,7 @@ import copy
|
|||
import hashlib
|
||||
import json
|
||||
import mimetypes
|
||||
import os
|
||||
import re
|
||||
import xml.etree.ElementTree as ET
|
||||
from enum import Enum
|
||||
|
|
@ -33,6 +34,7 @@ from litellm.types.llms.openai import (
|
|||
ChatCompletionToolCallFunctionChunk,
|
||||
ChatCompletionToolMessage,
|
||||
ChatCompletionUserMessage,
|
||||
ChatCompletionVideoObject,
|
||||
OpenAIMessageContentListBlock,
|
||||
)
|
||||
from litellm.types.llms.vertex_ai import FunctionCall as VertexFunctionCall
|
||||
|
|
@ -3468,6 +3470,7 @@ from litellm.types.llms.bedrock import (
|
|||
from litellm.types.llms.bedrock import ContentBlock as BedrockContentBlock
|
||||
from litellm.types.llms.bedrock import DocumentBlock as BedrockDocumentBlock
|
||||
from litellm.types.llms.bedrock import ImageBlock as BedrockImageBlock
|
||||
from litellm.types.llms.bedrock import S3Location as BedrockS3Location
|
||||
from litellm.types.llms.bedrock import SourceBlock as BedrockSourceBlock
|
||||
from litellm.types.llms.bedrock import ToolBlock as BedrockToolBlock
|
||||
from litellm.types.llms.bedrock import (
|
||||
|
|
@ -3481,6 +3484,7 @@ from litellm.types.llms.bedrock import (
|
|||
from litellm.types.llms.bedrock import ToolSpecBlock as BedrockToolSpecBlock
|
||||
from litellm.types.llms.bedrock import ToolUseBlock as BedrockToolUseBlock
|
||||
from litellm.types.llms.bedrock import VideoBlock as BedrockVideoBlock
|
||||
from litellm.utils import supports_s3_input
|
||||
|
||||
|
||||
def _parse_content_type(content_type: str) -> str:
|
||||
|
|
@ -3498,7 +3502,14 @@ def _parse_mime_type(base64_data: str) -> Optional[str]:
|
|||
|
||||
|
||||
class BedrockImageProcessor:
|
||||
"""Handles both sync and async image processing for Bedrock conversations."""
|
||||
"""Handles both sync and async image/media processing for Bedrock conversations."""
|
||||
|
||||
@staticmethod
|
||||
def _extract_video_url(element: Union[dict, "ChatCompletionVideoObject"]) -> str:
|
||||
"""Extract the URL string from a video_url content element."""
|
||||
if isinstance(element["video_url"], dict):
|
||||
return element["video_url"]["url"]
|
||||
return element["video_url"]
|
||||
|
||||
@staticmethod
|
||||
def _post_call_image_processing(
|
||||
|
|
@ -3717,20 +3728,98 @@ class BedrockImageProcessor:
|
|||
image=BedrockImageBlock(source=_blob, format=image_format)
|
||||
)
|
||||
|
||||
@staticmethod
|
||||
def _get_bedrock_format_from_extension(extension: str) -> str:
|
||||
"""Map file extension to Bedrock format enum value."""
|
||||
# 3gpp is an alias for 3gp
|
||||
if extension == "3gpp":
|
||||
return "3gp"
|
||||
# jpg → jpeg
|
||||
if extension == "jpg":
|
||||
return "jpeg"
|
||||
return extension
|
||||
|
||||
@classmethod
|
||||
def _create_s3_bedrock_block(
|
||||
cls, s3_url: str, format: Optional[str] = None
|
||||
) -> BedrockContentBlock:
|
||||
"""
|
||||
Create a Bedrock content block referencing an S3 object.
|
||||
Determines block type (image/document/video) from file extension,
|
||||
or from the explicit ``format`` override if provided.
|
||||
"""
|
||||
# If explicit format override provided (e.g. from file.format field),
|
||||
# use it directly instead of requiring a file extension.
|
||||
if format:
|
||||
raw_format = format.split("/")[-1] if "/" in format else format
|
||||
bedrock_format = cls._get_bedrock_format_from_extension(raw_format.lower())
|
||||
else:
|
||||
extension = os.path.splitext(s3_url)[-1].lstrip(".").lower()
|
||||
if not extension:
|
||||
raise ValueError(
|
||||
f"Cannot determine file type from S3 URL (no extension): {s3_url}"
|
||||
)
|
||||
bedrock_format = cls._get_bedrock_format_from_extension(extension)
|
||||
|
||||
s3_source = BedrockSourceBlock(s3Location=BedrockS3Location(uri=s3_url))
|
||||
|
||||
config = litellm.AmazonConverseConfig()
|
||||
supported_image_formats = config.get_supported_image_types()
|
||||
supported_doc_formats = config.get_supported_document_types()
|
||||
supported_video_formats = config.get_supported_video_types()
|
||||
|
||||
if bedrock_format in supported_image_formats:
|
||||
return BedrockContentBlock(
|
||||
image=BedrockImageBlock(source=s3_source, format=bedrock_format)
|
||||
)
|
||||
elif bedrock_format in supported_doc_formats:
|
||||
doc_name = f"s3doc_{hashlib.sha256(s3_url.encode()).hexdigest()[:16]}_{bedrock_format}"
|
||||
return BedrockContentBlock(
|
||||
document=BedrockDocumentBlock(
|
||||
source=s3_source, format=bedrock_format, name=doc_name
|
||||
)
|
||||
)
|
||||
elif bedrock_format in supported_video_formats:
|
||||
return BedrockContentBlock(
|
||||
video=BedrockVideoBlock(source=s3_source, format=bedrock_format)
|
||||
)
|
||||
else:
|
||||
raise ValueError(
|
||||
f"Unsupported file format '{bedrock_format}' for Bedrock S3 content. "
|
||||
f"Supported: images={supported_image_formats}, "
|
||||
f"documents={supported_doc_formats}, "
|
||||
f"videos={supported_video_formats}"
|
||||
)
|
||||
|
||||
# TODO: Rename to process_media_sync/async — these methods now handle images, documents, and videos (not just images).
|
||||
@classmethod
|
||||
def process_image_sync(
|
||||
cls, image_url: str, format: Optional[str] = None
|
||||
cls,
|
||||
image_url: str,
|
||||
format: Optional[str] = None,
|
||||
model: Optional[str] = None,
|
||||
custom_llm_provider: Optional[str] = None,
|
||||
) -> BedrockContentBlock:
|
||||
"""Synchronous image processing."""
|
||||
"""Synchronous processing of media URLs (images, documents, videos) for Bedrock."""
|
||||
|
||||
if "base64" in image_url:
|
||||
if image_url.startswith("s3://"):
|
||||
if model and not supports_s3_input(
|
||||
model=model, custom_llm_provider=custom_llm_provider
|
||||
):
|
||||
raise ValueError(
|
||||
f"Model '{model}' does not support s3:// URLs. "
|
||||
"Only Amazon Nova models (with vision) support S3 input via Bedrock Converse API. "
|
||||
"Please use a base64-encoded or https:// URL instead."
|
||||
)
|
||||
return cls._create_s3_bedrock_block(image_url, format)
|
||||
elif "base64" in image_url:
|
||||
img_bytes, mime_type, image_format = cls._parse_base64_image(image_url)
|
||||
elif "http://" in image_url or "https://" in image_url:
|
||||
img_bytes, mime_type = BedrockImageProcessor.get_image_details(image_url)
|
||||
image_format = mime_type.split("/")[1]
|
||||
else:
|
||||
raise ValueError(
|
||||
"Unsupported image type. Expected either image url or base64 encoded string"
|
||||
"Unsupported image type. Expected either image url, base64 encoded string, or s3:// URL"
|
||||
)
|
||||
|
||||
if format:
|
||||
|
|
@ -3742,11 +3831,25 @@ class BedrockImageProcessor:
|
|||
|
||||
@classmethod
|
||||
async def process_image_async(
|
||||
cls, image_url: str, format: Optional[str]
|
||||
cls,
|
||||
image_url: str,
|
||||
format: Optional[str],
|
||||
model: Optional[str] = None,
|
||||
custom_llm_provider: Optional[str] = None,
|
||||
) -> BedrockContentBlock:
|
||||
"""Asynchronous image processing."""
|
||||
"""Asynchronous processing of media URLs (images, documents, videos) for Bedrock."""
|
||||
|
||||
if "base64" in image_url:
|
||||
if image_url.startswith("s3://"):
|
||||
if model and not supports_s3_input(
|
||||
model=model, custom_llm_provider=custom_llm_provider
|
||||
):
|
||||
raise ValueError(
|
||||
f"Model '{model}' does not support s3:// URLs. "
|
||||
"Only Amazon Nova models (with vision) support S3 input via Bedrock Converse API. "
|
||||
"Please use a base64-encoded or https:// URL instead."
|
||||
)
|
||||
return cls._create_s3_bedrock_block(image_url, format)
|
||||
elif "base64" in image_url:
|
||||
img_bytes, mime_type, image_format = cls._parse_base64_image(image_url)
|
||||
elif "http://" in image_url or "https://" in image_url:
|
||||
img_bytes, mime_type = await BedrockImageProcessor.get_image_details_async(
|
||||
|
|
@ -3755,7 +3858,7 @@ class BedrockImageProcessor:
|
|||
image_format = mime_type.split("/")[1]
|
||||
else:
|
||||
raise ValueError(
|
||||
"Unsupported image type. Expected either image url or base64 encoded string"
|
||||
"Unsupported image type. Expected either image url, base64 encoded string, or s3:// URL"
|
||||
)
|
||||
|
||||
if format: # override with user-defined params
|
||||
|
|
@ -4396,14 +4499,30 @@ class BedrockConverseMessagesProcessor:
|
|||
else:
|
||||
image_url = element["image_url"]
|
||||
_part = await BedrockImageProcessor.process_image_async( # type: ignore
|
||||
image_url=image_url, format=format
|
||||
image_url=image_url,
|
||||
format=format,
|
||||
model=model,
|
||||
custom_llm_provider=llm_provider,
|
||||
)
|
||||
_parts.append(_part) # type: ignore
|
||||
elif element["type"] == "file":
|
||||
_part = await BedrockConverseMessagesProcessor._async_process_file_message(
|
||||
message=cast(ChatCompletionFileObject, element)
|
||||
message=cast(ChatCompletionFileObject, element),
|
||||
model=model,
|
||||
llm_provider=llm_provider,
|
||||
)
|
||||
_parts.append(_part)
|
||||
elif element["type"] == "video_url":
|
||||
video_url = BedrockImageProcessor._extract_video_url(
|
||||
element
|
||||
)
|
||||
_part = await BedrockImageProcessor.process_image_async( # type: ignore
|
||||
image_url=video_url,
|
||||
format=None,
|
||||
model=model,
|
||||
custom_llm_provider=llm_provider,
|
||||
)
|
||||
_parts.append(_part) # type: ignore
|
||||
_cache_point_block = (
|
||||
litellm.AmazonConverseConfig()._get_cache_point_block(
|
||||
message_block=cast(
|
||||
|
|
@ -4645,7 +4764,11 @@ class BedrockConverseMessagesProcessor:
|
|||
return reasoning_content_blocks
|
||||
|
||||
@staticmethod
|
||||
def _process_file_message(message: ChatCompletionFileObject) -> BedrockContentBlock:
|
||||
def _process_file_message(
|
||||
message: ChatCompletionFileObject,
|
||||
model: Optional[str] = None,
|
||||
llm_provider: Optional[str] = None,
|
||||
) -> BedrockContentBlock:
|
||||
file_message = message["file"]
|
||||
file_data = file_message.get("file_data")
|
||||
file_id = file_message.get("file_id")
|
||||
|
|
@ -4660,12 +4783,17 @@ class BedrockConverseMessagesProcessor:
|
|||
)
|
||||
format = file_message.get("format")
|
||||
return BedrockImageProcessor.process_image_sync(
|
||||
image_url=cast(str, file_id or file_data), format=format
|
||||
image_url=cast(str, file_id or file_data),
|
||||
format=format,
|
||||
model=model,
|
||||
custom_llm_provider=llm_provider,
|
||||
)
|
||||
|
||||
@staticmethod
|
||||
async def _async_process_file_message(
|
||||
message: ChatCompletionFileObject,
|
||||
model: Optional[str] = None,
|
||||
llm_provider: Optional[str] = None,
|
||||
) -> BedrockContentBlock:
|
||||
file_message = message["file"]
|
||||
file_data = file_message.get("file_data")
|
||||
|
|
@ -4680,7 +4808,10 @@ class BedrockConverseMessagesProcessor:
|
|||
llm_provider="bedrock",
|
||||
)
|
||||
return await BedrockImageProcessor.process_image_async(
|
||||
image_url=cast(str, file_id or file_data), format=format
|
||||
image_url=cast(str, file_id or file_data),
|
||||
format=format,
|
||||
model=model,
|
||||
custom_llm_provider=llm_provider,
|
||||
)
|
||||
|
||||
@staticmethod
|
||||
|
|
@ -4771,15 +4902,30 @@ def _bedrock_converse_messages_pt( # noqa: PLR0915
|
|||
_part = BedrockImageProcessor.process_image_sync( # type: ignore
|
||||
image_url=image_url,
|
||||
format=format,
|
||||
model=model,
|
||||
custom_llm_provider=llm_provider,
|
||||
)
|
||||
_parts.append(_part) # type: ignore
|
||||
elif element["type"] == "file":
|
||||
_part = (
|
||||
BedrockConverseMessagesProcessor._process_file_message(
|
||||
message=cast(ChatCompletionFileObject, element)
|
||||
message=cast(ChatCompletionFileObject, element),
|
||||
model=model,
|
||||
llm_provider=llm_provider,
|
||||
)
|
||||
)
|
||||
_parts.append(_part)
|
||||
elif element["type"] == "video_url":
|
||||
video_url = BedrockImageProcessor._extract_video_url(
|
||||
element
|
||||
)
|
||||
_part = BedrockImageProcessor.process_image_sync( # type: ignore
|
||||
image_url=video_url,
|
||||
format=None,
|
||||
model=model,
|
||||
custom_llm_provider=llm_provider,
|
||||
)
|
||||
_parts.append(_part) # type: ignore
|
||||
_cache_point_block = (
|
||||
litellm.AmazonConverseConfig()._get_cache_point_block(
|
||||
message_block=cast(
|
||||
|
|
|
|||
|
|
@ -335,7 +335,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"amazon.nova-2-lite-v1:0": {
|
||||
"cache_read_input_token_cost": 7.5e-08,
|
||||
|
|
@ -352,7 +353,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"amazon.nova-2-pro-preview-20251202-v1:0": {
|
||||
"cache_read_input_token_cost": 5.46875e-07,
|
||||
|
|
@ -371,7 +373,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"apac.amazon.nova-2-lite-v1:0": {
|
||||
"cache_read_input_token_cost": 8.25e-08,
|
||||
|
|
@ -388,7 +391,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"apac.amazon.nova-2-pro-preview-20251202-v1:0": {
|
||||
"cache_read_input_token_cost": 5.46875e-07,
|
||||
|
|
@ -407,7 +411,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"eu.amazon.nova-2-lite-v1:0": {
|
||||
"cache_read_input_token_cost": 8.25e-08,
|
||||
|
|
@ -424,7 +429,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"eu.amazon.nova-2-pro-preview-20251202-v1:0": {
|
||||
"cache_read_input_token_cost": 5.46875e-07,
|
||||
|
|
@ -443,7 +449,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"us.amazon.nova-2-lite-v1:0": {
|
||||
"cache_read_input_token_cost": 8.25e-08,
|
||||
|
|
@ -460,7 +467,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"us.amazon.nova-2-pro-preview-20251202-v1:0": {
|
||||
"cache_read_input_token_cost": 5.46875e-07,
|
||||
|
|
@ -479,7 +487,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"amazon.nova-2-multimodal-embeddings-v1:0": {
|
||||
"litellm_provider": "bedrock",
|
||||
|
|
@ -522,7 +531,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"amazon.rerank-v1:0": {
|
||||
"input_cost_per_query": 0.001,
|
||||
|
|
@ -1478,7 +1488,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"apac.amazon.nova-micro-v1:0": {
|
||||
"input_cost_per_token": 3.7e-08,
|
||||
|
|
@ -1504,7 +1515,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"apac.anthropic.claude-3-5-sonnet-20240620-v1:0": {
|
||||
"input_cost_per_token": 3e-06,
|
||||
|
|
@ -7582,7 +7594,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"bedrock/us-gov-east-1/amazon.titan-embed-text-v1": {
|
||||
"input_cost_per_token": 1e-07,
|
||||
|
|
@ -7713,7 +7726,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"bedrock/us-gov-west-1/amazon.titan-embed-text-v1": {
|
||||
"input_cost_per_token": 1e-07,
|
||||
|
|
@ -12115,7 +12129,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"eu.amazon.nova-micro-v1:0": {
|
||||
"input_cost_per_token": 4.6e-08,
|
||||
|
|
@ -12142,7 +12157,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"eu.anthropic.claude-3-5-haiku-20241022-v1:0": {
|
||||
"input_cost_per_token": 2.5e-07,
|
||||
|
|
@ -14599,17 +14615,14 @@
|
|||
"uses_embed_content": true
|
||||
},
|
||||
"vertex_ai/gemini-embedding-2-preview": {
|
||||
"input_cost_per_audio_per_second": 0.00016,
|
||||
"input_cost_per_image": 0.00012,
|
||||
"input_cost_per_token": 2e-07,
|
||||
"input_cost_per_video_per_second": 0.00079,
|
||||
"input_cost_per_token": 1.5e-07,
|
||||
"litellm_provider": "vertex_ai",
|
||||
"max_input_tokens": 8192,
|
||||
"max_tokens": 8192,
|
||||
"mode": "embedding",
|
||||
"output_cost_per_token": 0,
|
||||
"output_vector_size": 3072,
|
||||
"source": "https://cloud.google.com/vertex-ai/generative-ai/pricing",
|
||||
"source": "https://ai.google.dev/gemini-api/docs/embeddings#multimodal",
|
||||
"supports_multimodal": true,
|
||||
"uses_embed_content": true
|
||||
},
|
||||
|
|
@ -14626,18 +14639,6 @@
|
|||
"source": "https://cloud.google.com/vertex-ai/generative-ai/pricing",
|
||||
"uses_embed_content": true
|
||||
},
|
||||
"vertex_ai/gemini-embedding-2-preview": {
|
||||
"input_cost_per_token": 1.5e-07,
|
||||
"litellm_provider": "vertex_ai",
|
||||
"max_input_tokens": 8192,
|
||||
"max_tokens": 8192,
|
||||
"mode": "embedding",
|
||||
"output_cost_per_token": 0,
|
||||
"output_vector_size": 3072,
|
||||
"source": "https://ai.google.dev/gemini-api/docs/embeddings#multimodal",
|
||||
"supports_multimodal": true,
|
||||
"uses_embed_content": true
|
||||
},
|
||||
"gemini/gemini-embedding-001": {
|
||||
"input_cost_per_token": 1.5e-07,
|
||||
"litellm_provider": "gemini",
|
||||
|
|
@ -16834,7 +16835,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"gpt-3.5-turbo": {
|
||||
"input_cost_per_token": 5e-07,
|
||||
|
|
@ -28050,7 +28052,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"us.amazon.nova-micro-v1:0": {
|
||||
"input_cost_per_token": 3.5e-08,
|
||||
|
|
@ -28076,7 +28079,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": false,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"us.amazon.nova-pro-v1:0": {
|
||||
"input_cost_per_token": 8e-07,
|
||||
|
|
@ -28090,7 +28094,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"us.anthropic.claude-3-5-haiku-20241022-v1:0": {
|
||||
"cache_creation_input_token_cost": 1e-06,
|
||||
|
|
@ -31013,7 +31018,9 @@
|
|||
"mode": "chat",
|
||||
"output_cost_per_token": 3.2e-06,
|
||||
"source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#glm-models",
|
||||
"supported_regions": ["global"],
|
||||
"supported_regions": [
|
||||
"global"
|
||||
],
|
||||
"supports_function_calling": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_reasoning": true,
|
||||
|
|
|
|||
|
|
@ -16,8 +16,14 @@ class SystemContentBlock(TypedDict, total=False):
|
|||
cachePoint: CachePointBlock
|
||||
|
||||
|
||||
class SourceBlock(TypedDict):
|
||||
class S3Location(TypedDict, total=False):
|
||||
uri: Required[str] # s3://bucket/key
|
||||
|
||||
|
||||
class SourceBlock(TypedDict, total=False):
|
||||
# Mutually exclusive: provide either `bytes` (inline base64) or `s3Location` (S3 reference), not both.
|
||||
bytes: Optional[str] # base 64 encoded string
|
||||
s3Location: S3Location # S3 reference for large content
|
||||
|
||||
|
||||
BedrockImageTypes = Literal["png", "jpeg", "gif", "webp"]
|
||||
|
|
|
|||
|
|
@ -131,6 +131,7 @@ class ProviderSpecificModelInfo(TypedDict, total=False):
|
|||
supports_embedding_image_input: Optional[bool]
|
||||
supports_audio_output: Optional[bool]
|
||||
supports_pdf_input: Optional[bool]
|
||||
supports_s3_input: Optional[bool]
|
||||
supports_native_streaming: Optional[bool]
|
||||
supports_parallel_function_calling: Optional[bool]
|
||||
supports_web_search: Optional[bool]
|
||||
|
|
|
|||
|
|
@ -2590,6 +2590,13 @@ def supports_pdf_input(model: str, custom_llm_provider: Optional[str] = None) ->
|
|||
)
|
||||
|
||||
|
||||
def supports_s3_input(model: str, custom_llm_provider: Optional[str] = None) -> bool:
|
||||
"""Check if a given model supports s3:// URLs as content source in Bedrock Converse API"""
|
||||
return _supports_factory(
|
||||
model=model, custom_llm_provider=custom_llm_provider, key="supports_s3_input"
|
||||
)
|
||||
|
||||
|
||||
def supports_audio_output(
|
||||
model: str, custom_llm_provider: Optional[str] = None
|
||||
) -> bool:
|
||||
|
|
|
|||
|
|
@ -335,7 +335,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"amazon.nova-2-lite-v1:0": {
|
||||
"cache_read_input_token_cost": 7.5e-08,
|
||||
|
|
@ -352,7 +353,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"amazon.nova-2-pro-preview-20251202-v1:0": {
|
||||
"cache_read_input_token_cost": 5.46875e-07,
|
||||
|
|
@ -371,7 +373,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"apac.amazon.nova-2-lite-v1:0": {
|
||||
"cache_read_input_token_cost": 8.25e-08,
|
||||
|
|
@ -388,7 +391,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"apac.amazon.nova-2-pro-preview-20251202-v1:0": {
|
||||
"cache_read_input_token_cost": 5.46875e-07,
|
||||
|
|
@ -407,7 +411,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"eu.amazon.nova-2-lite-v1:0": {
|
||||
"cache_read_input_token_cost": 8.25e-08,
|
||||
|
|
@ -424,7 +429,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"eu.amazon.nova-2-pro-preview-20251202-v1:0": {
|
||||
"cache_read_input_token_cost": 5.46875e-07,
|
||||
|
|
@ -443,7 +449,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"us.amazon.nova-2-lite-v1:0": {
|
||||
"cache_read_input_token_cost": 8.25e-08,
|
||||
|
|
@ -460,7 +467,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"us.amazon.nova-2-pro-preview-20251202-v1:0": {
|
||||
"cache_read_input_token_cost": 5.46875e-07,
|
||||
|
|
@ -479,7 +487,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"amazon.nova-2-multimodal-embeddings-v1:0": {
|
||||
"litellm_provider": "bedrock",
|
||||
|
|
@ -522,7 +531,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"amazon.rerank-v1:0": {
|
||||
"input_cost_per_query": 0.001,
|
||||
|
|
@ -1478,7 +1488,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"apac.amazon.nova-micro-v1:0": {
|
||||
"input_cost_per_token": 3.7e-08,
|
||||
|
|
@ -1504,7 +1515,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"apac.anthropic.claude-3-5-sonnet-20240620-v1:0": {
|
||||
"input_cost_per_token": 3e-06,
|
||||
|
|
@ -7582,7 +7594,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"bedrock/us-gov-east-1/amazon.titan-embed-text-v1": {
|
||||
"input_cost_per_token": 1e-07,
|
||||
|
|
@ -7713,7 +7726,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"bedrock/us-gov-west-1/amazon.titan-embed-text-v1": {
|
||||
"input_cost_per_token": 1e-07,
|
||||
|
|
@ -12115,7 +12129,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"eu.amazon.nova-micro-v1:0": {
|
||||
"input_cost_per_token": 4.6e-08,
|
||||
|
|
@ -12142,7 +12157,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"eu.anthropic.claude-3-5-haiku-20241022-v1:0": {
|
||||
"input_cost_per_token": 2.5e-07,
|
||||
|
|
@ -14599,17 +14615,14 @@
|
|||
"uses_embed_content": true
|
||||
},
|
||||
"vertex_ai/gemini-embedding-2-preview": {
|
||||
"input_cost_per_audio_per_second": 0.00016,
|
||||
"input_cost_per_image": 0.00012,
|
||||
"input_cost_per_token": 2e-07,
|
||||
"input_cost_per_video_per_second": 0.00079,
|
||||
"input_cost_per_token": 1.5e-07,
|
||||
"litellm_provider": "vertex_ai",
|
||||
"max_input_tokens": 8192,
|
||||
"max_tokens": 8192,
|
||||
"mode": "embedding",
|
||||
"output_cost_per_token": 0,
|
||||
"output_vector_size": 3072,
|
||||
"source": "https://cloud.google.com/vertex-ai/generative-ai/pricing",
|
||||
"source": "https://ai.google.dev/gemini-api/docs/embeddings#multimodal",
|
||||
"supports_multimodal": true,
|
||||
"uses_embed_content": true
|
||||
},
|
||||
|
|
@ -14626,18 +14639,6 @@
|
|||
"source": "https://cloud.google.com/vertex-ai/generative-ai/pricing",
|
||||
"uses_embed_content": true
|
||||
},
|
||||
"vertex_ai/gemini-embedding-2-preview": {
|
||||
"input_cost_per_token": 1.5e-07,
|
||||
"litellm_provider": "vertex_ai",
|
||||
"max_input_tokens": 8192,
|
||||
"max_tokens": 8192,
|
||||
"mode": "embedding",
|
||||
"output_cost_per_token": 0,
|
||||
"output_vector_size": 3072,
|
||||
"source": "https://ai.google.dev/gemini-api/docs/embeddings#multimodal",
|
||||
"supports_multimodal": true,
|
||||
"uses_embed_content": true
|
||||
},
|
||||
"gemini/gemini-embedding-001": {
|
||||
"input_cost_per_token": 1.5e-07,
|
||||
"litellm_provider": "gemini",
|
||||
|
|
@ -16834,7 +16835,8 @@
|
|||
"supports_reasoning": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_video_input": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"gpt-3.5-turbo": {
|
||||
"input_cost_per_token": 5e-07,
|
||||
|
|
@ -28050,7 +28052,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"us.amazon.nova-micro-v1:0": {
|
||||
"input_cost_per_token": 3.5e-08,
|
||||
|
|
@ -28076,7 +28079,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": false,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"us.amazon.nova-pro-v1:0": {
|
||||
"input_cost_per_token": 8e-07,
|
||||
|
|
@ -28090,7 +28094,8 @@
|
|||
"supports_pdf_input": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_response_schema": true,
|
||||
"supports_vision": true
|
||||
"supports_vision": true,
|
||||
"supports_s3_input": true
|
||||
},
|
||||
"us.anthropic.claude-3-5-haiku-20241022-v1:0": {
|
||||
"cache_creation_input_token_cost": 1e-06,
|
||||
|
|
@ -31013,7 +31018,9 @@
|
|||
"mode": "chat",
|
||||
"output_cost_per_token": 3.2e-06,
|
||||
"source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#glm-models",
|
||||
"supported_regions": ["global"],
|
||||
"supported_regions": [
|
||||
"global"
|
||||
],
|
||||
"supports_function_calling": true,
|
||||
"supports_prompt_caching": true,
|
||||
"supports_reasoning": true,
|
||||
|
|
|
|||
183
tests/local_testing/test_bedrock_s3_content.py
Normal file
183
tests/local_testing/test_bedrock_s3_content.py
Normal file
|
|
@ -0,0 +1,183 @@
|
|||
"""
|
||||
Integration tests for Bedrock S3 URL pass-through support.
|
||||
|
||||
These tests make real requests to AWS Bedrock and S3.
|
||||
Requires AWS credentials and a pre-populated S3 bucket.
|
||||
|
||||
To run:
|
||||
AWS_PROFILE=<your-profile> BEDROCK_S3_TEST_BUCKET=<your-bucket> \
|
||||
pytest tests/local_testing/test_bedrock_s3_content.py -v
|
||||
|
||||
Required env vars (one of):
|
||||
AWS_PROFILE — named profile with Bedrock + S3 access
|
||||
AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY
|
||||
|
||||
Required:
|
||||
BEDROCK_S3_TEST_BUCKET — S3 bucket containing test files
|
||||
|
||||
Expected test files in the bucket:
|
||||
photos/cat.jpg, docs/report.pdf, videos/demo.mp4
|
||||
"""
|
||||
|
||||
import os
|
||||
|
||||
import pytest
|
||||
|
||||
_has_aws_creds = bool(
|
||||
os.environ.get("AWS_PROFILE")
|
||||
or (os.environ.get("AWS_ACCESS_KEY_ID") and os.environ.get("AWS_SECRET_ACCESS_KEY"))
|
||||
)
|
||||
_s3_bucket = os.environ.get("BEDROCK_S3_TEST_BUCKET", "")
|
||||
_skip_reason = "Requires AWS credentials (AWS_PROFILE or AWS_ACCESS_KEY_ID/SECRET) and BEDROCK_S3_TEST_BUCKET"
|
||||
|
||||
|
||||
@pytest.fixture(autouse=True)
|
||||
def _ensure_s3_input_flag(monkeypatch):
|
||||
"""Patch supports_s3_input in factory.py for Nova models.
|
||||
|
||||
Until this PR merges upstream, the remote model_prices JSON (fetched at
|
||||
import time) won't contain the supports_s3_input flag. We monkeypatch
|
||||
the function directly in factory.py where it's imported.
|
||||
"""
|
||||
|
||||
def _patched(model, custom_llm_provider=None):
|
||||
model_lower = model.lower()
|
||||
return (
|
||||
"nova" in model_lower
|
||||
and "micro" not in model_lower
|
||||
and "embed" not in model_lower
|
||||
)
|
||||
|
||||
monkeypatch.setattr(
|
||||
"litellm.litellm_core_utils.prompt_templates.factory.supports_s3_input",
|
||||
_patched,
|
||||
)
|
||||
|
||||
|
||||
@pytest.mark.skipif(not (_has_aws_creds and _s3_bucket), reason=_skip_reason)
|
||||
class TestS3IntegrationBedrock:
|
||||
"""Integration tests that send real requests to Bedrock with S3 content."""
|
||||
|
||||
_MODELS = [
|
||||
"bedrock/us.amazon.nova-lite-v1:0",
|
||||
"bedrock/us.amazon.nova-pro-v1:0",
|
||||
"bedrock/us.amazon.nova-2-lite-v1:0",
|
||||
]
|
||||
|
||||
@pytest.mark.parametrize("model", _MODELS)
|
||||
def test_should_send_s3_image(self, model):
|
||||
import litellm
|
||||
|
||||
response = litellm.completion(
|
||||
model=model,
|
||||
messages=[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "Describe this image in one sentence."},
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": f"s3://{_s3_bucket}/photos/cat.jpg"},
|
||||
},
|
||||
],
|
||||
}
|
||||
],
|
||||
max_tokens=100,
|
||||
)
|
||||
assert response.choices[0].message.content is not None
|
||||
assert len(response.choices[0].message.content) > 0
|
||||
|
||||
@pytest.mark.parametrize("model", _MODELS)
|
||||
def test_should_send_s3_pdf(self, model):
|
||||
import litellm
|
||||
|
||||
response = litellm.completion(
|
||||
model=model,
|
||||
messages=[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "Summarize this document in one sentence."},
|
||||
{
|
||||
"type": "file",
|
||||
"file": {"file_data": f"s3://{_s3_bucket}/docs/report.pdf"},
|
||||
},
|
||||
],
|
||||
}
|
||||
],
|
||||
max_tokens=100,
|
||||
)
|
||||
assert response.choices[0].message.content is not None
|
||||
assert len(response.choices[0].message.content) > 0
|
||||
|
||||
@pytest.mark.parametrize("model", _MODELS)
|
||||
def test_should_send_s3_video(self, model):
|
||||
import litellm
|
||||
|
||||
response = litellm.completion(
|
||||
model=model,
|
||||
messages=[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "Describe this video in one sentence."},
|
||||
{
|
||||
"type": "video_url",
|
||||
"video_url": {"url": f"s3://{_s3_bucket}/videos/demo.mp4"},
|
||||
},
|
||||
],
|
||||
}
|
||||
],
|
||||
max_tokens=100,
|
||||
)
|
||||
assert response.choices[0].message.content is not None
|
||||
assert len(response.choices[0].message.content) > 0
|
||||
|
||||
@pytest.mark.parametrize("model", _MODELS)
|
||||
def test_should_send_multiple_s3_files_in_one_message(self, model):
|
||||
import litellm
|
||||
|
||||
response = litellm.completion(
|
||||
model=model,
|
||||
messages=[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "Describe what you see in these two items."},
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": f"s3://{_s3_bucket}/photos/cat.jpg"},
|
||||
},
|
||||
{
|
||||
"type": "file",
|
||||
"file": {"file_data": f"s3://{_s3_bucket}/docs/report.pdf"},
|
||||
},
|
||||
],
|
||||
}
|
||||
],
|
||||
max_tokens=200,
|
||||
)
|
||||
assert response.choices[0].message.content is not None
|
||||
assert len(response.choices[0].message.content) > 0
|
||||
|
||||
def test_should_reject_s3_url_for_unsupported_model(self):
|
||||
"""Claude models don't support s3Location — should get a clear error."""
