Viewing trace details and latency in LangChain - Step-by-Step Execution

Concept Flow - Viewing trace details and latency

Start LangChain Call

↓

Initialize Trace

↓

Execute Chain Step

↓

Record Step Details & Latency

↓

Check More Steps?

Yes→Execute Chain Step

No↓

Compile Trace Summary

↓

Return Trace Data & Latency Info

↓

End

Shows how LangChain starts a call, records each step's details and latency, then compiles and returns the full trace.

Execution Sample

LangChain

from langchain.chains import LLMChain
from langchain.callbacks import get_openai_callback

chain = LLMChain(...)  # Initialize your chain here

with get_openai_callback() as cb:
    result = chain.run("Hello")
    print(cb)

Runs a LangChain chain with a callback that tracks and prints trace details and latency.

Execution Table

Step	Action	Trace Detail Recorded	Latency (ms)	Notes
1	Start chain run	Trace initialized	0	Begin tracking
2	Call LLM with input 'Hello'	Input recorded	120	LLM processing time
3	Receive LLM output	Output recorded	5	Output received
4	Post-process output	Post-processing step recorded	10	Formatting output
5	End chain run	Total latency calculated	135	Sum of all steps
6	Print callback info	Trace summary printed	0	Shows tokens used and latency

💡 Chain run completes after all steps, total latency 135 ms recorded

Variable Tracker

Variable	Start	After Step 2	After Step 3	After Step 4	Final
trace_details	{}	{"input": "Hello"}	{"output": "Hi!"}	{"post_process": "Formatted"}	{"input": "Hello", "output": "Hi!", "post_process": "Formatted"}
latency_ms	0	120	125	135	135

Key Moments - 2 Insights

Why does latency increase after each step in the execution_table?

Each step adds its processing time to the total latency, as shown in rows 2 to 5 where times accumulate.

What does the trace_details variable hold after step 4?

It holds all recorded data from input, output, and post-processing steps, as seen in variable_tracker after step 4.

Visual Quiz - 3 Questions

Look at the execution_table at step 3, what trace detail is recorded?

A Input to the LLM

B Post-processing details

C Output from the LLM

D Total latency

💡 Hint

Check the 'Trace Detail Recorded' column at step 3 in execution_table

At which step does the total latency get calculated?

A Step 3

B Step 5

C Step 2

D Step 6

💡 Hint

Look for 'Total latency calculated' in the 'Trace Detail Recorded' column

If the LLM call latency increased to 200 ms, how would latency_ms change after step 2?

A It would increase to 200 ms

B It would decrease

C It would stay 120 ms

D It would reset to 0

💡 Hint

Refer to latency_ms values in variable_tracker after step 2

Concept Snapshot

LangChain tracing captures each step's input, output, and processing time.
Use callbacks like get_openai_callback() to record trace details.
Latency accumulates as each step runs.
Trace data helps debug and optimize chain performance.
Print callback info to see tokens used and total latency.

Full Transcript

This visual trace shows how LangChain records detailed trace information and latency during a chain run. The process starts by initializing the trace, then executing each step such as calling the language model, receiving output, and post-processing. Each step's details and latency are recorded and accumulated. After all steps complete, the total latency is calculated and the trace summary is printed. Variables like trace_details hold the input, output, and processing info, while latency_ms tracks the time spent. This helps learners see how LangChain tracks performance and data flow step-by-step.