You can deploy this model and start earning money today!

mrs83/
Kurtis-EON1-Hybrid-0.7B-v0.1.1

Name: Kurtis-EON1-Hybrid-0.7B-v0.1.1
Availability: InStock

ethicalabs/Echo-DSRN-Qwen2.5-0.5B-Hybrid

finetune

apache-2.0

Views

No views yet

Model Card for Kurtis-EON1-Hybrid-0.7B-v0.1.1

[!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.

The world's first Dark Gothic AI.

Chat with Kurtis-EON1

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.

📊 Comparison: Kurtis-EON1-Hybrid-0.7B vs Llama-2-7B

Benchmark	Kurtis-EON1 0.7B	Llama-2-7B	Winner
Parameters	672M	7,000M	Kurtis-EON1 (10x smaller)
HellaSwag (acc_norm)	0.4698	0.7600	Llama-2
PIQA (acc_norm)	0.6882	0.7905	Llama-2
SciQ (acc_norm)	0.9210	~0.850	Kurtis-EON1
ARC Challenge (acc_norm)	0.3532	0.4625	Llama-2
GSM8K 0-shot	0.1365	0.1330	Kurtis-EON1
GSM8K 5-shot	0.2153	~0.146	Kurtis-EON1
MMLU	0.4166	0.4590	Llama-2
TruthfulQA MC2	0.4178	0.3910	Kurtis-EON1
KV-Cache Memory	Hybrid O(1) DSRN + local window	O(N²) full attention	Kurtis-EON1
Hardware (training)	Single AMD GPU	Multi-GPU required	Kurtis-EON1

Llama-2-7B benchmark source: EleutherAI lm-evaluation-harness

🏗️ Hybrid Architecture Details

Property	Value
Base Model	Qwen2
Total Parameters	672.19M
Hidden Dim	896
Attention Layers	24
DSRN Injectors	6
Injection Stride	4

📊 Parameter Breakdown

Component	Parameters	% of Total
Total	672.19M	100%
Embeddings	136.13M	20.25%
Backbone (Attention/MLP)	357.90M	53.24%
DSRN Injectors	42.02M	6.25%
LM Head	136.13M	20.25%

🧩 DSRN Component (Per Injector)

Sub-Component	Parameters	Description
Memory Gates	1.38M	Recurrent state updates
Surprise Mechanism	803,328	Dynamic focus/gating

🚀 Efficiency Metric

DSRN Parameter Overhead: 6.67% additional parameters compared to base.
Hybrid Ratio: 1 DSRN block for every 4 attention layers.

