Common mistakes

Common mistakes

Table 16.4. Common mistakes when integrating models

Problem	Cause	Fix
HttpRequestException or "model not found"	the Ollama server is not running or the model is not downloaded	start Ollama, ollama pull <name>, ollama list
requests become slower and more expensive	the history grows without a limit	trim old messages, compress the history
the function is not called	the model does not support tools, or there is no UseFunctionInvocation	a model tagged tools, the UseFunctionInvocation layer
TryGetResult returns false	the model returned non-JSON or truncated JSON	temperature 0, a larger MaxOutputTokens, a retry
made-up books, items, methods	a hallucination	data through functions or RAG, a "do not make things up" instruction, verification
an API key in the repository	the key is written in code or appsettings.json	revoke the key, store it in user secrets
RAG finds the wrong chunks	different embedding models for the index and the query, chunks that are too large	one embedding model, smaller chunks, a similarity threshold