|
||||
import litellm
|
||||
|
||||
with pytest.raises(Exception, match="does not support s3://"):
|
||||
litellm.completion(
|
||||
model="bedrock/anthropic.claude-3-sonnet-20240229-v1:0",
|
||||
messages=[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "Describe this image"},
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": f"s3://{_s3_bucket}/photos/cat.jpg"},
|
||||
},
|
||||
],
|
||||
}
|
||||
],
|
||||
max_tokens=50,
|
||||
)
|
||||
468
tests/test_litellm/llms/bedrock/chat/test_s3_content_blocks.py
Normal file
468
tests/test_litellm/llms/bedrock/chat/test_s3_content_blocks.py
Normal file
|
|
@ -0,0 +1,468 @@
|
|||
"""
|
||||
Tests for Bedrock S3 URL pass-through support (PR1).
|
||||
|
||||
When users provide s3:// URLs in OpenAI-format messages, LiteLLM should
|
||||
map them to Bedrock's native s3Location blocks without downloading or
|
||||
base64-encoding.
|
||||
"""
|
||||
|
||||
import hashlib
|
||||
import os
|
||||
from unittest.mock import patch
|
||||
|
||||
import pytest
|
||||
|
||||
from litellm.litellm_core_utils.prompt_templates.factory import BedrockImageProcessor
|
||||
|
||||
|
||||
def _mock_supports_s3(model: str, custom_llm_provider=None) -> bool:
|
||||
"""Mock supports_s3_input that returns True for Nova models."""
|
||||
return "nova" in model.lower() and "micro" not in model.lower()
|
||||
|
||||
|
||||
class TestS3ImageBlocks:
|
||||
"""Test S3 URL → ImageBlock creation."""
|
||||
|
||||
def test_should_create_s3_image_block_from_jpg_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/image.jpg")
|
||||
assert result["image"]["format"] == "jpeg"
|
||||
assert result["image"]["source"]["s3Location"]["uri"] == "s3://bucket/image.jpg"
|
||||
assert "bytes" not in result["image"]["source"]
|
||||
|
||||
def test_should_create_s3_image_block_from_png_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/image.png")
|
||||
assert result["image"]["format"] == "png"
|
||||
assert result["image"]["source"]["s3Location"]["uri"] == "s3://bucket/image.png"
|
||||
|
||||
def test_should_create_s3_image_block_from_gif_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/image.gif")
|
||||
assert result["image"]["format"] == "gif"
|
||||
|
||||
def test_should_create_s3_image_block_from_webp_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/image.webp")
|
||||
assert result["image"]["format"] == "webp"
|
||||
|
||||
def test_should_create_s3_image_block_from_jpeg_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/photo.jpeg")
|
||||
assert result["image"]["format"] == "jpeg"
|
||||
|
||||
|
||||
class TestS3DocumentBlocks:
|
||||
"""Test S3 URL → DocumentBlock creation."""
|
||||
|
||||
def test_should_create_s3_document_block_from_pdf_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/doc.pdf")
|
||||
assert result["document"]["format"] == "pdf"
|
||||
assert (
|
||||
result["document"]["source"]["s3Location"]["uri"] == "s3://bucket/doc.pdf"
|
||||
)
|
||||
assert "name" in result["document"]
|
||||
assert "bytes" not in result["document"]["source"]
|
||||
|
||||
def test_should_create_s3_document_block_from_docx_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/doc.docx")
|
||||
assert result["document"]["format"] == "docx"
|
||||
|
||||
def test_should_create_s3_document_block_from_xlsx_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/data.xlsx")
|
||||
assert result["document"]["format"] == "xlsx"
|
||||
|
||||
def test_should_create_s3_document_block_from_md_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/readme.md")
|
||||
assert result["document"]["format"] == "md"
|
||||
|
||||
def test_should_create_s3_document_block_from_txt_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/notes.txt")
|
||||
assert result["document"]["format"] == "txt"
|
||||
|
||||
def test_should_create_s3_document_block_from_csv_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/data.csv")
|
||||
assert result["document"]["format"] == "csv"
|
||||
|
||||
def test_should_create_s3_document_block_from_html_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/page.html")
|
||||
assert result["document"]["format"] == "html"
|
||||
|
||||
def test_should_generate_deterministic_document_names(self):
|
||||
r1 = BedrockImageProcessor.process_image_sync("s3://bucket/doc.pdf")
|
||||
r2 = BedrockImageProcessor.process_image_sync("s3://bucket/doc.pdf")
|
||||
assert r1["document"]["name"] == r2["document"]["name"]
|
||||
|
||||
r3 = BedrockImageProcessor.process_image_sync("s3://bucket/other.pdf")
|
||||
assert r1["document"]["name"] != r3["document"]["name"]
|
||||
|
||||
|
||||
class TestS3VideoBlocks:
|
||||
"""Test S3 URL → VideoBlock creation."""
|
||||
|
||||
def test_should_create_s3_video_block_from_mp4_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/video.mp4")
|
||||
assert result["video"]["format"] == "mp4"
|
||||
assert result["video"]["source"]["s3Location"]["uri"] == "s3://bucket/video.mp4"
|
||||
assert "bytes" not in result["video"]["source"]
|
||||
|
||||
def test_should_create_s3_video_block_from_mov_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/video.mov")
|
||||
assert result["video"]["format"] == "mov"
|
||||
|
||||
def test_should_create_s3_video_block_from_mkv_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/video.mkv")
|
||||
assert result["video"]["format"] == "mkv"
|
||||
|
||||
def test_should_create_s3_video_block_from_3gp_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/video.3gp")
|
||||
assert result["video"]["format"] == "3gp"
|
||||
|
||||
def test_should_create_s3_video_block_from_3gpp_url(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/video.3gpp")
|
||||
assert result["video"]["format"] == "3gp"
|
||||
|
||||
|
||||
class TestS3ErrorCases:
|
||||
"""Test error handling for S3 URLs."""
|
||||
|
||||
def test_should_raise_error_for_s3_url_without_extension(self):
|
||||
with pytest.raises(ValueError, match="no extension"):
|
||||
BedrockImageProcessor.process_image_sync("s3://bucket/noext")
|
||||
|
||||
def test_should_raise_error_for_unsupported_extension(self):
|
||||
with pytest.raises(ValueError, match="Unsupported file format"):
|
||||
BedrockImageProcessor.process_image_sync("s3://bucket/file.xyz")
|
||||
|
||||
def test_should_not_raise_for_s3_url_without_extension_when_format_provided(self):
|
||||
"""format override bypasses extension requirement."""
|
||||
result = BedrockImageProcessor.process_image_sync(
|
||||
"s3://bucket/blob", format="image/png"
|
||||
)
|
||||
assert result["image"]["format"] == "png"
|
||||
|
||||
|
||||
class TestS3EdgeCases:
|
||||
"""Test edge cases: uppercase extensions, dots in path, etc."""
|
||||
|
||||
def test_should_handle_uppercase_extension(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/IMAGE.JPG")
|
||||
assert result["image"]["format"] == "jpeg"
|
||||
|
||||
def test_should_handle_mixed_case_extension(self):
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/doc.Pdf")
|
||||
assert result["document"]["format"] == "pdf"
|
||||
|
||||
def test_should_handle_dots_in_path(self):
|
||||
result = BedrockImageProcessor.process_image_sync(
|
||||
"s3://bucket/path/to/file.with.dots.jpg"
|
||||
)
|
||||
assert result["image"]["format"] == "jpeg"
|
||||
assert (
|
||||
result["image"]["source"]["s3Location"]["uri"]
|
||||
== "s3://bucket/path/to/file.with.dots.jpg"
|
||||
)
|
||||
|
||||
def test_should_handle_deeply_nested_s3_path(self):
|
||||
url = "s3://my-bucket/a/b/c/d/report.pdf"
|
||||
result = BedrockImageProcessor.process_image_sync(url)
|
||||
assert result["document"]["format"] == "pdf"
|
||||
assert result["document"]["source"]["s3Location"]["uri"] == url
|
||||
|
||||
|
||||
class TestS3FormatOverride:
|
||||
"""Test explicit format override for S3 URLs."""
|
||||
|
||||
def test_should_handle_s3_url_with_explicit_format_override(self):
|
||||
# Even though extension is .dat, explicit format says pdf
|
||||
result = BedrockImageProcessor.process_image_sync(
|
||||
"s3://bucket/file.dat", format="application/pdf"
|
||||
)
|
||||
assert result["document"]["format"] == "pdf"
|
||||
|
||||
def test_should_handle_s3_url_with_simple_format_override(self):
|
||||
result = BedrockImageProcessor.process_image_sync(
|
||||
"s3://bucket/file.dat", format="png"
|
||||
)
|
||||
assert result["image"]["format"] == "png"
|
||||
|
||||
|
||||
class TestS3AsyncProcessing:
|
||||
"""Test async S3 URL processing."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_should_create_s3_image_block_async(self):
|
||||
result = await BedrockImageProcessor.process_image_async(
|
||||
"s3://bucket/image.jpg", format=None
|
||||
)
|
||||
assert result["image"]["format"] == "jpeg"
|
||||
assert result["image"]["source"]["s3Location"]["uri"] == "s3://bucket/image.jpg"
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_should_create_s3_document_block_async(self):
|
||||
result = await BedrockImageProcessor.process_image_async(
|
||||
"s3://bucket/doc.pdf", format=None
|
||||
)
|
||||
assert result["document"]["format"] == "pdf"
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_should_create_s3_video_block_async(self):
|
||||
result = await BedrockImageProcessor.process_image_async(
|
||||
"s3://bucket/video.mp4", format=None
|
||||
)
|
||||
assert result["video"]["format"] == "mp4"
|
||||
|
||||
|
||||
class TestExistingPathsUnbroken:
|
||||
"""Ensure existing base64 and HTTPS paths still work."""
|
||||
|
||||
def test_should_still_handle_base64_urls(self):
|
||||
# Minimal valid base64 PNG
|
||||
b64_url = "data:image/png;base64,iVBORw0KGgo="
|
||||
result = BedrockImageProcessor.process_image_sync(b64_url)
|
||||
assert result["image"]["format"] == "png"
|
||||
assert "bytes" in result["image"]["source"]
|
||||
assert "s3Location" not in result["image"]["source"]
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.litellm_core_utils.prompt_templates.factory.supports_s3_input",
|
||||
_mock_supports_s3,
|
||||
)
|
||||
class TestFullMessageTransformation:
|
||||
"""Test S3 URLs through full Bedrock message transformation."""
|
||||
|
||||
def test_should_pass_s3_image_url_through_full_message_transformation(self):
|
||||
from litellm.litellm_core_utils.prompt_templates.factory import (
|
||||
_bedrock_converse_messages_pt,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "Describe this image"},
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": "s3://my-bucket/photos/cat.jpg"},
|
||||
},
|
||||
],
|
||||
}
|
||||
]
|
||||
result = _bedrock_converse_messages_pt(
|
||||
messages=messages,
|
||||
model="us.amazon.nova-lite-v1:0",
|
||||
llm_provider="bedrock",
|
||||
)
|
||||
# Find the image block in the result
|
||||
content_blocks = result[0]["content"]
|
||||
image_blocks = [b for b in content_blocks if "image" in b]
|
||||
assert len(image_blocks) == 1
|
||||
assert (
|
||||
image_blocks[0]["image"]["source"]["s3Location"]["uri"]
|
||||
== "s3://my-bucket/photos/cat.jpg"
|
||||
)
|
||||
assert image_blocks[0]["image"]["format"] == "jpeg"
|
||||
|
||||
def test_should_pass_s3_video_url_through_full_message_transformation(self):
|
||||
from litellm.litellm_core_utils.prompt_templates.factory import (
|
||||
_bedrock_converse_messages_pt,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "Describe this video"},
|
||||
{
|
||||
"type": "video_url",
|
||||
"video_url": {"url": "s3://my-bucket/videos/demo.mp4"},
|
||||
},
|
||||
],
|
||||
}
|
||||
]
|
||||
result = _bedrock_converse_messages_pt(
|
||||
messages=messages,
|
||||
model="us.amazon.nova-pro-v1:0",
|
||||
llm_provider="bedrock",
|
||||
)
|
||||
content_blocks = result[0]["content"]
|
||||
video_blocks = [b for b in content_blocks if "video" in b]
|
||||
assert len(video_blocks) == 1
|
||||
assert (
|
||||
video_blocks[0]["video"]["source"]["s3Location"]["uri"]
|
||||
== "s3://my-bucket/videos/demo.mp4"
|
||||
)
|
||||
assert video_blocks[0]["video"]["format"] == "mp4"
|
||||
|
||||
def test_should_pass_s3_file_url_through_full_message_transformation(self):
|
||||
from litellm.litellm_core_utils.prompt_templates.factory import (
|
||||
_bedrock_converse_messages_pt,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "Summarize this document"},
|
||||
{
|
||||
"type": "file",
|
||||
"file": {
|
||||
"file_data": "s3://my-bucket/docs/report.pdf",
|
||||
},
|
||||
},
|
||||
],
|
||||
}
|
||||
]
|
||||
result = _bedrock_converse_messages_pt(
|
||||
messages=messages,
|
||||
model="us.amazon.nova-lite-v1:0",
|
||||
llm_provider="bedrock",
|
||||
)
|
||||
content_blocks = result[0]["content"]
|
||||
doc_blocks = [b for b in content_blocks if "document" in b]
|
||||
assert len(doc_blocks) == 1
|
||||
assert (
|
||||
doc_blocks[0]["document"]["source"]["s3Location"]["uri"]
|
||||
== "s3://my-bucket/docs/report.pdf"
|
||||
)
|
||||
assert doc_blocks[0]["document"]["format"] == "pdf"
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_should_pass_s3_video_url_through_async_message_transformation(self):
|
||||
from litellm.litellm_core_utils.prompt_templates.factory import (
|
||||
BedrockConverseMessagesProcessor,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "Describe this video"},
|
||||
{
|
||||
"type": "video_url",
|
||||
"video_url": {"url": "s3://my-bucket/videos/demo.mp4"},
|
||||
},
|
||||
],
|
||||
}
|
||||
]
|
||||
result = (
|
||||
await BedrockConverseMessagesProcessor._bedrock_converse_messages_pt_async(
|
||||
messages=messages,
|
||||
model="us.amazon.nova-pro-v1:0",
|
||||
llm_provider="bedrock",
|
||||
)
|
||||
)
|
||||
content_blocks = result[0]["content"]
|
||||
video_blocks = [b for b in content_blocks if "video" in b]
|
||||
assert len(video_blocks) == 1
|
||||
assert (
|
||||
video_blocks[0]["video"]["source"]["s3Location"]["uri"]
|
||||
== "s3://my-bucket/videos/demo.mp4"
|
||||
)
|
||||
assert video_blocks[0]["video"]["format"] == "mp4"
|
||||
|
||||
def test_should_handle_video_url_as_plain_string(self):
|
||||
"""video_url can be a plain string instead of a dict."""
