Pomona Nutrient / pH-EC Reasoner v0.1.1 MLX 8-bit
Experimental runtime artifact. Prepared locally only; not uploaded. This
MLX package is an 8-bit conversion of the merged Pomona Nutrient/pH-EC v0.1.1
correction model, based on Qwen/Qwen2.5-0.5B-Instruct.
The source PEFT adapter passed its independent 140-case semantic evaluation
with valid JSON, allowed labels/actions, nutrient label F1, blocked-action F1,
and human-review match all at 1.0000. Conversion quality must be measured
separately.
Runtime Evaluation
The MLX conversion passed a five-case smoke test for valid JSON, required
fields, allowed values, blocked-action F1, and human-review match. Its nutrient
label F1 was 0.8667, so it is not release-ready and has not passed the
full 140-case holdout. Do not treat this package as equivalent to the source
LoRA.
The Pomona guarded hybrid path applies deterministic rules after model
generation. With that guard enabled, the same 140-case holdout reached
valid JSON, allowed labels/actions, nutrient label F1, blocked-action F1, and
human-review match of 1.0000 in every category. This measures the complete
Pomona deployment path, not standalone MLX reasoning.
Scope And Safety
This is a narrow structured-output advisory reasoner for pH and EC telemetry. It must not change fertigation, inject nutrients, dose chemicals, control actuators, or provide a definitive disease diagnosis. Keep it behind Pomona's deterministic nutrient validation, safety checker, and human approval workflow.
Local Use
Serve locally with MLX-LM:
mlx_lm.server --model pomona-nutrient-ph-ec-reasoner-v0.1.1-8bit --port 8083
Use the exact training prompt and deterministic validation. The package is intended for local Apple Silicon experimentation until a full independent runtime evaluation passes.
For the guarded path, use scripts/models/guard_nutrient_ph_ec_output.py or
the Pomona model-router deterministic route. guarded_evaluation.json records
the guarded result; evaluation.json records the unguarded model-only result.
Related Projects
- Platform and deterministic route: okyanu/pomona
- Base model: Qwen/Qwen2.5-0.5B-Instruct
- Source LoRA:
pomona-nutrient-ph-ec-reasoner-v0.1.1-correction-lora
- Downloads last month
- 10
8-bit
Model tree for Okyanus/pomona-nutrient-ph-ec-reasoner-v0.1.1-MLX
Base model
Qwen/Qwen2.5-0.5B
# Make sure mlx-lm is installed # pip install --upgrade mlx-lm # Generate text with mlx-lm from mlx_lm import load, generate model, tokenizer = load("Okyanus/pomona-nutrient-ph-ec-reasoner-v0.1.1-MLX") prompt = "Write a story about Einstein" messages = [{"role": "user", "content": prompt}] prompt = tokenizer.apply_chat_template( messages, add_generation_prompt=True ) text = generate(model, tokenizer, prompt=prompt, verbose=True)