Skip to content

fix(anthropic): ensure read/write cache tokens are recorded with streaming methods #2234

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix(anthropic): ensure read/write cache tokens are recorded with streaming methods #2234

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 30 additions & 2 deletions ...nference-instrumentation-anthropic/src/openinference/instrumentation/anthropic/_stream.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -249,6 +249,8 @@ def __init__(self) -> None:
),
stop_reason=_SimpleStringReplace(),
input_tokens=_SimpleStringReplace(),
cache_creation_input_tokens=_SimpleStringReplace(),
cache_read_input_tokens=_SimpleStringReplace(),
output_tokens=_SimpleStringReplace(),
),
),
Expand All @@ -272,6 +274,13 @@ def process_chunk(self, chunk: "RawContentBlockDeltaEvent") -> None:
"input_tokens": str(chunk.message.usage.input_tokens),
}
}
# Capture Anthropic prompt cache usage details when present
cache_write = getattr(chunk.message.usage, "cache_creation_input_tokens", None)
cache_read = getattr(chunk.message.usage, "cache_read_input_tokens", None)
if cache_write is not None:
value["messages"]["cache_creation_input_tokens"] = str(cache_write)
if cache_read is not None:
value["messages"]["cache_read_input_tokens"] = str(cache_read)
self._values += value
elif isinstance(chunk, RawContentBlockStartEvent):
self._current_content_block_type = chunk.content_block
Expand Down Expand Up @@ -341,6 +350,8 @@ def get_extra_attributes(self) -> Iterator[Tuple[str, AttributeValue]]:
idx = 0
total_completion_token_count = 0
total_prompt_token_count = 0
total_cache_read_token_count = 0
total_cache_write_token_count = 0
# TODO(harrison): figure out if we should always assume messages is 1.
# The current non streaming implementation assumes the same
for message in messages:
Expand All @@ -353,6 +364,10 @@ def get_extra_attributes(self) -> Iterator[Tuple[str, AttributeValue]]:
total_completion_token_count += int(output_tokens)
if input_tokens := message.get("input_tokens"):
total_prompt_token_count += int(input_tokens)
if cache_read_tokens := message.get("cache_read_input_tokens"):
total_cache_read_token_count += int(cache_read_tokens)
if cache_write_tokens := message.get("cache_creation_input_tokens"):
total_cache_write_token_count += int(cache_write_tokens)

# TODO(harrison): figure out if we should always assume the first message
# will always be a message output generally this block feels really
Expand All @@ -377,10 +392,23 @@ def get_extra_attributes(self) -> Iterator[Tuple[str, AttributeValue]]:
tool_idx += 1
idx += 1
yield SpanAttributes.LLM_TOKEN_COUNT_COMPLETION, total_completion_token_count
yield SpanAttributes.LLM_TOKEN_COUNT_PROMPT, total_prompt_token_count
total_prompt_with_cache = (
total_prompt_token_count + total_cache_read_token_count + total_cache_write_token_count
)
yield SpanAttributes.LLM_TOKEN_COUNT_PROMPT, total_prompt_with_cache
if total_cache_read_token_count:
yield (
SpanAttributes.LLM_TOKEN_COUNT_PROMPT_DETAILS_CACHE_READ,
total_cache_read_token_count,
)
if total_cache_write_token_count:
yield (
SpanAttributes.LLM_TOKEN_COUNT_PROMPT_DETAILS_CACHE_WRITE,
total_cache_write_token_count,
)
yield (
SpanAttributes.LLM_TOKEN_COUNT_TOTAL,
total_completion_token_count + total_prompt_token_count,
total_completion_token_count + total_prompt_with_cache,
)


Expand Down
Loading