# Ollama Modelfile for Nemotron-3-Nano-Omni-30B-A3B-Reasoning (text-only GGUF)
#
# Substitute FROM with whichever quant you've downloaded.
# Build:  ollama create nemotron-3-nano-omni-text:Q4_K_M -f Modelfile

FROM ./Nemotron-3-Nano-Omni-30B-A3B-Reasoning-text-only.Q4_K_M.gguf

# Go-template equivalent of the model's native Jinja chat template.
# The Jinja template embedded in the GGUF includes multimodal helpers Ollama can't parse;
# this pares it to the text-only chat structure that matches the model's training format.
TEMPLATE """{{- if .System }}<|im_start|>system
{{ .System }}<|im_end|>
{{ end }}
{{- range .Messages }}<|im_start|>{{ .Role }}
{{ .Content }}<|im_end|>
{{ end }}<|im_start|>assistant
<think></think>"""

PARAMETER num_ctx 32768
PARAMETER temperature 0.6
PARAMETER top_p 0.95
PARAMETER repeat_penalty 1.05
PARAMETER stop "<|im_end|>"
PARAMETER stop "<|endoftext|>"

# Notes:
# - The TEMPLATE here suppresses thinking by default (uses <think></think> empty block).
#   To enable thinking, replace the trailing `<think></think>` with `<think>` (model fills in).
# - To suppress thinking via API instead: pass "think": false at the TOP LEVEL of /api/chat (NOT in options).