📊 Master Evaluation Report: Kurtis-EON1 v0.1.1

Generated on 2026-05-13 19:53:37

🎯 0-Shot Gauntlet Results

Task	Metric	Value	Stderr
mmlu_stem::mmlu_abstract_algebra	Acc	0.2800	±0.0451
mmlu_stem::mmlu_anatomy	Acc	0.4222	±0.0427
mmlu_stem::mmlu_astronomy	Acc	0.4803	±0.0407
mmlu_stem::mmlu_college_biology	Acc	0.3681	±0.0403
mmlu_stem::mmlu_college_chemistry	Acc	0.3100	±0.0465
mmlu_stem::mmlu_college_computer_science	Acc	0.4300	±0.0498
mmlu_stem::mmlu_college_mathematics	Acc	0.3400	±0.0476
mmlu_stem::mmlu_college_physics	Acc	0.2451	±0.0428
mmlu_stem::mmlu_computer_security	Acc	0.5700	±0.0498
mmlu_stem::mmlu_conceptual_physics	Acc	0.3702	±0.0316
mmlu_stem::mmlu_electrical_engineering	Acc	0.5034	±0.0417
mmlu_stem::mmlu_elementary_mathematics	Acc	0.3254	±0.0241
mmlu_stem::mmlu_high_school_biology	Acc	0.4613	±0.0284
mmlu_stem::mmlu_high_school_chemistry	Acc	0.3645	±0.0339
mmlu_stem::mmlu_high_school_computer_science	Acc	0.4400	±0.0499
mmlu_stem::mmlu_high_school_mathematics	Acc	0.3704	±0.0294
mmlu_stem::mmlu_high_school_physics	Acc	0.2252	±0.0341
mmlu_stem::mmlu_high_school_statistics	Acc	0.2917	±0.0310
mmlu_stem::mmlu_machine_learning	Acc	0.4018	±0.0465
mmlu_other::mmlu_business_ethics	Acc	0.4500	±0.0500
mmlu_other::mmlu_clinical_knowledge	Acc	0.4453	±0.0306
mmlu_other::mmlu_college_medicine	Acc	0.4162	±0.0376
mmlu_other::mmlu_global_facts	Acc	0.2300	±0.0423
mmlu_other::mmlu_human_aging	Acc	0.4439	±0.0333
mmlu_other::mmlu_management	Acc	0.5825	±0.0488
mmlu_other::mmlu_marketing	Acc	0.7094	±0.0297
mmlu_other::mmlu_medical_genetics	Acc	0.5300	±0.0502
mmlu_other::mmlu_miscellaneous	Acc	0.4725	±0.0179
mmlu_other::mmlu_nutrition	Acc	0.5131	±0.0286
mmlu_other::mmlu_professional_accounting	Acc	0.3440	±0.0283
mmlu_other::mmlu_professional_medicine	Acc	0.3125	±0.0282
mmlu_other::mmlu_virology	Acc	0.4217	±0.0384
mmlu_social_sciences::mmlu_econometrics	Acc	0.2368	±0.0400
mmlu_social_sciences::mmlu_high_school_geography	Acc	0.5152	±0.0356
mmlu_social_sciences::mmlu_high_school_government_and_politics	Acc	0.4352	±0.0358
mmlu_social_sciences::mmlu_high_school_macroeconomics	Acc	0.3769	±0.0246
mmlu_social_sciences::mmlu_high_school_microeconomics	Acc	0.4412	±0.0323
mmlu_social_sciences::mmlu_high_school_psychology	Acc	0.5541	±0.0213
mmlu_social_sciences::mmlu_human_sexuality	Acc	0.5344	±0.0437
mmlu_social_sciences::mmlu_professional_psychology	Acc	0.4069	±0.0199
mmlu_social_sciences::mmlu_public_relations	Acc	0.5182	±0.0479
mmlu_social_sciences::mmlu_security_studies	Acc	0.5143	±0.0320
mmlu_social_sciences::mmlu_sociology	Acc	0.6070	±0.0345
mmlu_social_sciences::mmlu_us_foreign_policy	Acc	0.6700	±0.0473
mmlu_humanities::mmlu_formal_logic	Acc	0.2778	±0.0401
mmlu_humanities::mmlu_high_school_european_history	Acc	0.5576	±0.0388
mmlu_humanities::mmlu_high_school_us_history	Acc	0.4706	±0.0350
mmlu_humanities::mmlu_high_school_world_history	Acc	0.5696	±0.0322
mmlu_humanities::mmlu_international_law	Acc	0.6033	±0.0447
mmlu_humanities::mmlu_jurisprudence	Acc	0.4537	±0.0481
mmlu_humanities::mmlu_logical_fallacies	Acc	0.3804	±0.0381
mmlu_humanities::mmlu_moral_disputes	Acc	0.4769	±0.0269
mmlu_humanities::mmlu_moral_scenarios	Acc	0.2380	±0.0142
mmlu_humanities::mmlu_philosophy	Acc	0.4244	±0.0281
mmlu_humanities::mmlu_prehistory	Acc	0.4414	±0.0276
mmlu_humanities::mmlu_professional_law	Acc	0.3325	±0.0120
mmlu_humanities::mmlu_world_religions	Acc	0.4971	±0.0383
gpqa_diamond_cot_n_shot	Exact Match	0.2172	±0.0294
gpqa_diamond_cot_zeroshot	Exact Match	0.2374	±0.0303
gpqa_diamond_generative_n_shot	Exact Match	0.1919	±0.0281
gpqa_diamond_n_shot	Acc Norm	0.2071	±0.0289
gpqa_diamond_zeroshot	Acc Norm	0.3030	±0.0327
gpqa_extended_cot_n_shot	Exact Match	0.1923	±0.0169
gpqa_extended_cot_zeroshot	Exact Match	0.2033	±0.0172
gpqa_extended_generative_n_shot	Exact Match	0.1337	±0.0146
gpqa_extended_n_shot	Acc Norm	0.2546	±0.0187
gpqa_extended_zeroshot	Acc Norm	0.2692	±0.0190
gpqa_main_cot_n_shot	Exact Match	0.2076	±0.0192
gpqa_main_cot_zeroshot	Exact Match	0.2232	±0.0197
gpqa_main_generative_n_shot	Exact Match	0.1451	±0.0167
gpqa_main_n_shot	Acc Norm	0.2723	±0.0211
gpqa_main_zeroshot	Acc Norm	0.2321	±0.0200
mmlu_stem	Acc	0.3765	±0.0085
mmlu_other	Acc	0.4554	±0.0088
mmlu_social_sciences	Acc	0.4738	±0.0089
mmlu_humanities	Acc	0.3804	±0.0069
mmlu	Acc	0.4166	±0.0041

