diff --git a/litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py b/litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py index 35cb1abbe60..14bd2bee6cf 100644 --- a/litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py +++ b/litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py @@ -770,9 +770,7 @@ class AmazonAnthropicClaudeMessagesConfig( aws_decoder: Final = AmazonAnthropicClaudeMessagesStreamDecoder( model=model, ) - completion_stream: Final = aws_decoder.aiter_bytes( - httpx_response.aiter_bytes() - ) + completion_stream: Final = aws_decoder.aiter_bytes(httpx_response.aiter_bytes()) # Convert decoded Bedrock events to Server-Sent Events expected by Anthropic clients. return self.bedrock_sse_wrapper( completion_stream=completion_stream, diff --git a/tests/test_litellm/llms/bedrock/messages/invoke_transformations/test_anthropic_claude3_transformation.py b/tests/test_litellm/llms/bedrock/messages/invoke_transformations/test_anthropic_claude3_transformation.py index c1db6e25af7..76a50e5165f 100644 --- a/tests/test_litellm/llms/bedrock/messages/invoke_transformations/test_anthropic_claude3_transformation.py +++ b/tests/test_litellm/llms/bedrock/messages/invoke_transformations/test_anthropic_claude3_transformation.py @@ -53,9 +53,7 @@ async def test_bedrock_sse_wrapper_encodes_dict_chunks(): _dummy_stream(), litellm_logging_obj=LiteLLMLoggingObj( model="bedrock/invoke/anthropic.claude-3-sonnet-20240229-v1:0", - messages=[ - {"role": "user", "content": "Hello, can you tell me a short joke?"} - ], + messages=[{"role": "user", "content": "Hello, can you tell me a short joke?"}], stream=True, call_type="chat", start_time=datetime.now(), @@ -232,9 +230,7 @@ async def test_bedrock_sse_wrapper_keeps_usage_in_message_start_and_message_delt def test_chunk_parser_usage_transformation(): """Ensure Bedrock invocation metrics are transformed to Anthropic usage keys.""" - decoder = AmazonAnthropicClaudeMessagesStreamDecoder( - model="bedrock/invoke/anthropic.claude-3-sonnet-20240229-v1:0" - ) + decoder = AmazonAnthropicClaudeMessagesStreamDecoder(model="bedrock/invoke/anthropic.claude-3-sonnet-20240229-v1:0") chunk = { "type": "message_delta", @@ -263,9 +259,7 @@ def test_chunk_parser_preserves_cache_usage_fields_with_invocation_metrics(): fields and cache tokens end up billed at $0. """ - decoder = AmazonAnthropicClaudeMessagesStreamDecoder( - model="bedrock/invoke/anthropic.claude-sonnet-4-6" - ) + decoder = AmazonAnthropicClaudeMessagesStreamDecoder(model="bedrock/invoke/anthropic.claude-sonnet-4-6") chunk = { "type": "message_stop", @@ -291,9 +285,7 @@ def test_chunk_parser_preserves_cache_usage_fields_with_invocation_metrics(): def test_chunk_parser_maps_cache_token_counts_from_invocation_metrics(): """Cache itemization inside invocationMetrics maps to Anthropic usage keys.""" - decoder = AmazonAnthropicClaudeMessagesStreamDecoder( - model="bedrock/invoke/anthropic.claude-sonnet-4-6" - ) + decoder = AmazonAnthropicClaudeMessagesStreamDecoder(model="bedrock/invoke/anthropic.claude-sonnet-4-6") chunk = { "type": "message_stop", @@ -316,9 +308,7 @@ def test_chunk_parser_maps_cache_token_counts_from_invocation_metrics(): def test_chunk_parser_keeps_existing_token_counts_over_invocation_metrics(): """Token counts reported in the chunk's own usage block win over invocationMetrics.""" - decoder = AmazonAnthropicClaudeMessagesStreamDecoder( - model="bedrock/invoke/anthropic.claude-sonnet-4-6" - ) + decoder = AmazonAnthropicClaudeMessagesStreamDecoder(model="bedrock/invoke/anthropic.claude-sonnet-4-6") chunk = { "type": "message_stop", @@ -353,9 +343,7 @@ async def test_bedrock_sse_wrapper_preserves_cache_usage_with_invocation_metrics final usage billed cache reads and writes at $0. """ - decoder = AmazonAnthropicClaudeMessagesStreamDecoder( - model="bedrock/invoke/anthropic.claude-sonnet-4-6" - ) + decoder = AmazonAnthropicClaudeMessagesStreamDecoder(model="bedrock/invoke/anthropic.claude-sonnet-4-6") cfg = AmazonAnthropicClaudeMessagesConfig() raw_chunks = [ @@ -565,11 +553,7 @@ def test_normalize_custom_field_on_tools(): assert request4["tools"] is None # Case 5: an explicit top-level flag wins over a conflicting wrapped one - request5 = { - "tools": [ - {"name": "Read", "defer_loading": False, "custom": {"defer_loading": True}} - ] - } + request5 = {"tools": [{"name": "Read", "defer_loading": False, "custom": {"defer_loading": True}}]} normalize_custom_field_on_tools(request5) assert request5["tools"][0] == {"name": "Read", "defer_loading": False} @@ -590,9 +574,7 @@ def test_normalize_custom_field_on_tools(): assert request7["tools"] == [{"name": "Read"}, {"name": "Write"}] -@pytest.mark.parametrize( - "deferred_marker", [{"custom": {"defer_loading": True}}, {"defer_loading": True}] -) +@pytest.mark.parametrize("deferred_marker", [{"custom": {"defer_loading": True}}, {"defer_loading": True}]) def test_bedrock_invoke_messages_transform_emits_top_level_defer_loading( deferred_marker, ): @@ -725,9 +707,7 @@ def test_bedrock_invoke_messages_skips_thinking_injection_when_already_enabled( "max_tokens": 32000, "stream": False, "thinking": {"type": "enabled", "budget_tokens": 2048}, - "context_management": { - "edits": [{"type": "clear_thinking_20251015", "keep": "all"}] - }, + "context_management": {"edits": [{"type": "clear_thinking_20251015", "keep": "all"}]}, } result = cfg.transform_anthropic_messages_request( model="global.anthropic.claude-sonnet-4-6-v1:0", @@ -829,9 +809,7 @@ def test_remove_ttl_from_cache_control_processes_tools(local_model_cost_map): "messages": [], } - cfg._remove_ttl_from_cache_control( - request, model="anthropic.claude-3-5-sonnet-20241022-v2:0" - ) + cfg._remove_ttl_from_cache_control(request, model="anthropic.claude-3-5-sonnet-20241022-v2:0") # Tool ttl should be stripped assert "ttl" not in request["tools"][0]["cache_control"] @@ -867,9 +845,7 @@ def test_remove_ttl_from_cache_control_preserves_tools_ttl_for_claude_4_5(local_ ], } - cfg._remove_ttl_from_cache_control( - request, model="us.anthropic.claude-sonnet-4-5-20250929-v1:0" - ) + cfg._remove_ttl_from_cache_control(request, model="us.anthropic.claude-sonnet-4-5-20250929-v1:0") # Both tools and system should preserve ttl for Claude 4.5 assert request["tools"][0]["cache_control"]["ttl"] == "1h" @@ -953,9 +929,7 @@ def test_bedrock_messages_strips_output_config(): headers={}, ) - assert "output_config" not in result, ( - "output_config should be stripped for models that don't support it" - ) + assert "output_config" not in result, "output_config should be stripped for models that don't support it" assert result.get("max_tokens") == 4096 @@ -988,9 +962,7 @@ def test_bedrock_messages_preserves_output_config_for_claude_4_6(): headers={}, ) - assert "output_config" in result, ( - "output_config should be preserved for supported models" - ) + assert "output_config" in