Skip to content

Fix final prompt token accounting #39

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

Merged
merged 1 commit into from
Jul 21, 2025
Merged

Fix final prompt token accounting #39

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: 24 additions & 8 deletions middleware/quota.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -359,11 +359,21 @@ async def finalize() -> None:
model = usage.get("model", "unknown")
parsed: dict | None = None
if resp_is_text:
# -- Robustly extract the last JSON object -----------------
text_tail = streamer.get_tail().decode("utf-8", "ignore")
idx = max(text_tail.rfind("{"), text_tail.rfind('{"'))
if idx != -1:
# 1. Split SSE frames if present: keep only the part after the final "data: "
if "data:" in text_tail:
*_, last_frame = text_tail.strip().split("data:")
text_tail = last_frame.strip()
# 2. Strip the trailing '[DONE]' token if it exists
if text_tail.endswith("[DONE]"):
text_tail = text_tail[: text_tail.rfind("[DONE]")].rstrip()
# 3. Find the first '{' from the *left* (because the frame has been cleaned)
brace = text_tail.find("{")
parsed = None
if brace != -1:
try:
parsed = json.loads(text_tail[idx:])
parsed = json.loads(text_tail[brace:])
except Exception:
parsed = None
if isinstance(parsed, dict):
Expand All @@ -372,9 +382,10 @@ async def finalize() -> None:
u = parsed.get("usage") or {}
tokens_out = int(u.get("completion_tokens", 0))
prompt_tokens = int(u.get("prompt_tokens", tokens_in))
delta_prompt = prompt_tokens - tokens_in
tokens_in = prompt_tokens
delta_prompt = prompt_tokens - tokens_in # adjust quota window
tokens_in = prompt_tokens # ← canonical value
usage["tokens_in"] = tokens_in
# Replace provisional count with canonical one in the meter
await self.store.adjust(user, delta_prompt)
elif "choices" in parsed:
msgs = [
Expand All @@ -390,15 +401,20 @@ async def finalize() -> None:
usage["tokens_out"] = tokens_out
usage["model"] = model

# All *out* tokens are new – add them once.
await self.store.adjust(user, tokens_out)

total = await self.store.peek_total(user)
if total > self.max_tokens:
usage["detail"] = "token quota exceeded post-stream"

response.headers["x-llm-model"] = model
response.headers["x-tokens-in"] = str(tokens_in)
response.headers["x-tokens-out"] = str(tokens_out)
response.headers.update(
{
"x-llm-model": model,
"x-tokens-in": str(tokens_in),
"x-tokens-out": str(tokens_out),
}
)

usage["ts"] = time.time()
await request.app.state.usage.record(**usage)
Expand Down