mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-17 09:34:48 +00:00
fix(vertex_ai,azure_ai): auto-add thinking display and tool change betas, redact streamed compaction summaries
Folds the header changes from #44103 into this PR. Vertex AI and Azure AI Claude routes now add thinking-display-updates-2026-08-18 when thinking.display is "updates" and mid-conversation-tool-changes-2026-07-01 when a tool_addition block is present, matching the Anthropic and Bedrock routes. With turn_off_message_logging on, compaction summaries from streamed messages and streamed chat completions are now redacted in spend logs and callback payloads, like the non-streamed response already was.
This commit is contained in:
parent
90b5ccb89c
commit
fbbcd52439
16 changed files with 309 additions and 22 deletions
|
|
@ -258,14 +258,18 @@ impl AnthropicBeta {
|
|||
Self::InterleavedThinking20250514 | Self::WebSearch20250305 => {
|
||||
&[BetaProvider::BedrockConverse, BetaProvider::Bedrock]
|
||||
}
|
||||
Self::MidConversationOutputConfig20260701
|
||||
| Self::MidConversationToolChanges20260701 => &[
|
||||
Self::MidConversationOutputConfig20260701 => &[
|
||||
BetaProvider::AzureAi,
|
||||
BetaProvider::BedrockConverse,
|
||||
BetaProvider::BedrockMantle,
|
||||
BetaProvider::VertexAi,
|
||||
BetaProvider::Databricks,
|
||||
],
|
||||
Self::MidConversationToolChanges20260701 => &[
|
||||
BetaProvider::BedrockConverse,
|
||||
BetaProvider::BedrockMantle,
|
||||
BetaProvider::Databricks,
|
||||
],
|
||||
Self::Output128k20250219 | Self::TokenEfficientTools20250219 => &[
|
||||
BetaProvider::AzureAi,
|
||||
BetaProvider::BedrockConverse,
|
||||
|
|
|
|||
|
|
@ -60,6 +60,7 @@
|
|||
"mcp-client-2025-11-20": "mcp-client-2025-11-20",
|
||||
"mcp-client-2025-04-04": "mcp-client-2025-04-04",
|
||||
"mcp-servers-2025-12-04": null,
|
||||
"mid-conversation-tool-changes-2026-07-01": "mid-conversation-tool-changes-2026-07-01",
|
||||
"output-128k-2025-02-19": null,
|
||||
"structured-output-2024-03-01": null,
|
||||
"per-turn-control-2026-07-01": "per-turn-control-2026-07-01",
|
||||
|
|
@ -204,6 +205,7 @@
|
|||
"mcp-client-2025-11-20": null,
|
||||
"mcp-client-2025-04-04": null,
|
||||
"mcp-servers-2025-12-04": null,
|
||||
"mid-conversation-tool-changes-2026-07-01": "mid-conversation-tool-changes-2026-07-01",
|
||||
"output-128k-2025-02-19": null,
|
||||
"structured-output-2024-03-01": null,
|
||||
"per-turn-control-2026-07-01": "per-turn-control-2026-07-01",
|
||||
|
|
|
|||
|
|
@ -149,7 +149,14 @@ def _redacted_provider_specific_fields(fields: Mapping[str, object]) -> Mapping[
|
|||
**({"reasoning_content": REDACTED_BY_LITELLM} if fields.get("reasoning_content") is not None else {}),
|
||||
**{
|
||||
key: None
|
||||
for key in ("thinking_blocks", "reasoningContentBlocks", "citations", "compaction_blocks")
|
||||
for key in (
|
||||
"thinking_blocks",
|
||||