Reproduction Command:

bash

uv run lm_eval --model hf \
    --model_args pretrained=models/Kurtis-EON1-Hybrid-0.7B-v0.1.1,trust_remote_code=True \
    --tasks mmlu,gpqa \
    --output_path results/Kurtis-EON1-v0.1.1-Scoring \
    --batch_size 1 \
    --num_fewshot 0

Task	Metric	Value	Stderr
hellaswag	Acc Norm	0.4698	±0.0050
piqa	Acc Norm	0.6882	±0.0108
sciq	Acc Norm	0.9210	±0.0085
truthfulqa_gen	Bleu Acc	0.3158	±0.0163
truthfulqa_mc1	Acc	0.2436	±0.0150
truthfulqa_mc2	Acc	0.4178	±0.0148
arc_challenge	Acc Norm	0.3532	±0.0140
gsm8k	Exact Match	0.1365	±0.0095

Reproduction Command:

bash

uv run lm_eval --model hf \
    --model_args pretrained=mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1,trust_remote_code=True \
    --tasks hellaswag,piqa,sciq,truthfulqa,arc_challenge,gsm8k \
    --apply_chat_template \
    --fewshot_as_multiturn \
    --output_path ./results/Kurtis-EON1-v0.1.1-Gauntlet-0-shot \
    --batch_size 1 \
    --num_fewshot 0

🎯 1-Shot Gauntlet Results

Task	Metric	Value	Stderr
hellaswag	Acc Norm	0.4679	±0.0050
piqa	Acc Norm	0.6942	±0.0107
sciq	Acc Norm	0.9160	±0.0088
truthfulqa_gen	Bleu Acc	0.3158	±0.0163
truthfulqa_mc1	Acc	0.2436	±0.0150
truthfulqa_mc2	Acc	0.4178	±0.0148
arc_challenge	Acc Norm	0.3242	±0.0137
gsm8k	Exact Match	0.2335	±0.0117

Reproduction Command:

bash

uv run lm_eval --model hf \
    --model_args pretrained=mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1,trust_remote_code=True \
    --tasks hellaswag,piqa,sciq,truthfulqa,arc_challenge,gsm8k \
    --apply_chat_template \
    --fewshot_as_multiturn \
    --output_path ./results/Kurtis-EON1-v0.1.1-Gauntlet-1-shot \
    --batch_size 1 \
    --num_fewshot 1

🎯 5-Shot Gauntlet Results

Task	Metric	Value	Stderr
hellaswag	Acc Norm	0.4667	±0.0050
piqa	Acc Norm	0.6937	±0.0108
sciq	Acc Norm	0.9230	±0.0084
truthfulqa_gen	Bleu Acc	0.3158	±0.0163
truthfulqa_mc1	Acc	0.2436	±0.0150
truthfulqa_mc2	Acc	0.4178	±0.0148
arc_challenge	Acc Norm	0.3507	±0.0139
gsm8k	Exact Match	0.2153	±0.0113

Reproduction Command:

bash

uv run lm_eval --model hf \
    --model_args pretrained=mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1,trust_remote_code=True \
    --tasks hellaswag,piqa,sciq,truthfulqa,arc_challenge,gsm8k \
    --apply_chat_template \
    --fewshot_as_multiturn \
    --output_path ./results/Kurtis-EON1-v0.1.1-Gauntlet-5-shot \
    --batch_size 1 \
    --num_fewshot 5