Delentia

delentia-slm-jitna-router

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Container

Run this model inference with full control and performance in your environment.

Talk with our engineer to get a quote for reserved GPU instances with discounts.

Technical Specifications

Base Model: unsloth/Meta-Llama-3.1-8B-bnb-4bit
Fine-Tuning Method: Sequence Classification QLoRA (SEQ_CLS adapter)
Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Output Labels:
- 0: Executor (Tool / JSON Execution)
- 1: Router Base (Conversational / Standard Prompt)
- 2: Guardian (Safety Shield evaluation)
- 3: Scribe (Context compression/summarization)

Zero Static Padding: Re-engineered training pipeline removes static padding and reduces input context to max_length=512.
Inference Latency: 20-50 milliseconds on consumer-grade local hardware, enabling instantaneous routing decisions.
Accuracy Gate: Achieved $\geq 96%$ intent classification accuracy on JITNA router evaluation dataset.