"reasoningContentBlocks",
|
||||
"citations",
|
||||
"compaction_blocks",
|
||||
"compaction_start",
|
||||
"compaction_delta",
|
||||
)
|
||||
if key in fields
|
||||
},
|
||||
}
|
||||
|
|
|
|||
|
|
@ -52,6 +52,7 @@ from litellm.types.llms.anthropic import (
|
|||
ANTHROPIC_SERVER_TOOL_USE_ID_PREFIX,
|
||||
ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER,
|
||||
ANTHROPIC_TOOL_SEARCH_TOOL_TYPES,
|
||||
LEGACY_THINKING_DISPLAY_UPDATES_PROVIDERS,
|
||||
AllAnthropicToolsValues,
|
||||
AnthropicMcpServerTool,
|
||||
AnthropicMessagesToolChoice,
|
||||
|
|
@ -548,10 +549,12 @@ class AnthropicModelInfo(BaseLLMModelInfo):
|
|||
file_ids: Final = get_file_ids_from_messages(messages)
|
||||
return len(file_ids) > 0
|
||||
|
||||
def is_thinking_display_updates_used(self, thinking: AnthropicThinkingParam | None) -> bool:
|
||||
if not isinstance(thinking, dict):
|
||||
return False
|
||||
return thinking.get("type") in ("adaptive", "enabled") and thinking.get("display") == "updates"
|
||||
def is_thinking_display_updates_used(self, thinking: object) -> bool:
|
||||
match thinking:
|
||||
case {"type": "adaptive" | "enabled", "display": "updates"}:
|
||||
return True
|
||||
case _:
|
||||
return False
|
||||
|
||||
def is_mid_conversation_tool_change_used(self, messages: Sequence[object]) -> bool:
|
||||
for message in messages:
|
||||
|
|
@ -1003,7 +1006,10 @@ class AnthropicModelInfo(BaseLLMModelInfo):
|
|||
|
||||
budget_tokens: Final = _legacy_budget_tokens(thinking.get("budget_tokens", 0), drop_params=drop_params)
|
||||
display: Final = thinking.get("display")
|
||||
supported_display: Final = {"display": display} if display in ("summarized", "omitted") else {}
|
||||
keeps_display: Final = display in ("summarized", "omitted") or (
|
||||
display == "updates" and custom_llm_provider in LEGACY_THINKING_DISPLAY_UPDATES_PROVIDERS
|
||||
)
|
||||
supported_display: Final = {"display": display} if keeps_display else {}
|
||||
carried_keys: Final = carried_thinking_keys(
|
||||
thinking, rebuilt_keys=frozenset(("type", "budget_tokens", "display")), drop_params=drop_params
|
||||
)
|
||||
|
|
|
|||
|
|
@ -2,13 +2,18 @@
|
|||
Azure Anthropic transformation config - extends AnthropicConfig with Azure authentication
|
||||
"""
|
||||
|
||||
from collections.abc import Mapping
|
||||
from typing import Final
|
||||
|
||||
from pydantic import TypeAdapter
|
||||
|
||||
from litellm.llms.anthropic.chat.transformation import AnthropicConfig
|
||||
from litellm.llms.azure.common_utils import BaseAzureLLM
|
||||
from litellm.types.llms.openai import AllMessageValues
|
||||
from litellm.types.router import GenericLiteLLMParams
|
||||
|
||||
_OPTIONAL_PARAMS: Final = TypeAdapter(Mapping[str, object])
|
||||
|
||||
|
||||
def _promote_extra_body_to_optional_params(optional_params: dict) -> None:
|
||||
"""Promote anthropic-native passthrough keys out of ``extra_body``.