result, "output_config should be preserved for supported models" assert result["output_config"] == {"effort": "high"} assert result.get("max_tokens") == 4096 @@ -1142,9 +1114,7 @@ def test_bedrock_messages_converts_output_config_format_to_inline_schema(): ("anthropic.claude-opus-4-7", "xhigh"), ], ) -def test_bedrock_messages_normalizes_output_config_effort_for_opus( - model, expected_effort -): +def test_bedrock_messages_normalizes_output_config_effort_for_opus(model, expected_effort): """Bedrock /v1/messages accepts ``xhigh`` and forwards the provider-safe effort.""" from unittest.mock import patch @@ -1202,9 +1172,7 @@ def test_bedrock_messages_does_not_mutate_callers_messages_when_embedding_schema headers={}, ) - assert caller_messages == [ - {"role": "user", "content": [{"type": "text", "text": "Hello"}]} - ] + assert caller_messages == [{"role": "user", "content": [{"type": "text", "text": "Hello"}]}] assert caller_message == { "role": "user", "content": [{"type": "text", "text": "Hello"}], @@ -1520,9 +1488,7 @@ def test_bedrock_messages_strips_context_management(): messages = [{"role": "user", "content": [{"type": "text", "text": "Hello"}]}] optional_params = { "max_tokens": 4096, - "context_management": { - "edits": [{"type": "clear_thinking_20251015", "keep": "all"}] - }, + "context_management": {"edits": [{"type": "clear_thinking_20251015", "keep": "all"}]}, } result = cfg.transform_anthropic_messages_request( @@ -1533,9 +1499,7 @@ def test_bedrock_messages_strips_context_management(): headers={}, ) - assert "context_management" not in result, ( - "context_management should be stripped — Bedrock Invoke rejects it" - ) + assert "context_management" not in result, "context_management should be stripped — Bedrock Invoke rejects it" assert result.get("max_tokens") == 4096 @@ -1660,7 +1624,9 @@ def test_bedrock_messages_allowlist_filters_anthropic_only_fields(): ["dangerous-tool-use-2026-09-03,interleaved-thinking-2025-05-14", "interleaved-thinking-2025-05-14"], ids=["client_sends_beta", "client_omits_beta"], ) -def test_bedrock_messages_forwards_safeguards_with_dangerous_tool_use_beta(local_beta_headers_config, client_beta_header): +def test_bedrock_messages_forwards_safeguards_with_dangerous_tool_use_beta( + local_beta_headers_config, client_beta_header +): """ Claude Code's server-side auto-mode classifier sends `safeguards` alongside the dangerous-tool-use-2026-09-03 beta. Bedrock Invoke accepts the pair, answers @@ -1768,12 +1734,8 @@ def test_bedrock_messages_filters_user_provided_unsupported_beta_header(): ) betas = result.get("anthropic_beta") or [] - assert "advisor-tool-2026-03-01" not in betas, ( - "user-provided beta not in the Bedrock mapping must be dropped" - ) - assert "context-1m-2025-08-07" in betas, ( - "user-provided beta that IS in the Bedrock mapping should survive" - ) + assert "advisor-tool-2026-03-01" not in betas, "user-provided beta not in the Bedrock mapping must be dropped" + assert "context-1m-2025-08-07" in betas, "user-provided beta that IS in the Bedrock mapping should survive" def test_bedrock_messages_renames_user_provided_aliased_beta_header(): @@ -1801,9 +1763,7 @@ def test_bedrock_messages_renames_user_provided_aliased_beta_header(): assert "advanced-tool-use-2025-11-20" not in betas, ( "Anthropic-direct spelling should be rewritten, not forwarded verbatim" ) - assert "tool-search-tool-2025-10-19" in betas, ( - "user-provided