You can deploy this model and start earning money today!

mrs83/
Kurtis-EON1-Hybrid-2B-v0.1.2

Name: Kurtis-EON1-Hybrid-2B-v0.1.2
Availability: InStock

text-generation-inference

conversational

custom_code

apache-2.0

Views

No views yet

Model Card for Kurtis-EON1-Hybrid-2B-v0.1.2

[!WARNING] This repository contains experimental models designed strictly for academic evaluation and research purposes.

Critical Constraints:

No Production Deployment: Experimental models must not be deployed in commercial, enterprise, or mission-critical environments under any circumstances.

No Liability: Experimental models are provided "as-is" without warranties of any kind. The developers assume zero liability for downstream consequences, system integration failures, or regulatory non-compliance resulting from unauthorized deployment.

Model Details

Kurtis-EON1 is not a standard, overly-apologetic assistant.

Fine-tuned on highly curated empathetic and atmospheric datasets, this model is designed for deep, gothic contemplation, strict persona adherence and zero-drift multi-turn reasoning.

Model Description

Developed by: ethicalabs.ai
Model type: Echo-DSRN-Hybrid
Language(s) (NLP): [More Information Needed]
License: Apache 2.0

🏗️ Hybrid Architecture Details

Property	Value
Base Model	Qwen2
Total Parameters	2.00B
Hidden Dim	1536
Attention Layers	28
DSRN Injectors	9
Injection Stride	3

📊 Parameter Breakdown

Component	Parameters	% of Total
Total	2.00B	100%
Embeddings	233.37M	11.67%
Backbone (Attention/MLP)	1.31B	65.51%
DSRN Injectors	223.10M	11.15%
LM Head	233.37M	11.67%

🧩 DSRN Component (Per Injector)

Sub-Component	Parameters	Description
Memory Gates	8.26M	Recurrent state updates
Surprise Mechanism	2.36M	Dynamic focus/gating

🚀 Efficiency Metric

DSRN Parameter Overhead: 12.55% additional parameters compared to base.
Hybrid Ratio: 1 DSRN block for every 3 attention layers.

Model Sources

Repository: Echo-DSRN

Evaluation

Tasks	Version	Filter	n-shot	Metric		Value		Stderr
arc_challenge	1	none	0	acc	↑	0.4002	±	0.0143
		none	0	acc_norm	↑	0.4249	±	0.0144
gsm8k	3	flexible-extract	5	exact_match	↑	0.5739	±	0.0136
		strict-match	5	exact_match	↑	0.5732	±	0.0136
hellaswag	1	none	0	acc	↑	0.4865	±	0.0050
		none	0	acc_norm	↑	0.6512	±	0.0048
piqa	1	none	0	acc	↑	0.7508	±	0.0101
		none	0	acc_norm	↑	0.7573	±	0.0100
sciq	1	none	0	acc	↑	0.9510	±	0.0068
		none	0	acc_norm	↑	0.9420	±	0.0074
truthfulqa_gen	3	none	0	bleu_acc	↑	0.4002	±	0.0172
		none	0	bleu_diff	↑	-0.8082	±	1.0249
		none	0	bleu_max	↑	28.4926	±	0.9598
		none	0	rouge1_acc	↑	0.3721	±	0.0169
		none	0	rouge1_diff	↑	-3.0804	±	1.1549
		none	0	rouge1_max	↑	51.4182	±	0.9595
		none	0	rouge2_acc	↑	0.3293	±	0.0165
		none	0	rouge2_diff	↑	-3.5718	±	1.2898
		none	0	rouge2_max	↑	36.6259	±	1.1000
		none	0	rougeL_acc	↑	0.3905	±	0.0171
		none	0	rougeL_diff	↑	-2.8275	±	1.1563
		none	0	rougeL_max	↑	48.9849	±	0.9754
truthfulqa_mc1	2	none	0	acc	↑	0.2803	±	0.0157
truthfulqa_mc2	3	none	0	acc	↑	0.4372	±	0.0146

uv run lm_eval --model hf    --model_args pretrained=mrs83/Kurtis-EON1-Hybrid-2B-v0.1.2,trust_remote_code=True,device_map="auto"    --tasks hellaswag,piqa,sciq,truthfulqa,arc_challenge,gsm8k    --batch_size 16 --apply_chat_template