|
||||
|
|
@ -96,9 +101,16 @@ class AzureAnthropicConfig(AnthropicConfig):
|
|||
is_vertex_request=optional_params.get("is_vertex_request", False),
|
||||
user_anthropic_beta_headers=user_anthropic_beta_headers,
|
||||
mcp_server_used=mcp_server_used,
|
||||
is_thinking_display_updates_used=self.is_thinking_display_updates_used(
|
||||
_OPTIONAL_PARAMS.validate_python(optional_params).get("thinking")
|
||||
),
|
||||
)
|
||||
# Merge headers - Azure auth (api-key or Authorization) takes precedence
|
||||
headers = {**anthropic_headers, **headers}
|
||||
litellm_beta: Final = _OPTIONAL_PARAMS.validate_python(anthropic_headers).get("anthropic-beta")
|
||||
litellm_betas: Final[Mapping[str, str]] = (
|
||||
{"anthropic-beta": litellm_beta} if isinstance(litellm_beta, str) else {}
|
||||
)
|
||||
headers = {**anthropic_headers, **headers, **litellm_betas}
|
||||
|
||||
# Ensure anthropic-version header is set
|
||||
if "anthropic-version" not in headers:
|
||||
|
|
|
|||
|
|
@ -11,6 +11,8 @@ from litellm.llms.anthropic.pass_through.messages.transformation import (
|
|||
from litellm.types.llms.anthropic import (
|
||||
ANTHROPIC_BETA_HEADER_VALUES,
|
||||
ANTHROPIC_HOSTED_TOOLS,
|
||||
ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER,
|
||||
ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER,
|
||||
)
|
||||
from litellm.types.llms.anthropic_tool_search import get_tool_search_beta_header
|
||||
from litellm.types.llms.vertex_ai import VertexPartnerProvider
|
||||
|
|
@ -120,11 +122,23 @@ class VertexAIPartnerModelsAnthropicMessagesConfig(AnthropicMessagesConfig, Vert
|
|||
if messages_carry_output_config(messages):
|
||||
beta_values.add(ANTHROPIC_BETA_HEADER_VALUES.PER_TURN_CONTROL_2026_07_01.value)
|
||||
|
||||
if requires_native_compaction_beta("vertex_ai", _OPTIONAL_PARAMS.validate_python(optional_params), messages):
|
||||
typed_params: Final = _OPTIONAL_PARAMS.validate_python(optional_params)
|
||||
if requires_native_compaction_beta("vertex_ai", typed_params, messages):
|
||||
beta_values.add(ANTHROPIC_BETA_HEADER_VALUES.COMPACT_2026_09_04.value)
|
||||
|
||||
if beta_values:
|
||||
headers["anthropic-beta"] = ",".join(beta_values)
|
||||
thinking_display_betas: Final = (
|
||||
(ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER,)
|
||||
if anthropic_model_info.is_thinking_display_updates_used(typed_params.get("thinking"))
|
||||
else ()
|
||||
)
|
||||
tool_change_betas: Final = (
|
||||
(ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER,)
|
||||
if anthropic_model_info.is_mid_conversation_tool_change_used(messages)
|
||||
else ()
|
||||
)
|
||||
all_beta_values: Final = beta_values.union(thinking_display_betas, tool_change_betas)
|
||||
if all_beta_values:
|
||||
headers["anthropic-beta"] = ",".join(sorted(all_beta_values))
|
||||
|
||||
return headers, api_base
|
||||
|
||||
|
|
|
|||
|
|
@ -156,6 +156,9 @@ class VertexAIAnthropicConfig(AnthropicConfig):
|
|||
file_id_used=self.is_file_id_used(messages),
|
||||
mcp_server_used=self.is_mcp_server_used(optional_params.get("mcp_servers")),
|
||||
custom_llm_provider="vertex_ai",
|
||||
is_thinking_display_updates_used=self.is_thinking_display_updates_used(
|
||||
_NON_DEFAULT_PARAMS.validate_python(data).get("thinking")
|
||||
),
|
||||
)
|
||||
|
||||
beta_set: Final = set(auto_betas)
|
||||
|
|
|
|||
|
|
@ -702,7 +702,7 @@ class AnthropicResponseContentBlockCompaction(LiteLLMBaseModel):
|
|||
model_config = ConfigDict(frozen=True)
