14 lines
574 B
Plaintext
14 lines
574 B
Plaintext
# Optional Ollama Modelfile for building a pre-tuned Marcus VL model.
|
|
# Not used at runtime — Marcus loads the stock `qwen2.5vl:3b` tag directly
|
|
# via Config/config_Brain.json::ollama_model. Keep this file only if you
|
|
# want to build a custom image with `ollama create marcus-vl -f Modelfile`.
|
|
#
|
|
# Runtime parameters (num_batch, num_ctx, num_predict) are overridden on
|
|
# every call by API/llava_api.py, so the PARAMETER lines below are just
|
|
# defaults for `ollama run` shell use.
|
|
|
|
FROM qwen2.5vl:3b
|
|
PARAMETER num_ctx 2048
|
|
PARAMETER num_predict 120
|
|
PARAMETER num_batch 128
|