beta should be renamed to the Bedrock-side spelling" - ) + assert "tool-search-tool-2025-10-19" in betas, "user-provided beta should be renamed to the Bedrock-side spelling" @pytest.mark.asyncio @@ -2065,9 +2025,7 @@ async def test_unified_bedrock_messages_sse_usage_and_cost_claude_sonnet_46(): "global.anthropic.claude-fable-5", ], ) -def test_bedrock_clear_thinking_injects_adaptive_with_effort_for_adaptive_models( - local_model_cost_map, model -): +def test_bedrock_clear_thinking_injects_adaptive_with_effort_for_adaptive_models(local_model_cost_map, model): """clear_thinking_20251015 without a top-level ``thinking`` field must inject ``thinking.type=adaptive`` plus ``output_config.effort`` on adaptive-thinking models (Opus 4.7/4.8, Fable 5). The legacy ``thinking.type=enabled`` shape is @@ -2077,9 +2035,7 @@ def test_bedrock_clear_thinking_injects_adaptive_with_effort_for_adaptive_models cfg = AmazonAnthropicClaudeMessagesConfig() request = { "max_tokens": 32000, - "context_management": { - "edits": [{"type": "clear_thinking_20251015", "keep": "all"}] - }, + "context_management": {"edits": [{"type": "clear_thinking_20251015", "keep": "all"}]}, } changed = cfg._ensure_thinking_for_clear_thinking_context_management( @@ -2102,9 +2058,7 @@ def test_bedrock_clear_thinking_converts_legacy_enabled_budget_to_effort(): "type": "enabled", "budget_tokens": DEFAULT_REASONING_EFFORT_HIGH_THINKING_BUDGET, }, - "context_management": { - "edits": [{"type": "clear_thinking_20251015", "keep": "all"}] - }, + "context_management": {"edits": [{"type": "clear_thinking_20251015", "keep": "all"}]}, } changed = cfg._ensure_thinking_for_clear_thinking_context_management( @@ -2122,10 +2076,7 @@ def test_resolve_clear_thinking_budget_tokens_honors_explicit_zero(): and only fall back to the minimum when the caller omits the budget.""" cfg = AmazonAnthropicClaudeMessagesConfig() assert cfg._resolve_clear_thinking_budget_tokens(0) == 0 - assert ( - cfg._resolve_clear_thinking_budget_tokens(None) - == BEDROCK_MIN_THINKING_BUDGET_TOKENS - ) + assert cfg._resolve_clear_thinking_budget_tokens(None) == BEDROCK_MIN_THINKING_BUDGET_TOKENS assert cfg._resolve_clear_thinking_budget_tokens(12000) == 12000 @@ -2135,9 +2086,7 @@ def test_bedrock_clear_thinking_keeps_enabled_for_non_adaptive_models(): cfg = AmazonAnthropicClaudeMessagesConfig() request = { "max_tokens": 32000, - "context_management": { - "edits": [{"type": "clear_thinking_20251015", "keep": "all"}] - }, + "context_management": {"edits": [{"type": "clear_thinking_20251015", "keep": "all"}]}, } changed = cfg._ensure_thinking_for_clear_thinking_context_management( @@ -2162,9 +2111,7 @@ def test_bedrock_invoke_transform_emits_adaptive_thinking_for_opus_4_8(): optional_params = { "max_tokens": 32000, "stream": False, - "context_management": { - "edits": [{"type": "clear_thinking_20251015", "keep": "all"}] - }, + "context_management": {"edits": [{"type": "clear_thinking_20251015", "keep": "all"}]}, } result = cfg.transform_anthropic_messages_request( @@ -2201,9 +2148,7 @@ def test_bedrock_invoke_transform_normalizes_system_role_message_into_system(): assert all(m.get("role") != "system" for m in result["messages"]) assert result["messages"] == [{"role": "user", "content": "hi"}] - assert result["system"] == [ - {"type": "text", "text": "You are a careful assistant."} - ] + assert result["system"] == [{"type": "text", "text": "You are a careful assistant."}] def test_bedrock_invoke_transform_merges_system_role_into_existing_system(): @@ -2318,9 +2263,7 @@ def test_bedrock_invoke_transform_keeps_mid_conversation_system_role_in_place(lo ) assert result["messages"] == messages - assert result["system"] == [ - {"type": "text", "text": "Base.", "cache_control": {"type": "ephemeral"}} - ] + assert result["system"] == [{"type": "text", "text": "Base.", "cache_control": {"type": "ephemeral"}}] def test_bedrock_invoke_transform_hoists_only_leading_system_run(local_model_cost_map): @@ -2503,13 +2446,13 @@ def test_bedrock_invoke_transform_converted_system_carries_only_its_content(loca assert result["messages"][2] == { "role": "user", "content": [ - { - "type": "text", - "text": ( - "Operator note (not from the user): the following was " - "originally a mid-conversation system-role reminder." - ), - }, + { + "type": "text", + "text": ( + "Operator note (not from the user): the following was " + "originally a mid-conversation system-role reminder." + ), + }, {"type": "text", "text": "[Truncated: PARTIAL view of big1.txt]"}, ], } @@ -2645,10 +2588,7 @@ def test_as_system_content_blocks_handles_each_shape(): def test_effort_from_thinking_budget_tiers(budget_tokens, expected_effort): """The budget -> effort mapping pins each tier boundary so a shifted threshold is caught.""" - assert ( - AmazonAnthropicClaudeMessagesConfig._effort_from_thinking_budget(budget_tokens) - == expected_effort - ) + assert AmazonAnthropicClaudeMessagesConfig._effort_from_thinking_budget(budget_tokens) == expected_effort def test_inject_adaptive_thinking_preserves_existing_effort(): @@ -2657,9 +2597,7 @@ def test_inject_adaptive_thinking_preserves_existing_effort(): cfg = AmazonAnthropicClaudeMessagesConfig() request = {"output_config": {"effort": "max", "other": "keep"}} - cfg._inject_adaptive_thinking_for_clear_thinking( - request, budget_tokens=24000, model="us.anthropic.claude-fable-5" - ) + cfg._inject_adaptive_thinking_for_clear_thinking(request, budget_tokens=24000, model="us.anthropic.claude-fable-5") assert request["thinking"] == {"type": "adaptive"} assert request["output_config"] == {"effort": "max", "other": "keep"} @@ -2672,9 +2610,7 @@ def test_bedrock_clear_thinking_noops_when_thinking_already_adaptive(): request = { "max_tokens": 32000, "thinking": {"type": "adaptive"}, - "context_management": { - "edits": [{"type": "clear_thinking_20251015", "keep": "all"}] - }, + "context_management": {"edits": [{"type": "clear_thinking_20251015", "keep": "all"}]}, } changed = cfg._ensure_thinking_for_clear_thinking_context_management( @@ -2694,9 +2630,7 @@ def test_bedrock_clear_thinking_replaces_disabled_thinking_on_adaptive_model(): request = { "max_tokens": 32000, "thinking": {"type": "disabled"}, - "context_management": { - "edits": [{"type": "clear_thinking_20251015", "keep": "all"}] - }, + "context_management": {"edits": [{"type": "clear_thinking_20251015", "keep": "all"}]}, } changed = cfg._ensure_thinking_for_clear_thinking_context_management( @@ -2716,9 +2650,7 @@ def test_bedrock_clear_thinking_leaves_enabled_thinking_on_non_adaptive_model(): request = { "max_tokens": 32000, "thinking": {"type": "enabled", "budget_tokens": 8000}, - "context_management": { - "edits": [{"type": "clear_thinking_20251015", "keep": "all"}] - }, + "context_management": {"edits": [{"type": "clear_thinking_20251015", "keep": "all"}]}, } changed = cfg._ensure_thinking_for_clear_thinking_context_management( @@ -2753,9 +2685,7 @@ def test_bedrock_messages_preserves_clear_tool_uses_context_management_and_adds_ messages = [{"role": "user", "content": [{"type": "text", "text": "Hi"}]}] optional_params = { "max_tokens": 4096, - "context_management": { - "edits": [{"type": "clear_tool_uses_20250919"}] - }, + "context_management": {"edits": [{"type": "clear_tool_uses_20250919"}]}, } result = cfg.transform_anthropic_messages_request( @@ -2766,12 +2696,11 @@ def test_bedrock_messages_preserves_clear_tool_uses_context_management_and_adds_ headers={}, ) - assert result.get("context_management") == { - "edits": [{"type": "clear_tool_uses_20250919"}] - }, "clear_tool_uses_20250919 edit must reach Bedrock InvokeModel body" + assert result.get("context_management") == {"edits": [{"type": "clear_tool_uses_20250919"}]}, ( + "clear_tool_uses_20250919 edit must reach Bedrock InvokeModel body" + ) assert "context-management-2025-06-27" in result.get("anthropic_beta", []), ( - "context-management-2025-06-27 beta must reach the InvokeModel body so " - "the tool-call-clearing edit is accepted" + "context-management-2025-06-27 beta must reach the InvokeModel body so the tool-call-clearing edit is accepted" ) @@ -2848,9 +2777,9 @@ def test_bedrock_messages_filters_clear_thinking_keeps_clear_tool_uses( cm = result.get("context_management") assert cm is not None - assert [e.get("type") for e in cm["edits"]] == [ - "clear_tool_uses_20250919" - ], "clear_thinking_20251015 must still be stripped (LiteLLM-internal)" + assert [e.get("type") for e in cm["edits"]] == ["clear_tool_uses_20250919"], ( + "clear_thinking_20251015 must still be stripped (LiteLLM-internal)" + ) betas = result.get("anthropic_beta", []) assert "context-management-2025-06-27" in betas @@ -2991,9 +2920,7 @@ def test_bedrock_messages_tool_search_follows_claude_tool_search_rule(local_mode assert cfg._supports_tool_search_on_bedrock(model) is expected -def test_bedrock_messages_thinking_shape_follows_exact_bedrock_entry_flag( - local_model_cost_map, monkeypatch -): +def test_bedrock_messages_thinking_shape_follows_exact_bedrock_entry_flag(local_model_cost_map, monkeypatch): """The outbound thinking payload must follow the exact Bedrock cost-map entry. Before threading the caller's provider through the capability probes, the probe was pinned to ``"anthropic"``: the exact ``global.anthropic.claude-opus-4-8`` @@ -3403,22 +3330,14 @@ def test_bedrock_invoke_eager_input_streaming_beta_not_duplicated_with_client_he def _bedrock_event_frame(payload: dict) -> bytes: def _header(name: str, value: str) -> bytes: - return ( - bytes([len(name)]) - + name.encode() - + bytes([7]) - + struct.pack(">H", len(value)) - + value.encode() - ) + return bytes([len(name)]) + name.encode() + bytes([7]) + struct.pack(">H", len(value)) + value.encode() headers: Final = ( _header(":message-type", "event") + _header(":event-type", "chunk") + _header(":content-type", "application/json") ) - body: Final = json.dumps( - {"bytes": base64.b64encode(json.dumps(payload).encode()).decode()} - ).encode() + body: Final = json.dumps({"bytes": base64.b64encode(json.dumps(payload).encode()).decode()}).encode() prelude: Final = struct.pack(">II", 12 + len(headers) + len(body) + 4, len(headers)) prelude_crc: Final = struct.pack(">I", zlib.crc32(prelude)) message_crc: Final = struct.pack(">I", zlib.crc32(prelude + prelude_crc + headers + body))