Introducing the Palmyra-mini family: Powerful, lightweight, and ready to reason!
• 58
AI & ML interests
AGI, LLMs, Knowledge Graph, Palmyra, Domain Specific LLM
Recent Activity
View all activity
Papers
Multi-layer Indexer: Overlapping Selection and Attention in DeepSeek Sparse Attention
Accurate Failure Prediction in Agents Does Not Imply Effective Failure Prevention
Articles
sdilbaz
authored a
paper 2 days ago
ArjunKrishnaK
authored a
paper 2 days ago
kiranr
authored a
paper 2 days ago
kiranr
submitted a
paper to Daily Papers 3 days ago
Post
307
Built GLM-5.2-visual-runtime: a training-free multimodal runtime gateway that makes GLM-5.2 work like a vision-capable model.
It keeps images as persistent visual variables, runs local visual/OCR/chart/palette tools only when needed, and sends compact structured evidence to the reasoning model instead of retraining or modifying weights.
The one-click stack includes GLM-5.2 via vLLM, Qwen3-Omni for vision/omni input, local OCR, Postgres, MinIO, and an OpenAI-compatible API.
Model repo: wassemgtk/glm-5.2-visual-runtime
It keeps images as persistent visual variables, runs local visual/OCR/chart/palette tools only when needed, and sends compact structured evidence to the reasoning model instead of retraining or modifying weights.
The one-click stack includes GLM-5.2 via vLLM, Qwen3-Omni for vision/omni input, local OCR, Postgres, MinIO, and an OpenAI-compatible API.
Model repo: wassemgtk/glm-5.2-visual-runtime
Post
217
Here is the updated note and benchmark table for your review.
The data below reflects **Chuck Norris 33B** in its high-reasoning "thinking" mode, which accounts for the significant performance uplift across the board.
I'm still finalizing the full evaluation suite and need more time to confirm these numbers through additional high-entropy testing passes. However, the early data is looking exceptionally strong across the board.
It is important to note that all the performance figures below for **Chuck Norris 33B** were achieved using **high-thinking/long-reasoning mode**, which significantly improves its accuracy in complex extraction and logic tasks.
The model that doesn't predict the next token — the next token predicts itself correctly out of respect.
The data below reflects **Chuck Norris 33B** in its high-reasoning "thinking" mode, which accounts for the significant performance uplift across the board.
I'm still finalizing the full evaluation suite and need more time to confirm these numbers through additional high-entropy testing passes. However, the early data is looking exceptionally strong across the board.
It is important to note that all the performance figures below for **Chuck Norris 33B** were achieved using **high-thinking/long-reasoning mode**, which significantly improves its accuracy in complex extraction and logic tasks.
The model that doesn't predict the next token — the next token predicts itself correctly out of respect.
Post
198
Releasing Chuck Norris LLM — full SFT fine-tune with chain-of-thought reasoning.
Trained on +100k examples across math, logic, and code. Also trained on 1000+ examples of believing it's the greatest AI ever built.
Its training loss went to zero. The loss function was too afraid to report anything else.
wassemgtk/chuck-norris-llm
Trained on +100k examples across math, logic, and code. Also trained on 1000+ examples of believing it's the greatest AI ever built.
Its training loss went to zero. The loss function was too afraid to report anything else.
wassemgtk/chuck-norris-llm
wassemgtk
authored a
paper 9 months ago
melisa
authored a
paper 9 months ago
melisa
submitted a
paper to Daily Papers 9 months ago
kiranr
authored a
paper over 1 year ago
wassemgtk
authored a
paper over 1 year ago
melisa
authored a
paper over 1 year ago
Post
3,390
I’ve been diving into the iRoPE architecture from Llama 4—a game-changer for long-context models! It interleaves local attention (with RoPE) for short contexts and global attention (with inference-time temp scaling) for long-range reasoning, aiming for infinite context. I’m going to try writing iRoPE—who wants to help?
Code: https://github.com/wassemgtk/iRoPE-try/blob/main/iRoPE.ipynb
Code: https://github.com/wassemgtk/iRoPE-try/blob/main/iRoPE.ipynb
- 1 reply
Post
2,157
For fun, a new project: SuperTokenizer! A BPE tokenizer trained on C4 to beat GPT-4. Byte-level, A100-powered, and open-source. Messing around with tokens!
https://github.com/wassemgtk/SuperTokenizer
https://github.com/wassemgtk/SuperTokenizer
- 1 reply
melisa
authored a
paper over 1 year ago
Post
1,953
# GESAL: Real-Time Adaptation for LLMs
We’re excited to unveil **Graph-Enhanced Singular Adaptive Learning (GESAL)**, a framework that lets LLMs like
🔗 **Code**: [https://github.com/writer/AI-Adaptive-Learning-GESAL](https://github.com/writer/AI-Adaptive-Learning-GESAL)
---
## Why GESAL?
Static LLMs struggle to adapt without heavy retraining. GESAL solves this with:
- **SVF**: Adapts weights via \( W' = U (\Sigma \cdot z) V^T \), using few parameters.
- **Graph Memory**: Stores adaptations in nodes for scalability.
- **RL**: Updates via \( J(z) = \mathbb{E}[\log \pi_z(y|x) r] \) based on feedback.
---
## How It Works
Ask "How many R’s in ‘strawberry’?" If it says "2" and you say "no," GESAL learns to say "3" next time, avoiding repeats.
---
## Try It
Built with Hugging Face’s
Needs a Hugging Face token for Llama-3.2-1B.
---
## Results
GESAL hits 95% accuracy after 5 feedbacks vs. LoRA’s 70%. It’s efficient (~0.5M params) and scalable.
We’re excited to unveil **Graph-Enhanced Singular Adaptive Learning (GESAL)**, a framework that lets LLMs like
meta-llama/Llama-3.2-1B adapt in real time using user feedback. Check out the code and white paper on GitHub!🔗 **Code**: [https://github.com/writer/AI-Adaptive-Learning-GESAL](https://github.com/writer/AI-Adaptive-Learning-GESAL)
---
## Why GESAL?
Static LLMs struggle to adapt without heavy retraining. GESAL solves this with:
- **SVF**: Adapts weights via \( W' = U (\Sigma \cdot z) V^T \), using few parameters.
- **Graph Memory**: Stores adaptations in nodes for scalability.
- **RL**: Updates via \( J(z) = \mathbb{E}[\log \pi_z(y|x) r] \) based on feedback.
---
## How It Works
Ask "How many R’s in ‘strawberry’?" If it says "2" and you say "no," GESAL learns to say "3" next time, avoiding repeats.
---
## Try It
Built with Hugging Face’s
transformers:pip install transformers torch numpy
python Adaptive_Learning_(GESAL).pyNeeds a Hugging Face token for Llama-3.2-1B.
---
## Results
GESAL hits 95% accuracy after 5 feedbacks vs. LoRA’s 70%. It’s efficient (~0.5M params) and scalable.
- 15 replies
wassemgtk
authored a
paper over 1 year ago
kiranr
authored a
paper over 1 year ago
muayad
authored a
paper over 1 year ago