|
||||
|
||||
type: Literal["compaction"]
|
||||
content: str | None
|
||||
content: str | None = None
|
||||
signature: str | None = None
|
||||
|
||||
|
||||
|
|
@ -783,6 +783,9 @@ class AnthropicThinkingParam(TypedDict, total=False):
|
|||
block_binding: ReadOnly[AnthropicThinkingBlockBinding]
|
||||
|
||||
|
||||
LEGACY_THINKING_DISPLAY_UPDATES_PROVIDERS: Final = frozenset(("azure_ai", "vertex_ai"))
|
||||
|
||||
|
||||
class ANTHROPIC_HOSTED_TOOLS(str, Enum):
|
||||
WEB_SEARCH = "web_search"
|
||||
BASH = "bash"
|
||||
|
|
|
|||
|
|
@ -6512,10 +6512,19 @@ def test_handle_anthropic_messages_response_logging_reparses_the_served_body_wit
|
|||
assert logged.usage.completion_tokens == 7 # type: ignore[attr-defined]
|
||||
|
||||
|
||||
def test_handle_anthropic_messages_response_logging_reparses_a_served_compaction_response():
|
||||
@pytest.mark.parametrize(
|
||||
"compaction_block",
|
||||
[
|
||||
{"type": "compaction", "content": "Summary of the chat so far.", "signature": "sig-abc"},
|
||||
{"type": "compaction", "signature": "sig-abc"},
|
||||
],
|
||||
ids=["with_summary", "without_content"],
|
||||
)
|
||||
def test_handle_anthropic_messages_response_logging_reparses_a_served_compaction_response(
|
||||
compaction_block: dict[str, str],
|
||||
):
|
||||
logging_obj = _anthropic_messages_logging_obj()
|
||||
logging_obj.optional_params = {}
|
||||
compaction_block = {"type": "compaction", "content": "Summary of the chat so far.", "signature": "sig-abc"}
|
||||
|
||||
logged = logging_obj._handle_anthropic_messages_response_logging(
|
||||
result={
|
||||
|
|
@ -6534,7 +6543,9 @@ def test_handle_anthropic_messages_response_logging_reparses_a_served_compaction
|
|||
assert logged.id == "msg_compacted"
|
||||
assert logged.usage.prompt_tokens == 410 # type: ignore[attr-defined]
|
||||
assert logged.usage.completion_tokens == 95 # type: ignore[attr-defined]
|
||||
assert logged.choices[0].message.provider_specific_fields["compaction_blocks"] == [compaction_block] # type: ignore[union-attr]
|
||||
assert logged.choices[0].message.provider_specific_fields["compaction_blocks"] == [ # type: ignore[union-attr]
|
||||
{"content": None, **compaction_block}
|
||||
]
|
||||
|
||||
|
||||
def test_handle_anthropic_messages_response_logging_passes_model_response_through():
|
||||
|
|
|
|||
|
|
@ -55,6 +55,8 @@ def _provider_fields() -> dict[str, object]:
|
|||
"reasoningContentBlocks": [{"reasoningText": {"text": SECRET}}],
|
||||
"citations": [{"cited_text": SECRET}],
|
||||
"compaction_blocks": [{"type": "compaction", "content": SECRET, "signature": "compaction-signature"}],
|
||||
"compaction_start": {"type": "compaction", "content": SECRET},
|
||||
"compaction_delta": {"type": "compaction_delta", "content": SECRET},
|
||||
"thought_signatures": [SIGNATURE],
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -4236,15 +4236,17 @@ def test_thinking_display_beta_requires_active_thinking(thinking: object, expect
|
|||
|
||||
@pytest.mark.usefixtures("local_model_cost_map")
|
||||
@pytest.mark.parametrize(
|
||||
("display", "expected_thinking"),
|
||||
("provider", "display", "expected_thinking"),
|
||||
(
|
||||
("summarized", {"type": "adaptive", "display": "summarized"}),
|
||||
("omitted", {"type": "adaptive", "display": "omitted"}),
|
||||
("updates", {"type": "adaptive"}),
|
||||
("azure_ai", "summarized", {"type": "adaptive", "display": "summarized"}),
|
||||
