Calculated available context size -N was not non-negative

TL;DR

Fix: Collaborator logan-markewich diagnosed: the 'Calculated available context size -N' error appears when a custom LLM does not declare an accurate contextwindow in its LLMMetadata - set contextwindow to the real model size (e.g.

The error

Issue run-llama/llama_index#16235 (closed, 16 comments): ### Question Validation - [X] I have searched both the documentation and discord for an answer. ### Question Hi, I'm having this issue whenever I reach to the point that my chat exceeds X amount of tokens. I was assuming `as_chat_engine()` set up with `max_prompt` which preserves chat history, would automatically handle this situation. What am I missing ?

Fix

1. Collaborator logan-markewich diagnosed: the 'Calculated available context size -N' error appears when a custom LLM does not declare an accurate contextwindow in its LLMMetadata - set contextwindow to the real model size (e.g.

Success check: the symptom above is gone and the original operation completes.

When to use

• You hit this exact symptom in llama_index.
• You want the verified fix without digging through the thread.

When not to use

• A different error in llama_index, even a similar-looking one.
• General llama_index setup or credential questions.

Compatibility

• Llama Index (run-llama/llama_index)

Variant phrasings

• How to fix "calculated available context size -N was not non-negative" in llama_index
• llama_index: calculated available context size -N was not non-negative

Edge cases

• If the chat history still overflows, lower the ChatMemoryBuffer token limit, e.g. aschatengine(..., memory=ChatMemoryBuffer.fromdefaults(tokenlimit=(50000)).

Source

• https://github.com/run-llama/llama_index/issues/16235

In gh:run-llama/llamaindex#16235 (https://github.com/run-llama/llamaindex/issues/16235), someone reported: Issue run-llama/llamaindex#16235 (closed, 16 comments): ### Question Validation - [X] I have searched both the documentation and discord for an answer. ### Question Hi, I'm having this issue whenever I reach to the point that my chat exceeds X amount of tokens. I was assuming `aschatengine() set up with maxprompt` which preserves chat history, would automatically handle this situation. What am I missing ? The verified resolution from the thread: Collaborator logan-markewich diagnosed: the 'Calculated available context size -N' error appears when a custom LLM does not declare an accurate contextwindow in its LLMMetadata - set contextwindow to the real model size (e.g. LLMMetadata(contextwindow=128000, modelname=...)) and numoutput. If the chat history still overflows, lower the ChatMemoryBuffer token limit, e.g. aschatengine(..., memory=ChatMemoryBuffer.fromdefaults(token_limit=50000)). The chat engine cannot auto-truncate when it doesn't know what the real window is.