samerzaher80/AetherMind_SRL
Text Classification
•
0.2B
•
Updated
•
24
•
1
SameerNajm PRO
samerzaher80
AI & ML interests
AI & ML interests
Recent Activity
updated
a model
23 days ago
samerzaher80/AetherMind_SRL
upvoted
an
article
25 days ago
AetherMind_SRL: Self-Reflective Learning for Robust Natural Language Inference
posted
an
update
about 1 month ago
AetherMind_SRL: How I beat 7B models on MMLU with 184M params and a $300 GPU
I’m Sameer, a solo researcher from Iraq working on a single RTX 3050 8GB laptop.Today I’m releasing AetherMind_SRL – a 184M-parameter NLI model that was trained only on tasks (SNLI, MNLI, ANLI, and a small clinical Alzheimer’s dataset).
It was never fine-tuned or even shown a single MMLU question during training.Yet here are the zero-shot MMLU (57 subjects) results:Model
MMLU Zero-Shot
Training Data
AetherMind_SRL (me)
184M
36.05 %
Only NLI (SNLI/MNLI/ANLI + ADNI)
DeBERTa-v3-base
278M
~30.8 %
General pre-training
BERT-large
340M
27–30 %
General pre-training
LLaMA-1 7B
7B
34–35 %
Massive text corpus
LLaMA-2 7B
7B
~45 %
Bigger + better data
Yes – my 184M model beats every classic 300–400M model and the original 7-billion-parameter LLaMA-1, all while running at 300+ samples/sec on a $300 laptop GPU.How did this happen?I built a standardized self-improvement loop called AetherMind Self-Reflective Learning (SRL) v1.0:Train normally on NLI
Let the model predict on hard adversarial data (ANLI)
Log every mistake + low-confidence case
Build a balanced “SMART” buffer (60% errors + 40% correct anchors)
Fine-tune with tiny LR and error-weighted loss
Repeat until stable
That’s it. No external knowledge, no MMLU data, no cluster.
Just pure reasoning transfer from entailment/contradiction patterns → real-world knowledge.Try it yourself python
from transformers import pipeline
import torch
nli_pipeline = pipeline(
"text-classification",
model="samerzaher80/AetherMind_SRL",
device=0 if torch.cuda.is_available() else -1
)
# DEFINE YOUR TEST HERE
premise = "Patient shows progressive memory decline."
hypothesis = "Patient shows progressive memory decline."
input_text = f"{premise} [SEP] {hypothesis}"
result = nli_pipeline(input_text)[0]
print(f"Prediction: {result['label']}")
print(f"Confidence: {result['score']:
Model: https://huggingface.co/samerzaher80/AetherMind_SRL
Organizations
None yet
updated
a
model
23 days ago
upvoted
an
article
25 days ago
Article
AetherMind_SRL: Self-Reflective Learning for Robust Natural Language Inference
•
2
posted
an
update
about 1 month ago
Post
1627
AetherMind_SRL: How I beat 7B models on MMLU with 184M params and a $300 GPU
I’m Sameer, a solo researcher from Iraq working on a single RTX 3050 8GB laptop.Today I’m releasing AetherMind_SRL – a 184M-parameter NLI model that was trained only on tasks (SNLI, MNLI, ANLI, and a small clinical Alzheimer’s dataset).
It was never fine-tuned or even shown a single MMLU question during training.Yet here are the zero-shot MMLU (57 subjects) results:Model
MMLU Zero-Shot
Training Data
AetherMind_SRL (me)
184M
36.05 %
Only NLI (SNLI/MNLI/ANLI + ADNI)
DeBERTa-v3-base
278M
~30.8 %
General pre-training
BERT-large
340M
27–30 %
General pre-training
LLaMA-1 7B
7B
34–35 %
Massive text corpus
LLaMA-2 7B
7B
~45 %
Bigger + better data
Yes – my 184M model beats every classic 300–400M model and the original 7-billion-parameter LLaMA-1, all while running at 300+ samples/sec on a $300 laptop GPU.How did this happen?I built a standardized self-improvement loop called AetherMind Self-Reflective Learning (SRL) v1.0:Train normally on NLI
Let the model predict on hard adversarial data (ANLI)
Log every mistake + low-confidence case
Build a balanced “SMART” buffer (60% errors + 40% correct anchors)
Fine-tune with tiny LR and error-weighted loss
Repeat until stable
That’s it. No external knowledge, no MMLU data, no cluster.
Just pure reasoning transfer from entailment/contradiction patterns → real-world knowledge.Try it yourself python
from transformers import pipeline
import torch
nli_pipeline = pipeline(
"text-classification",
model="samerzaher80/AetherMind_SRL",
device=0 if torch.cuda.is_available() else -1
)
# DEFINE YOUR TEST HERE
premise = "Patient shows progressive memory decline."
hypothesis = "Patient shows progressive memory decline."
input_text = f"{premise} [SEP] {hypothesis}"
result = nli_pipeline(input_text)[0]
print(f"Prediction: {result['label']}")
print(f"Confidence: {result['score']:
Model: samerzaher80/AetherMind_SRL
I’m Sameer, a solo researcher from Iraq working on a single RTX 3050 8GB laptop.Today I’m releasing AetherMind_SRL – a 184M-parameter NLI model that was trained only on tasks (SNLI, MNLI, ANLI, and a small clinical Alzheimer’s dataset).
