# Modelfile for ThoxMicro-125M on Ollama.
# Build:   ollama create thox-ai/thox-micro -f Modelfile
# Push:    ollama push thox-ai/thox-micro
# Pull:    ollama pull thox-ai/thox-micro
# Run:     ollama run thox-ai/thox-micro "the THOX agent fleet is"
#
# Requires: thox-micro-125m.q4_k_m.gguf (or another quant) sitting next to this Modelfile.
# Default points at Q4_K_M (~80 MB). Override FROM to use Q5_K_M, Q8_0, or F16.

FROM ./thox-micro-125m.q4_k_m.gguf

# THOX-tuned generation defaults. Conservative for a 125M base model that
# is NOT instruction-tuned. Override at runtime with `ollama run --temp ...`.
PARAMETER temperature  0.7
PARAMETER top_p        0.9
PARAMETER top_k        40
PARAMETER repeat_penalty 1.1
PARAMETER num_ctx      1024
PARAMETER num_predict  128
PARAMETER stop         "<|endoftext|>"
PARAMETER stop         "</s>"

# This model is a pretrained base, not a chat model. We do not ship a chat
# template, on purpose. Wrap with your own system prompt at runtime if you
# want chat behavior; for raw next-token use, leave TEMPLATE off.
TEMPLATE """{{ .Prompt }}"""

SYSTEM """ThoxMicro-125M is a pretrained base model from Thox.ai. It has no instruction tuning. Treat the prompt as the start of a document and the response as the continuation."""

# Provenance and licensing surfaced inside `ollama show`.
LICENSE """Apache-2.0
Copyright 2026 Thox.ai LLC
Inventors: Craig Ross and Phamy Xaypanya
THOX IP-017
Source: https://huggingface.co/thox-ai/thox-micro-125m
Training pipeline: https://github.com/ttracx/thox-llm
"""

MESSAGE system You are reading a pretrained base model. It can complete the prompt but it does not chat.