---
language:
- en
- de
license: apache-2.0
base_model: allenai/Olmo-3-7B-Instruct
tags:
- compound-ai
- planner
- orchestrator
- moe
- gguf
- lora
- lumi-g
- moe-sovereign
- open-source
pipeline_tag: text-generation
library_name: transformers
---

# MoE Sovereign Planner 7B -- OLMo-3 (`moe-sovereign-planner-olmo3-7b`)
*Task Decomposition & Orchestration*

[![License: Apache 2.0](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
[![Base Model: OLMo-3-7B-Instruct](https://img.shields.io/badge/Base_Model-OLMo--3--7B--Instruct-violet.svg)](https://huggingface.co/allenai/Olmo-3-7B-Instruct)

---

## Model Summary

`moe-sovereign-planner-olmo3-7b` is a LoRA fine-tune of **OLMo-3-7B-Instruct**, specialized as the orchestrator/planner of the MoE Sovereign compound-AI system: it decomposes an incoming request into 1-4 subtasks for the domain experts, extracting and propagating explicit numerical constraints so experts cannot hallucinate default values.

Genuinely open-source base (Ai2 -- weights, training data, and training code all publicly documented), the Spur-2 (open-source) counterpart to the parallel Qwen3.5-9B planner trained on the open-weight track.

## Training Configuration

| Parameter | Value |
|---|---|
| Method | LoRA (rank 16, alpha 32, dropout 0.05) |
| Trainable parameters | 39,976,960 of 7,337,988,096 (0.54%) |
| Epochs | 3 |
| Effective batch size | 128 (micro-batch 4 x 8 GPUs x grad-accum 4) |
| Learning rate | 1.5e-5 |
| Training sequence length | 4,096 tokens |
| Compute | EuroHPC LUMI-G, 8x AMD Instinct MI250X GCDs, ROCm |
| Training examples | 4,726 curated decomposition examples |

### Observed Training Trajectory

Training loss: 2.111 -> 1.094 -> 0.737. Smooth decline, no overfitting signature.

## Prompt Format

ChatML. System prompt:

```
You are the orchestrator of a Mixture-of-Experts system.
Decompose the following request into 1-4 subtasks.

Mandatorily extract all numerical constraints and technical parameters from the request (e.g. model sizes, MTU values, protocol overheads, chemical doses, bitrates). Integrate these as IMMUTABLE_CONSTANTS directly into each subtask description for the experts, so experts cannot hallucinate default values.
```

## Available Formats

| File | Notes |
|---|---|
| `moe-sovereign-planner-olmo3-7b-Q4_K_M.gguf` | Recommended for deployment |
| `moe-sovereign-planner-olmo3-7b-Q8_0.gguf` | Higher-fidelity reference quantization |

## Hardware Guidance

OLMo-3-7B's native context is 65,536 tokens (via YaRN extension). Fits comfortably on a single modern 12GB+ GPU at Q4_K_M.

### Ollama `Modelfile`

```dockerfile
FROM ./moe-sovereign-planner-olmo3-7b-Q4_K_M.gguf
SYSTEM """You are the orchestrator of a Mixture-of-Experts system.
Decompose the following request into 1-4 subtasks.

Mandatorily extract all numerical constraints and technical parameters from the request (e.g. model sizes, MTU values, protocol overheads, chemical doses, bitrates). Integrate these as IMMUTABLE_CONSTANTS directly into each subtask description for the experts, so experts cannot hallucinate default values."""
TEMPLATE """{{ if .System }}<|im_start|>system
{{ .System }}<|im_end|>
{{ end }}{{ if .Prompt }}<|im_start|>user
{{ .Prompt }}<|im_end|>
{{ end }}<|im_start|>assistant
{{ .Response }}<|im_end|>"""
PARAMETER stop "<|im_end|>"
PARAMETER temperature 0.2
PARAMETER num_ctx 32768
```

## Limitations

- Decomposition quality depends on the request containing extractable constraints; ambiguous requests may yield underspecified subtasks.
- Does not execute the subtasks itself -- routes to the domain experts.

## License

Apache 2.0, inherited from the OLMo-3-7B-Instruct base model.