github/AutoGPT
1
1
Fork 0
mirror of https://github.com/Significant-Gravitas/AutoGPT.git synced 2026-04-08 03:00:28 -04:00
Code Issues Packages Projects Releases Wiki Activity

refactor(copilot): use tiktoken for fallback token estimation

Browse Source 
Replace rough chars/4 heuristic with proper tiktoken tokenizer via
estimate_token_count/estimate_token_count_str from backend.util.prompt.
This commit is contained in:
Zamil Majdy
2026-03-13 05:24:53 +07:00
parent 4ceb15b3f1
commit 976443bf6e
1 changed files with 13 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
autogpt_platform/backend/backend/copilot/baseline/service.py
View File

@@ -430,16 +430,23 @@ async def stream_chat_completion_baseline(
except Exception:
logger.warning("[Baseline] Langfuse trace context teardown failed")
# Fallback: estimate tokens from text length when the provider
# does not honour stream_options={"include_usage": True}.
# Fallback: estimate tokens via tiktoken when the provider does
# not honour stream_options={"include_usage": True}.
# Only count messages added during *this* turn (user message +
# tool rounds), not the full conversation history.
# Rough estimate: 1 token ≈ 4 characters.
if turn_prompt_tokens == 0 and turn_completion_tokens == 0:
from backend.util.prompt import (
estimate_token_count,
estimate_token_count_str,
)
turn_messages = openai_messages[_msgs_before_turn - 1 :]
prompt_chars = sum(len(m.get("content") or "") for m in turn_messages)
turn_prompt_tokens = max(prompt_chars // 4, 1)
turn_completion_tokens = max(len(assistant_text) // 4, 1)
turn_prompt_tokens = max(
estimate_token_count(turn_messages, model=config.model), 1
)
turn_completion_tokens = max(
estimate_token_count_str(assistant_text, model=config.model), 1
)
logger.info(
"[Baseline] No streaming usage reported; estimated tokens: "
"prompt=%d, completion=%d",