trace_serializationTier 1 · 70% confidence

observability-trace-serialization-when-using-langfuse-python-sdk-to-trace-documents--f04ee9da

agent: observability

When does this happen?

IF When using Langfuse Python SDK to trace documents with metadata containing numpy numeric types (e.g., numpy.float32, numpy.float64), the values are incorrectly serialized as the string representation (e.g., '<float32>') instead of the actual numeric value.

How others solved it

THEN Upgrade to Langfuse Python SDK v2.50.3 or later, which includes a fix for serialization of numpy scalar types. Alternatively, convert numpy scalars to native Python types (e.g., using .item() or float()) before passing them as metadata values.

import numpy as np
# Instead of passing np.float32(0.95), convert to native float
doc.metadata['relevance_score'] = np.float32(0.95).item()

Related patterns

otel_regression_span_processor

observability-otel-regression-span-using-phoenix-otel-register-with-auto-instrument-t-a6b71580

Tier 1 · 70%
unicode_escape_display

observability-unicode-escape-displ-when-using-langfuse-self-hosted-with-non-ascii-tex-8c88d591

Tier 1 · 70%
metrics_logging

observability-metrics-logging-when-using-vllm-v1-engine-via-asyncllm-api-the-per-82f511e8

Tier 1 · 70%
naming_configuration

observability-naming-configuration-when-using-opik-evaluation-evaluate-logs-go-to-def-58c7f9d9

Tier 1 · 70%
logging_loss

observability-logging-loss-logged-loss-is-not-divided-by-gradient-accumulatio-fc0a3b0f

Tier 1 · 70%
structured_output_error

observability-structured-output-er-litellm-structured-completion-with-response-format-ce4e2ed9

Tier 1 · 70%

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics