343536373839404142434445464748495051

@dataclassclassRequestUsage:"""Usage details for a single API request."""input_tokens:int"""Input tokens for this individual request."""output_tokens:int"""Output tokens for this individual request."""total_tokens:int"""Total tokens (input + output) for this individual request."""input_tokens_details:InputTokensDetails"""Details about the input tokens for this individual request."""output_tokens_details:OutputTokensDetails"""Details about the output tokens for this individual request."""

 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165

@dataclassclassUsage:requests:int=0"""Total requests made to the LLM API."""input_tokens:int=0"""Total input tokens sent, across all requests."""input_tokens_details:Annotated[InputTokensDetails,BeforeValidator(_normalize_input_tokens_details)]=field(default_factory=lambda:InputTokensDetails(cached_tokens=0))"""Details about the input tokens, matching responses API usage details."""output_tokens:int=0"""Total output tokens received, across all requests."""output_tokens_details:Annotated[OutputTokensDetails,BeforeValidator(_normalize_output_tokens_details)]=field(default_factory=lambda:OutputTokensDetails(reasoning_tokens=0))"""Details about the output tokens, matching responses API usage details."""total_tokens:int=0"""Total tokens sent and received, across all requests."""request_usage_entries:list[RequestUsage]=field(default_factory=list)"""List of RequestUsage entries for accurate per-request cost calculation.    Each call to `add()` automatically creates an entry in this list if the added usage    represents a new request (i.e., has non-zero tokens).    Example:        For a run that makes 3 API calls with 100K, 150K, and 80K input tokens each,        the aggregated `input_tokens` would be 330K, but `request_usage_entries` would        preserve the [100K, 150K, 80K] breakdown, which could be helpful for detailed        cost calculation or context window management.    """def__post_init__(self)->None:# Some providers don't populate optional token detail fields# (cached_tokens, reasoning_tokens), and the OpenAI SDK's generated# code can bypass Pydantic validation (e.g., via model_construct),# allowing None values. We normalize these to 0 to prevent TypeErrors.input_details_none=self.input_tokens_detailsisNoneinput_cached_none=(notinput_details_noneandself.input_tokens_details.cached_tokensisNone)ifinput_details_noneorinput_cached_none:self.input_tokens_details=InputTokensDetails(cached_tokens=0)output_details_none=self.output_tokens_detailsisNoneoutput_reasoning_none=(notoutput_details_noneandself.output_tokens_details.reasoning_tokensisNone)ifoutput_details_noneoroutput_reasoning_none:self.output_tokens_details=OutputTokensDetails(reasoning_tokens=0)defadd(self,other:Usage)->None:"""Add another Usage object to this one, aggregating all fields.        This method automatically preserves request_usage_entries.        Args:            other: The Usage object to add to this one.        """self.requests+=other.requestsifother.requestselse0self.input_tokens+=other.input_tokensifother.input_tokenselse0self.output_tokens+=other.output_tokensifother.output_tokenselse0self.total_tokens+=other.total_tokensifother.total_tokenselse0# Null guards for nested token details (other may bypass validation via model_construct)other_cached=(other.input_tokens_details.cached_tokensifother.input_tokens_detailsandother.input_tokens_details.cached_tokenselse0)other_reasoning=(other.output_tokens_details.reasoning_tokensifother.output_tokens_detailsandother.output_tokens_details.reasoning_tokenselse0)self_cached=(self.input_tokens_details.cached_tokensifself.input_tokens_detailsandself.input_tokens_details.cached_tokenselse0)self_reasoning=(self.output_tokens_details.reasoning_tokensifself.output_tokens_detailsandself.output_tokens_details.reasoning_tokenselse0)self.input_tokens_details=InputTokensDetails(cached_tokens=self_cached+other_cached)self.output_tokens_details=OutputTokensDetails(reasoning_tokens=self_reasoning+other_reasoning)# Automatically preserve request_usage_entries.# If the other Usage represents a single request with tokens, record it.ifother.requests==1andother.total_tokens>0:input_details=other.input_tokens_detailsorInputTokensDetails(cached_tokens=0)output_details=other.output_tokens_detailsorOutputTokensDetails(reasoning_tokens=0)request_usage=RequestUsage(input_tokens=other.input_tokens,output_tokens=other.output_tokens,total_tokens=other.total_tokens,input_tokens_details=input_details,output_tokens_details=output_details,)self.request_usage_entries.append(request_usage)elifother.request_usage_entries:# If the other Usage already has individual request breakdowns, merge them.self.request_usage_entries.extend(other.request_usage_entries)