("azure_ai", "omitted", {"type": "adaptive", "display": "omitted"}),
|
||||
("azure_ai", "updates", {"type": "adaptive", "display": "updates"}),
|
||||
("vertex_ai", "updates", {"type": "adaptive", "display": "updates"}),
|
||||
("anthropic", "updates", {"type": "adaptive"}),
|
||||
),
|
||||
)
|
||||
def test_shared_legacy_thinking_translation_preserves_supported_display(
|
||||
display: str, expected_thinking: dict[str, str]
|
||||
provider: str, display: str, expected_thinking: dict[str, str]
|
||||
) -> None:
|
||||
from litellm.llms.anthropic.common_utils import AnthropicModelInfo
|
||||
|
||||
|
|
@ -4255,7 +4257,7 @@ def test_shared_legacy_thinking_translation_preserves_supported_display(
|
|||
AnthropicModelInfo.translate_legacy_thinking_for_adaptive_model(
|
||||
model="claude-opus-5",
|
||||
optional_params=optional_params,
|
||||
custom_llm_provider="azure_ai",
|
||||
custom_llm_provider=provider,
|
||||
drop_params=False,
|
||||
)
|
||||
|
||||
|
|
|
|||
|
|
@ -642,3 +642,45 @@ async def test_outgoing_azure_compaction_request_carries_the_compact_2026_09_04_
|
|||
|
||||
assert len(received) == 1, received
|
||||
assert received[0].headers["anthropic-beta"].split(",").count("compact-2026-09-04") == 1, dict(received[0].headers)
|
||||
|
||||
|
||||
@pytest.mark.usefixtures("local_model_cost_map")
|
||||
@pytest.mark.parametrize("feature", ("thinking", "output_config", "tool_addition", "tool_removal", "none"))
|
||||
@pytest.mark.parametrize("explicit_beta", (False, True))
|
||||
def test_messages_feature_beta_headers(feature: str, explicit_beta: bool) -> None:
|
||||
from typing import Final
|
||||
|
||||
from litellm.llms.azure_ai.anthropic.messages_transformation import AzureAnthropicMessagesConfig
|
||||
from litellm.types.llms.anthropic import (
|
||||
ANTHROPIC_BETA_HEADER_VALUES,
|
||||
ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER,
|
||||
ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER,
|
||||
)
|
||||
|
||||
beta: Final = (
|
||||
ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER
|
||||
if feature == "thinking"
|
||||
else ANTHROPIC_BETA_HEADER_VALUES.PER_TURN_CONTROL_2026_07_01.value
|
||||
if feature == "output_config"
|
||||
else ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER
|
||||
)
|
||||
message: Final = (
|
||||
{"role": "system", "content": [], "output_config": {"effort": "high"}}
|
||||
if feature == "output_config"
|
||||
else {"role": "system", "content": [{"type": feature, "tool": {"type": "tool_reference", "name": "ping"}}]}
|
||||
if feature in ("tool_addition", "tool_removal")
|
||||
else {"role": "user", "content": "Reply with OK"}
|
||||
)
|
||||
headers, _ = AzureAnthropicMessagesConfig().validate_anthropic_messages_environment(
|
||||
headers={"anthropic-beta": f"existing-beta,{beta}" if explicit_beta else "existing-beta"},
|
||||
model="claude-fable-5-1",
|
||||
messages=[{"role": "user", "content": "Hello"}, message],
|
||||
optional_params={"thinking": {"type": "adaptive", "display": "updates"}} if feature == "thinking" else {},
|
||||
litellm_params={"vertex_ai_project": "test-project", "vertex_ai_location": "us-east5"},
|
||||
api_key="test-key",
|
||||
api_base="https://example.com/anthropic",
|
||||
)
|
||||
|
||||
betas: Final = headers["anthropic-beta"].split(",")
|
||||
assert betas.count(beta) == int(feature != "none" or explicit_beta)
|
||||
assert betas.count("existing-beta") == 1
|
||||
|
|
|
|||
|
|
@ -514,3 +514,57 @@ def test_azure_chat_compaction_sends_the_compact_2026_09_04_beta(