It was never fine-tuned or even shown a single MMLU question during training.Yet here are the zero-shot MMLU (57 subjects) results:Model
MMLU Zero-Shot
Training Data
AetherMind_SRL (me)
184M
36.05 %
Only NLI (SNLI/MNLI/ANLI + ADNI)
DeBERTa-v3-base
278M
~30.8 %
General pre-training
BERT-large
340M
27–30 %
General pre-training
LLaMA-1 7B
7B
34–35 %
Massive text corpus
LLaMA-2 7B
7B
~45 %
Bigger + better data
Yes – my 184M model beats every classic 300–400M model and the original 7-billion-parameter LLaMA-1, all while running at 300+ samples/sec on a $300 laptop GPU.How did this happen?I built a standardized self-improvement loop called AetherMind Self-Reflective Learning (SRL) v1.0:Train normally on NLI
Let the model predict on hard adversarial data (ANLI)
Log every mistake + low-confidence case
Build a balanced “SMART” buffer (60% errors + 40% correct anchors)
Fine-tune with tiny LR and error-weighted loss
Repeat until stable
That’s it. No external knowledge, no MMLU data, no cluster.
Just pure reasoning transfer from entailment/contradiction patterns → real-world knowledge.Try it yourself python
from transformers import pipeline
import torch
nli_pipeline = pipeline(
"text-classification",
model="samerzaher80/AetherMind_SRL",
device=0 if torch.cuda.is_available() else -1
)
# DEFINE YOUR TEST HERE
premise = "Patient shows progressive memory decline."
hypothesis = "Patient shows progressive memory decline."
input_text = f"{premise} [SEP] {hypothesis}"
result = nli_pipeline(input_text)[0]
print(f"Prediction: {result['label']}")
print(f"Confidence: {result['score']:
Model: samerzaher80/AetherMind_SRL
posted
an
update
about 1 month ago
Post
275
Need Help Getting arXiv Endorsement for My AI Research Paper
Hi everyone,
I hope you're doing well. I’m trying to publish my new AI research paper on arXiv under the cs.AI category, but I currently need an endorser who is already authorized for cs.AI submissions.
If anyone here is registered as a cs.AI endorser and is willing to help, I would truly appreciate it.
Here is the official arXiv endorsement request link:
🔗 https://arxiv.org/auth/endorse?x=EZEMO7
(Backup: http://arxiv.org/auth/endorse.php — Code: EZEMO7)
My research:
It’s part of the AetherMind project — a self-reflective NLI reasoning system inspired by human cognitive consistency and used also in Alzheimer’s research. If needed, I can share the abstract or full PDF.
Thank you so much to anyone who can support.
— Sameer S.Najm
Hi everyone,
I hope you're doing well. I’m trying to publish my new AI research paper on arXiv under the cs.AI category, but I currently need an endorser who is already authorized for cs.AI submissions.
If anyone here is registered as a cs.AI endorser and is willing to help, I would truly appreciate it.
Here is the official arXiv endorsement request link:
🔗 https://arxiv.org/auth/endorse?x=EZEMO7
(Backup: http://arxiv.org/auth/endorse.php — Code: EZEMO7)
My research:
It’s part of the AetherMind project — a self-reflective NLI reasoning system inspired by human cognitive consistency and used also in Alzheimer’s research. If needed, I can share the abstract or full PDF.
Thank you so much to anyone who can support.
— Sameer S.Najm
posted
an
update
about 1 month ago
Post
1851
Need Help Getting arXiv Endorsement for My AI Research Paper
Hi everyone,
I hope you're doing well. I’m trying to publish my new AI research paper on arXiv under the cs.AI category, but I currently need an endorser who is already authorized for cs.AI submissions.
If anyone here is registered as a cs.AI endorser and is willing to help, I would truly appreciate it.
Here is the official arXiv endorsement request link:
🔗 https://arxiv.org/auth/endorse?x=EZEMO7
(Backup: http://arxiv.org/auth/endorse.php — Code: EZEMO7)
My research:
It’s part of the AetherMind project — a self-reflective NLI reasoning system inspired by human cognitive consistency and used also in Alzheimer’s research. If needed, I can share the abstract or full PDF.
Thank you so much to anyone who can support.
— Sameer S.Najm
Hi everyone,
I hope you're doing well. I’m trying to publish my new AI research paper on arXiv under the cs.AI category, but I currently need an endorser who is already authorized for cs.AI submissions.
If anyone here is registered as a cs.AI endorser and is willing to help, I would truly appreciate it.
Here is the official arXiv endorsement request link:
🔗 https://arxiv.org/auth/endorse?x=EZEMO7
(Backup: http://arxiv.org/auth/endorse.php — Code: EZEMO7)
My research:
It’s part of the AetherMind project — a self-reflective NLI reasoning system inspired by human cognitive consistency and used also in Alzheimer’s research. If needed, I can share the abstract or full PDF.
Thank you so much to anyone who can support.
— Sameer S.Najm
reacted to
aufklarer's
post with 🔥
about 1 month ago
Post
1448
Couple months ago I fine‑tuned Qwen3 Embeddings with LoRA on the LSPC dataset. This time I went the opposite way: a small, task‑specific 80M encoder with bidirectional attention, trained end‑to‑end. It outperforms the Qwen3 LoRA baseline on the same data (0.9315 macro‑F1 vs 0.8360). Details and code: https://blog.ivan.digital/beating-qwen3-lora-with-a-tiny-pytorch-encoder-on-the-large-scale-product-corpus-afe536de205f
posted
an
update
about 1 month ago
Post
868
New Research Release: AetherMind_SRL — Self-Reflective Learning for Robust Natural Language Inference (NLI)
By: Sameer S. Najm
After more than a year of continuous work, experimentation, and refinement, I am proud to officially publish AetherMind_SRL, a self-improving Transformer model trained using Self-Reflective Learning (SRL) — a technique that enables the model to learn from its own mistakes.
This research integrates:
🔹 Knowledge Distillation from DeBERTa-v3-base
🔹 Self-Reflective Learning loops
🔹 Adversarial ANLI training
🔹 Clinical Alzheimer’s reasoning (ADNI)
🔹 SMART Error Buffers for hard-example mining
What makes it unique?
AetherMind_SRL continuously improves through structured error logs, balanced correction buffers, and clinical-domain adaptation. It achieves strong performance across general NLI benchmarks, adversarial datasets, and Alzheimer’s-focused reasoning tasks.
Official Research Preprint (Zenodo DOI):
https://doi.org/10.5281/zenodo.17670971
Model Repository (Hugging Face):
samerzaher80/AetherMind_SRL
Full community article (Hugging Face Blog):
https://huggingface.co/blog/samerzaher80/aethermind-srl
By: Sameer S. Najm
After more than a year of continuous work, experimentation, and refinement, I am proud to officially publish AetherMind_SRL, a self-improving Transformer model trained using Self-Reflective Learning (SRL) — a technique that enables the model to learn from its own mistakes.
This research integrates:
🔹 Knowledge Distillation from DeBERTa-v3-base
🔹 Self-Reflective Learning loops
🔹 Adversarial ANLI training
🔹 Clinical Alzheimer’s reasoning (ADNI)
🔹 SMART Error Buffers for hard-example mining
What makes it unique?
AetherMind_SRL continuously improves through structured error logs, balanced correction buffers, and clinical-domain adaptation. It achieves strong performance across general NLI benchmarks, adversarial datasets, and Alzheimer’s-focused reasoning tasks.
Official Research Preprint (Zenodo DOI):
https://doi.org/10.5281/zenodo.17670971
Model Repository (Hugging Face):
samerzaher80/AetherMind_SRL
Full community article (Hugging Face Blog):
https://huggingface.co/blog/samerzaher80/aethermind-srl
AetherMind_SRL Released – SRL-ANLI Round 12 Paper (DOI Inside)
#1 opened about 1 month ago
by
samerzaher80
published
an
article
about 1 month ago
Article
AetherMind_SRL: Self-Reflective Learning for Robust Natural Language Inference
•
2
published
a
model
about 1 month ago
updated
a
dataset
about 1 month ago
published
a
dataset
about 1 month ago
published
an
article
about 1 month ago
Article
AetherMind-KD-Student: A Robust and Efficient Knowledge-Distilled NLI Model
posted
an
update
about 1 month ago
Post
281
AetherMind-KD-Student is a 184M-parameter Natural Language Inference (NLI) model distilled from a DeBERTa-v3 teacher using a multi-stage, adversarial-aware knowledge distillation pipeline.
The model is designed to provide:
High accuracy on standard NLI benchmarks
Strong robustness on adversarial datasets
Excellent zero-shot generalization to unseen datasets
High inference efficiency on consumer GPUs
This makes it suitable for research and practical applications that require fast and reliable sentence-level reasoning.
samerzaher80/AetherMind-KD-Student
The model is designed to provide:
High accuracy on standard NLI benchmarks
Strong robustness on adversarial datasets
Excellent zero-shot generalization to unseen datasets
High inference efficiency on consumer GPUs
This makes it suitable for research and practical applications that require fast and reliable sentence-level reasoning.
samerzaher80/AetherMind-KD-Student
reacted to
evalstate's
post with 🤗
about 1 month ago
Post
2357
Hugging Face MCP Server v0.2.46
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
- Add "discover" to Dynamic Space tool. Recommend deselecting "space_search" if using dynamic spaces.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
- Add "discover" to Dynamic Space tool. Recommend deselecting "space_search" if using dynamic spaces.
Official Release: AetherMind-KD-Student (184M)
1
#1 opened about 1 month ago
by
samerzaher80
Official Release: AetherMind-KD-Student (184M)
#2 opened about 1 month ago
by
samerzaher80
published
a
model
about 1 month ago
updated
a
model
about 1 month ago
updated
a
Space
7 months ago