Spaces:
Sleeping
Sleeping
File size: 3,389 Bytes
d86db02 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 | """
LangSmith observability for the MediShield classification pipeline.
Each pipeline stage is wrapped with @traceable so LangSmith records:
- inputs / outputs per stage
- latency per stage
- token usage for LLM calls
- parent/child span relationships (rules → ocr → llm under classify)
Required env vars:
LANGCHAIN_TRACING_V2=true
LANGCHAIN_API_KEY=<your-langsmith-key>
LANGCHAIN_PROJECT=medishield-classification
"""
import logging
import time
from typing import Any
from langsmith import traceable
logger = logging.getLogger(__name__)
@traceable(name="rules-engine", run_type="tool", tags=["stage:rules"])
def trace_rules_engine(filename: str, doc_type: str | None, send_to_llm: bool) -> dict:
"""Record the rules engine decision in LangSmith."""
return {
"filename": filename,
"doc_type": doc_type,
"send_to_llm": send_to_llm,
"stage": "rules",
}
@traceable(name="kyc-ocr", run_type="tool", tags=["stage:ocr"])
def trace_kyc_ocr(filename: str, doc_type: str | None, send_to_llm: bool, ocr_text: str) -> dict:
"""Record the KYC OCR decision in LangSmith."""
return {
"filename": filename,
"doc_type": doc_type,
"send_to_llm": send_to_llm,
"ocr_text_length": len(ocr_text),
"stage": "ocr",
}
@traceable(name="llm-classify", run_type="llm", tags=["stage:llm"])
def trace_llm_classify(
filename: str,
sub_type: str,
input_tokens: int,
output_tokens: int,
raw_response: str,
) -> dict:
"""Record the LLM classification result and token usage in LangSmith."""
return {
"filename": filename,
"sub_type": sub_type,
"usage": {
"input_tokens": input_tokens,
"output_tokens": output_tokens,
"total_tokens": input_tokens + output_tokens,
},
"raw_response": raw_response,
"stage": "llm",
}
@traceable(name="classify-document", run_type="chain", tags=["pipeline"])
def trace_classify(
filename: str,
doc_type: str,
sub_type: str | None,
method: str,
latency_ms: int,
input_tokens: int,
output_tokens: int,
) -> dict:
"""Top-level trace for a single document classification run."""
result = {
"filename": filename,
"doc_type": doc_type,
"sub_type": sub_type,
"method": method,
"latency_ms": latency_ms,
"input_tokens": input_tokens,
"output_tokens": output_tokens,
}
logger.info(
"classified",
extra={
"filename": filename,
"doc_type": doc_type,
"sub_type": sub_type,
"method": method,
"latency_ms": latency_ms,
"input_tokens": input_tokens,
"output_tokens": output_tokens,
},
)
return result
def record_token_usage(response_usage_metadata: Any) -> dict[str, int]:
"""Extract token counts from a Gemini usage_metadata object."""
if response_usage_metadata is None:
return {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0}
input_t = getattr(response_usage_metadata, "prompt_token_count", 0) or 0
output_t = getattr(response_usage_metadata, "candidates_token_count", 0) or 0
return {
"input_tokens": input_t,
"output_tokens": output_t,
"total_tokens": input_t + output_t,
}
|