Fixes#15263
This PR fixes the cost calculation for Bedrock Anthropic models with prompt caching.
**Root Cause:**
PR #9838 incorrectly removed adding `cacheWriteInputTokens` to `prompt_tokens`
for Bedrock, based on the assumption that it would cause double counting (similar
to an Anthropic API issue). However, Bedrock's token structure is different:
- **Bedrock API**: `inputTokens`, `cacheReadInputTokens`, and `cacheWriteInputTokens`
are ALL separate values that should be summed for total input tokens
- **Anthropic API**: Same structure - all three token types are separate
The fix in #9838 was later reverted for Anthropic (correctly re-adding
`cache_creation_input_tokens` to `prompt_tokens`), but Bedrock was never fixed.
**Changes:**
1. Re-add `cacheWriteInputTokens` to `input_tokens` in Bedrock transformation
2. Update test assertions to reflect correct behavior
3. Add regression test for prompt caching cost calculation
4. Fix typo in Anthropic transformation where `cache_creation_tokens` was
incorrectly set to `cache_read_input_tokens`
**Testing:**
- All existing Bedrock transformation tests pass
- New test validates correct cost calculation with prompt caching
- Verified costs are non-negative and accurate
* add supported_db_objects
* add _should_load_db_object
* add docs on storing MCP objects in DB
* test_should_load_db_object_with_supported_db_objects
* type fix
* perf(router): add model_name index for O(1) deployment lookups
Add model_name_to_deployment_indices mapping to optimize _get_all_deployments()
from O(n) to O(1) + O(k) lookups.
- Add model_name_to_deployment_indices: Dict[str, List[int]]
- Add _build_model_name_index() to build/maintain the index
- Update _add_model_to_list_and_index_map() to maintain both indices
- Refactor to use idx = len(self.model_list) before append (cleaner)
- Optimize _get_all_deployments() to use index instead of linear scan
* test(router): add test coverage for _build_model_name_index
Add single comprehensive test for _build_model_name_index() function to fix
code coverage CI failure.
The test verifies:
- Index correctly maps model_name to deployment indices
- Handles multiple deployments per model_name
- Clears and rebuilds index correctly
Fixes: CI code coverage error for _build_model_name_index
Add support for function calling (tools) with Snowflake Cortex models that support it (e.g., Claude 3.5 Sonnet).
Changes:
- Add 'tools' and 'tool_choice' to supported OpenAI parameters
- Implement request transformation: OpenAI function format → Snowflake tool_spec format
- Implement response transformation: Snowflake content_list with tool_use → OpenAI tool_calls
- Add tool_choice transformation: OpenAI nested format → Snowflake array format
Request transformation:
- Transform tools from nested {"type": "function", "function": {...}} to Snowflake's {"tool_spec": {"type": "generic", "name": "...", "input_schema": {...}}}
- Transform tool_choice from {"type": "function", "function": {"name": "..."}} to {"type": "tool", "name": ["..."]}
Response transformation:
- Parse Snowflake's content_list array containing tool_use objects
- Extract tool calls with tool_use_id, name, and input
- Convert to OpenAI's tool_calls format with proper JSON serialization
Testing:
- Add 7 unit tests covering request/response transformations
- Add integration test for Responses API with tool calling
- All tests passing
Fixes issue #15218🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Tim Elfrink