Pass through cachePoint and other Bedrock-native content blocks

Fixes #1954. The _to_bedrock_content_items function was raising
ValueError for unrecognized Bedrock-native dict content blocks
like cachePoint, guardContent, audio, and video. These are valid
Bedrock Converse API content types that should be passed through
to the API unchanged, just like text, image, and document blocks.

Instead of maintaining an explicit allowlist that breaks whenever
AWS adds new content types, pass through any dict that does not
match the OpenAI-style format (no "type" key). The Bedrock API
itself validates content blocks, so instructor does not need to
be the gatekeeper here.

Signed-off-by: debu-sinha <debusinha2009@gmail.com>

Author: debu-sinha

instructor/providers/bedrock/utils.py

@@ -225,10 +225,14 @@ def _to_bedrock_content_items(content: Any) -> list[dict[str, Any]]:
             {"text":"..."}
             {"image":{"format":"jpeg|png|gif|webp","source":{"bytes": <raw bytes>}}}
             {"document":{"format":"pdf|csv|doc|docx|xls|xlsx|html|txt|md","name":"...","source":{"bytes": <raw bytes>}}}
+            {"cachePoint":{"type":"default"}}
+            Any other valid Bedrock ContentBlock dict (guardContent, toolUse,
+            toolResult, audio, video, reasoningContent, etc.)
 
     Note:
-      - We do not validate or normalize Bedrock-native image/document blocks here.
-        Caller is responsible for providing valid 'format' and raw 'bytes'.
+      - We do not validate or normalize Bedrock-native blocks here.
+        Caller is responsible for providing valid structure.
+        The Bedrock API will reject malformed content blocks.
     """
     # Plain string
     if isinstance(content, str):
@@ -269,7 +273,10 @@ def _to_bedrock_content_items(content: Any) -> list[dict[str, Any]]:
                     items.append(p)
                     continue
 
-                raise ValueError(f"Unsupported dict content for Bedrock: {p}")
+                # Pass-through any other Bedrock-native content block (cachePoint,
+                # guardContent, toolUse, toolResult, audio, video, etc.)
+                items.append(p)
+                continue
 
             # Plain string elements inside list
             if isinstance(p, str):

tests/llm/test_bedrock/test_bedrock_native_passthrough.py

@@ -1,4 +1,7 @@
 from __future__ import annotations
+
+import pytest
+
 from instructor.providers.bedrock.utils import _to_bedrock_content_items
 
 
@@ -18,3 +21,49 @@ def test_bedrock_native_document_passthrough(tiny_pdf_bytes: bytes):
     native = {"document": {"format": "pdf", "source": {"bytes": tiny_pdf_bytes}}}
     items = _to_bedrock_content_items([native])
     assert items[0] == native
+
+
+def test_bedrock_native_cachepoint_passthrough():
+    """Regression test for #1954: cachePoint dicts must pass through."""
+    cache_point = {"cachePoint": {"type": "default"}}
+    items = _to_bedrock_content_items([cache_point])
+    assert items == [cache_point]
+
+
+def test_bedrock_native_cachepoint_with_ttl():
+    cache_point = {"cachePoint": {"type": "default", "ttl": "5m"}}
+    items = _to_bedrock_content_items([cache_point])
+    assert items == [cache_point]
+
+
+def test_bedrock_native_guard_content_passthrough():
+    guard = {"guardContent": {"text": {"text": "test content"}}}
+    items = _to_bedrock_content_items([guard])
+    assert items == [guard]
+
+
+@pytest.mark.parametrize(
+    "block",
+    [
+        {"cachePoint": {"type": "default"}},
+        {"guardContent": {"text": {"text": "check this"}}},
+        {"video": {"format": "mp4", "source": {"bytes": b"fake"}}},
+        {"audio": {"format": "mp3", "source": {"bytes": b"fake"}}},
+    ],
+    ids=["cachePoint", "guardContent", "video", "audio"],
+)
+def test_bedrock_native_content_block_passthrough(block: dict):
+    """All Bedrock-native content blocks should pass through unchanged."""
+    items = _to_bedrock_content_items([block])
+    assert items == [block]
+
+
+def test_mixed_content_with_cachepoint():
+    """Regression test for #1954: cachePoint mixed with text in real usage."""
+    content = [
+        {"text": "Say hello world."},
+        {"cachePoint": {"type": "default"}},
+        {"text": "This is a test message."},
+    ]
+    items = _to_bedrock_content_items(content)
+    assert items == content