OLLAMA_MODEL_NOT_FOUND info Ollama model not found locally
The model tag hasn't been pulled yet, or the tag name doesn't match the registry exactly.
What’s actually happening
This is almost always one of two causes: the model hasn’t been pulled yet, or the tag name in your request doesn’t exactly match what’s in Ollama’s local registry (including the quantization suffix, which is part of the tag, not a separate parameter).
Check what’s actually pulled
ollama list
Compare the exact tag shown here against what your request is asking
for. llama3:8b and llama3:8b-instruct-q4_0 are different tags —
requesting one when you’ve only pulled the other produces this error,
even though “the model” is technically present.
Fix
ollama pull llama3:8b-instruct-q4_0
Pull the exact tag your application requests. If you’re unsure of the
exact tag format for a model, check the model’s page on Ollama’s
registry (ollama.com/library) — tag naming conventions (quantization
suffix format, instruct vs. base model naming) vary by model family
and aren’t fully standardized across them.
If this happens inside a script or CI pipeline
Add an explicit pull step before the run step, rather than relying on Ollama’s automatic pull-on-first-use behavior — that behavior exists for interactive use, but in an unattended pipeline it turns a slow first run into an unexplained hang or timeout instead of a clear “pulling model” log line.
ollama pull llama3:8b-instruct-q4_0 && \
ollama run llama3:8b-instruct-q4_0 "your prompt here"
Frequently asked questions
I pulled the model, why does Ollama still say it's not found?
The tag you requested probably doesn't exactly match what you pulled — check ollama list and compare tags exactly, including the quantization suffix, since llama3:8b and llama3:8b-instruct-q4_0 are treated as different models.
Why does this only show up in my CI pipeline, not locally?
Locally, Ollama's automatic pull-on-first-use silently downloads a missing model. In an unattended pipeline that same behavior turns into a hang or timeout instead of a clear error — add an explicit ollama pull step before your run step to avoid this.