|
|||
filtered: Final = update_headers_with_filtered_beta(headers=headers, provider="azure_ai")
|
||||
|
||||
assert ("compact-2026-09-04" in filtered.get("anthropic-beta", "").split(",")) is expected
|
||||
|
||||
|
||||
@pytest.mark.usefixtures("local_model_cost_map")
|
||||
@pytest.mark.parametrize("thinking_type", ("adaptive", "enabled"))
|
||||
@pytest.mark.parametrize("display", ("summarized", "omitted", "updates"))
|
||||
def test_chat_preserves_thinking_display_and_beta(thinking_type: str, display: str) -> None:
|
||||
from typing import Final
|
||||
|
||||
from litellm.llms.azure_ai.anthropic.transformation import AzureAnthropicConfig
|
||||
from litellm.types.llms.anthropic import ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER
|
||||
|
||||
config: Final = AzureAnthropicConfig()
|
||||
optional_params: Final = config.map_openai_params(
|
||||
non_default_params={"thinking": {"type": thinking_type, "display": display, "budget_tokens": 2048}},
|
||||
optional_params={},
|
||||
model="claude-fable-5-1",
|
||||
drop_params=False,
|
||||
)
|
||||
headers: Final = config.validate_environment(
|
||||
headers={"anthropic-beta": "existing-beta"},
|
||||
model="claude-fable-5-1",
|
||||
messages=[{"role": "user", "content": "Reply with OK"}],
|
||||
optional_params=optional_params,
|
||||
litellm_params={},
|
||||
api_key="test-key",
|
||||
)
|
||||
result: Final = config.transform_request(
|
||||
model="claude-fable-5-1",
|
||||
messages=[{"role": "user", "content": "Reply with OK"}],
|
||||
optional_params=optional_params,
|
||||
litellm_params={},
|
||||
headers=headers,
|
||||
)
|
||||
|
||||
assert result["thinking"]["type"] == "adaptive"
|
||||
assert result["thinking"]["display"] == display
|
||||
assert (ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER in headers["anthropic-beta"].split(",")) == (display == "updates")
|
||||
assert "existing-beta" in headers["anthropic-beta"].split(",")
|
||||
|
||||
|
||||
@pytest.mark.parametrize("thinking", ("enabled", {"type": "bogus"}, {"type": "adaptive", "display": "future-mode"}))
|
||||
def test_beta_detection_preserves_unrecognized_thinking(thinking: object) -> None:
|
||||
from typing import Final
|
||||
|
||||
from litellm.types.llms.anthropic import ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER
|
||||
|
||||
params: Final = {"thinking": thinking}
|
||||
headers: Final = AzureAnthropicConfig().validate_environment(
|
||||
headers={}, model="claude-fable-5-1", messages=[], optional_params=params,
|
||||
litellm_params={}, api_key="test-key",
|
||||
)
|
||||
|
||||
assert params["thinking"] == thinking
|
||||
assert ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER not in headers.get("anthropic-beta", "").split(",")
|
||||
|
|
|
|||
|
|
@ -918,3 +918,79 @@ async def test_vertex_messages_request_sends_compact_2026_09_04_beta_on_the_wire
|
|||
assert sent["headers"]["anthropic-beta"].split(",").count("compact-2026-09-04") == 1
|
||||
assert sent_body["compaction"] == {"type": "summarize"}
|
||||
assert "anthropic_beta" not in sent_body
|
||||
|
||||
|
||||
@pytest.mark.usefixtures("local_model_cost_map", "local_beta_headers_config")
|
||||
@pytest.mark.parametrize("feature", ("thinking", "output_config", "tool_addition", "tool_removal", "none"))
|
||||
@pytest.mark.parametrize("explicit_beta", (False, True))
|
||||
def test_messages_feature_beta_headers(feature: str, explicit_beta: bool) -> None:
|
||||
from typing import Final
|
||||
from unittest.mock import patch
|
||||
|
||||
from google.oauth2.credentials import Credentials
|
||||
from litellm.anthropic_beta_headers_manager import update_headers_with_filtered_beta
|
||||
from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.experimental_pass_through.transformation import VertexAIPartnerModelsAnthropicMessagesConfig
|
||||
from litellm.types.llms.anthropic import (
|
||||
ANTHROPIC_BETA_HEADER_VALUES,
|
||||
ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER,
|
||||
ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER,
|
||||
)
|
||||
|
||||
beta: Final = (
|
||||
ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER
|
||||
if feature == "thinking"
|
||||
else ANTHROPIC_BETA_HEADER_VALUES.PER_TURN_CONTROL_2026_07_01.value
|
||||
if feature == "output_config"
|
||||
else ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER
|
||||
)
|
||||
message: Final = (
|
||||
{"role": "system", "content": [], "output_config": {"effort": "high"}}
|
||||
if feature == "output_config"
|
||||
else {"role": "system", "content": [{"type": feature, "tool": {"type": "tool_reference", "name": "ping"}}]}
|
||||
if feature in ("tool_addition", "tool_removal")
|
||||
else {"role": "user", "content": "Reply with OK"}
|
||||
)
|
||||
with patch(
|
||||
"google.auth.default",
|
||||
return_value=(MagicMock(spec=Credentials, token="test-token", expired=False), "test-project"),
|
||||
):
|
||||
headers, _ = VertexAIPartnerModelsAnthropicMessagesConfig().validate_anthropic_messages_environment(
|
||||
headers={"anthropic-beta": f"existing-beta,{beta}" if explicit_beta else "existing-beta"},
|
||||
model="claude-fable-5-1",
|
||||
messages=[{"role": "user", "content": "Hello"}, message],
|
||||
optional_params={"thinking": {"type": "adaptive", "display": "updates"}} if feature == "thinking" else {},
|
||||
litellm_params={"vertex_ai_project": "test-project", "vertex_ai_location": "us-east5"},
|
||||
api_key="test-key",
|
||||
api_base="https://example.com/anthropic",
|
||||
)
|
||||
|
||||
betas: Final = headers["anthropic-beta"].split(",")
|
||||
assert betas.count(beta) == int(feature != "none" or explicit_beta)
|
||||
assert betas.count("existing-beta") == 1
|
||||
|
||||
if feature in ("tool_addition", "tool_removal"):
|
||||
filtered_headers: Final = update_headers_with_filtered_beta(headers=headers, provider="vertex_ai")
|
||||
assert filtered_headers.get("anthropic-beta", "").split(",").count(beta) == 1
|
||||
|
||||
|
||||
@pytest.mark.parametrize("thinking", ("enabled", {"type": "bogus"}, {"type": "adaptive", "display": "future-mode"}))
|
||||
def test_beta_detection_preserves_unrecognized_thinking(thinking: object) -> None:
|
||||
from typing import Final
|
||||
from unittest.mock import patch
|
||||
|
||||
from google.oauth2.credentials import Credentials
|
||||
from litellm.types.llms.anthropic import ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER
|
||||
|
||||
params: Final = {"thinking": thinking}
|
||||
with patch(
|
||||
"google.auth.default",
|
||||
return_value=(MagicMock(spec=Credentials, token="test-token", expired=False), "test-project"),
|
||||
):
|
||||
headers, _ = VertexAIPartnerModelsAnthropicMessagesConfig().validate_anthropic_messages_environment(
|
||||
headers={}, model="claude-fable-5-1", messages=[], optional_params=params,
|
||||
litellm_params={"vertex_ai_project": "test-project", "vertex_ai_location": "us-east5"},