|
||||
from litellm.litellm_core_utils.prompt_templates.factory import (
|
||||
_bedrock_converse_messages_pt,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "Describe"},
|
||||
{"type": "video_url", "video_url": "s3://bucket/video.mp4"},
|
||||
],
|
||||
}
|
||||
]
|
||||
result = _bedrock_converse_messages_pt(
|
||||
messages=messages,
|
||||
model="us.amazon.nova-pro-v1:0",
|
||||
llm_provider="bedrock",
|
||||
)
|
||||
content_blocks = result[0]["content"]
|
||||
video_blocks = [b for b in content_blocks if "video" in b]
|
||||
assert len(video_blocks) == 1
|
||||
assert video_blocks[0]["video"]["format"] == "mp4"
|
||||
|
||||
def test_should_still_handle_https_image_url(self):
|
||||
"""Regression: HTTPS URLs should still be downloaded and base64-encoded, not treated as S3."""
|
||||
from unittest.mock import patch
|
||||
|
||||
from litellm.litellm_core_utils.prompt_templates.factory import (
|
||||
_bedrock_converse_messages_pt,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "Describe this."},
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": "https://example.com/photo.jpg"},
|
||||
},
|
||||
],
|
||||
}
|
||||
]
|
||||
|
||||
fake_bytes = b"\x89PNG\r\n\x1a\n" + b"\x00" * 100
|
||||
with patch.object(
|
||||
BedrockImageProcessor,
|
||||
"get_image_details",
|
||||
return_value=(fake_bytes, "image/jpeg"),
|
||||
):
|
||||
result = _bedrock_converse_messages_pt(
|
||||
messages=messages,
|
||||
model="us.amazon.nova-lite-v1:0",
|
||||
llm_provider="bedrock",
|
||||
)
|
||||
content_blocks = result[0]["content"]
|
||||
img_blocks = [b for b in content_blocks if "image" in b]
|
||||
assert len(img_blocks) == 1
|
||||
# Should have inline bytes, NOT s3Location
|
||||
assert "bytes" in img_blocks[0]["image"]["source"]
|
||||
assert "s3Location" not in img_blocks[0]["image"]["source"]
|
||||
|
||||
def test_should_accept_s3_video_url_in_extract(self):
|
||||
"""s3:// video URLs should pass through extraction."""
|
||||
url = BedrockImageProcessor._extract_video_url(
|
||||
{"type": "video_url", "video_url": {"url": "s3://bucket/video.mp4"}}
|
||||
)
|
||||
assert url == "s3://bucket/video.mp4"
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.litellm_core_utils.prompt_templates.factory.supports_s3_input",
|
||||
_mock_supports_s3,
|
||||
)
|
||||
class TestS3ModelGuard:
|
||||
"""Test that S3 URLs are rejected for models that don't support s3Location."""
|
||||
|
||||
def test_should_reject_s3_url_for_unsupported_model(self):
|
||||
with pytest.raises(ValueError, match="does not support s3://"):
|
||||
BedrockImageProcessor.process_image_sync(
|
||||
"s3://bucket/image.jpg",
|
||||
model="anthropic.claude-3-sonnet-20240229-v1:0",
|
||||
custom_llm_provider="bedrock",
|
||||
)
|
||||
|
||||
def test_should_allow_s3_url_for_nova_model(self):
|
||||
result = BedrockImageProcessor.process_image_sync(
|
||||
"s3://bucket/image.jpg",
|
||||
model="us.amazon.nova-lite-v1:0",
|
||||
custom_llm_provider="bedrock",
|
||||
)
|
||||
assert result["image"]["source"]["s3Location"]["uri"] == "s3://bucket/image.jpg"
|
||||
|
||||
def test_should_allow_s3_url_when_model_not_specified(self):
|
||||
"""When no model is provided, allow S3 pass-through (caller's responsibility)."""
|
||||
result = BedrockImageProcessor.process_image_sync("s3://bucket/image.jpg")
|
||||
assert result["image"]["source"]["s3Location"]["uri"] == "s3://bucket/image.jpg"
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_should_reject_s3_url_for_unsupported_model_async(self):
|
||||
with pytest.raises(ValueError, match="does not support s3://"):
|
||||
await BedrockImageProcessor.process_image_async(
|
||||
"s3://bucket/image.jpg",
|
||||
format=None,
|
||||
model="anthropic.claude-3-sonnet-20240229-v1:0",
|
||||
custom_llm_provider="bedrock",
|
||||
)
|
||||
|
|
@ -734,6 +734,7 @@ def test_aaamodel_prices_and_context_window_json_is_valid():
|
|||
"supports_pdf_input": {"type": "boolean"},
|
||||
"supports_prompt_caching": {"type": "boolean"},
|
||||
"supports_response_schema": {"type": "boolean"},
|
||||
"supports_s3_input": {"type": "boolean"},
|
||||
"supports_system_messages": {"type": "boolean"},
|
||||
"supports_tool_choice": {"type": "boolean"},
|
||||
"supports_video_input": {"type": "boolean"},
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue