h3rb3rn/smollm3-expert-omni-3b

MoE Sovereign Cross-Domain Synthesis Expert 3B -- SmolLM3 (smollm3-expert-omni-3b)

 

Model Summary

smollm3-expert-omni-3b is a LoRA fine-tune of the text-decoder of Qwen3.5-4B, specialized as the omni domain expert within the MoE Sovereign compound-AI system.

You are the cross-domain synthesis and interface-harmonization expert (moe-expert-omni-4b). Reconcile and integrate outputs from other specialists (coder, security, data infrastructure, governance, precision, graph retrieval) into one coherent, consistent result. Do not attempt deep mathematical proofs, raw kernel/driver code, or standalone regulatory audits yourself -- state explicitly when a dedicated specialist is actually required instead.

Base Architecture

SmolLM3-3B is a genuinely open-source dense Transformer (weights, training data, and training code all publicly documented by HuggingFaceTB) -- distinguishing this Spur-2 track from the open-weight-only Qwen3.5 base used in the parallel 4B expert line.

Training Configuration

Observed Training Trajectory

Training loss over the run (representative logged steps): 2.303 -> 1.475 -> 1.444. Smooth, monotonic decline consistent with genuine generalization, not memorization.

Prompt Format

ChatML:

<|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
{user_message}<|im_end|>
<|im_start|>assistant
{response}<|im_end|>

System Prompt

You are the cross-domain synthesis and interface-harmonization expert (moe-expert-omni-4b). Reconcile and integrate outputs from other specialists (coder, security, data infrastructure, governance, precision, graph retrieval) into one coherent, consistent result. Do not attempt deep mathematical proofs, raw kernel/driver code, or standalone regulatory audits yourself -- state explicitly when a dedicated specialist is actually required instead.

Available Formats

Hardware Guidance

SmolLM3-3B's native context window is 65,536 tokens. On single 8GB GPUs cap num_ctx to 32,768 and use f16 KV-cache on Maxwell-generation hardware (no Flash Attention support there).

Ollama Modelfile

FROM ./smollm3-expert-omni-3b-Q4_K_M.gguf
SYSTEM """You are the cross-domain synthesis and interface-harmonization expert (moe-expert-omni-4b). Reconcile and integrate outputs from other specialists (coder, security, data infrastructure, governance, precision, graph retrieval) into one coherent, consistent result. Do not attempt deep mathematical proofs, raw kernel/driver code, or standalone regulatory audits yourself -- state explicitly when a dedicated specialist is actually required instead."""
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

• —Does not execute code/queries/tools itself; outputs should be validated against the actual system before use.
• —Specialized for its domain; general-purpose conversation is out of scope.

License

Apache 2.0, inherited from the SmolLM3-3B base model.