|
||||
api_key="test-key", api_base="https://example.com/anthropic",
|
||||
)
|
||||
|
||||
assert params["thinking"] == thinking
|
||||
assert ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER not in headers.get("anthropic-beta", "").split(",")
|
||||
|
|
|
|||
|
|
@ -1431,3 +1431,51 @@ def test_vertex_ai_anthropic_completion_sends_compact_2026_09_04_beta_on_the_wir
|
|||
assert sent["headers"]["anthropic-beta"].split(",").count("compact-2026-09-04") == 1
|
||||
assert sent_body["compaction"] == {"type": "summarize"}
|
||||
assert "anthropic_beta" not in sent_body
|
||||
|
||||
|
||||
@pytest.mark.usefixtures("local_model_cost_map")
|
||||
@pytest.mark.parametrize("thinking_type", ("adaptive", "enabled"))
|
||||
@pytest.mark.parametrize("display", ("summarized", "omitted", "updates"))
|
||||
def test_chat_preserves_thinking_display_and_beta(thinking_type: str, display: str) -> None:
|
||||
from typing import Final
|
||||
|
||||
from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.transformation import VertexAIAnthropicConfig
|
||||
from litellm.types.llms.anthropic import ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER
|
||||
|
||||
config: Final = VertexAIAnthropicConfig()
|
||||
optional_params: Final = config.map_openai_params(
|
||||
non_default_params={"thinking": {"type": thinking_type, "display": display, "budget_tokens": 2048}},
|
||||
optional_params={},
|
||||
model="claude-fable-5-1",
|
||||
drop_params=False,
|
||||
)
|
||||
optional_params["extra_headers"] = {"anthropic-beta": "existing-beta"}
|
||||
headers: Final = {}
|
||||
result: Final = config.transform_request(
|
||||
model="claude-fable-5-1",
|
||||
messages=[{"role": "user", "content": "Reply with OK"}],
|
||||
optional_params=optional_params,
|
||||
litellm_params={},
|
||||
headers=headers,
|
||||
)
|
||||
|
||||
assert result["thinking"]["type"] == "adaptive"
|
||||
assert result["thinking"]["display"] == display
|
||||
assert (ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER in headers["anthropic-beta"].split(",")) == (display == "updates")
|
||||
assert "existing-beta" in headers["anthropic-beta"].split(",")
|
||||
|
||||
|
||||
@pytest.mark.parametrize("thinking", ("enabled", {"type": "bogus"}, {"type": "adaptive", "display": "future-mode"}))
|
||||
def test_beta_detection_preserves_unrecognized_thinking(thinking: object) -> None:
|
||||
from typing import Final
|
||||
|
||||
from litellm.types.llms.anthropic import ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER
|
||||
|
||||
headers: Final = {}
|
||||
result: Final = VertexAIAnthropicConfig().transform_request(
|
||||
model="claude-fable-5-1", messages=[{"role": "user", "content": "Hello"}],
|
||||
optional_params={"thinking": thinking}, litellm_params={}, headers=headers,
|
||||
)
|
||||
|
||||
assert result["thinking"] == thinking
|
||||
assert ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER not in headers.get("anthropic-beta", "").split(",")
|
||||
|
|
|
|||
|
|
@ -84,6 +84,7 @@ class TestAnthropicBetaHeadersFiltering:
|
|||
beta_headers=all_headers, provider=provider
|
||||
)
|
||||
assert ("compact-2026-09-04" in filtered) is (provider in ("anthropic", "azure_ai", "vertex_ai"))
|
||||
assert ("mid-conversation-tool-changes-2026-07-01" in filtered) is (provider != "bedrock_converse")
|
||||
|
||||
for header in unsupported_headers:
|
||||
assert (
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue