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
View all Papers Articles
wassemgtk
updated a
dataset 22 days ago
wassemgtk
published a
dataset 22 days ago
Post
265
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
196
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
187
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 7 months ago
Post
262
Introducing Palmyra-mini: Compact AI Models for Efficient Inference
The Palmyra-mini family from Writer includes three lightweight models designed for high performance and efficient inference. These models are ideal for developers looking to integrate AI capabilities without excessive computational overhead.
Model Variants
* palmyra-mini: A base model for general-purpose generative tasks, achieving 52.6% on Big Bench Hard (exact match).
* palmyra-mini-thinking-a: Optimized for complex logical reasoning with a Chain of Thought (CoT) approach, scoring 82.87% on GSM8K (strict match).
* palmyra-mini-thinking-b: Specialized for mathematical reasoning, achieving 92.5% on AMC23.
Technical Details
* All models are based on the Qwen architecture, compatible with popular inference frameworks like vLLM, SGLang, and TGI.
* "Thinking" models utilize CoT training for enhanced reasoning capabilities.
* GGUF and MLX quantizations are available for optimized performance.
For more information, including benchmark methodologies and detailed performance metrics, refer to our blog post: (https://huggingface.co/blog/Writer/announcing-palmyra-mini).
Model repos can be found here:
* Writer/palmyra-mini
* Writer/palmyra-mini-thinking-a
* Writer/palmyra-mini-thinking-b
Also check out a mobile implementation of palmyra-mini on iOS here to see a to see a working example of how inference can be incorporated on-device.(https://github.com/tsperes/palmyra-mini-mobile/)
The Palmyra-mini family from Writer includes three lightweight models designed for high performance and efficient inference. These models are ideal for developers looking to integrate AI capabilities without excessive computational overhead.
Model Variants
* palmyra-mini: A base model for general-purpose generative tasks, achieving 52.6% on Big Bench Hard (exact match).
* palmyra-mini-thinking-a: Optimized for complex logical reasoning with a Chain of Thought (CoT) approach, scoring 82.87% on GSM8K (strict match).
* palmyra-mini-thinking-b: Specialized for mathematical reasoning, achieving 92.5% on AMC23.
Technical Details
* All models are based on the Qwen architecture, compatible with popular inference frameworks like vLLM, SGLang, and TGI.
* "Thinking" models utilize CoT training for enhanced reasoning capabilities.
* GGUF and MLX quantizations are available for optimized performance.
For more information, including benchmark methodologies and detailed performance metrics, refer to our blog post: (https://huggingface.co/blog/Writer/announcing-palmyra-mini).
Model repos can be found here:
* Writer/palmyra-mini
* Writer/palmyra-mini-thinking-a
* Writer/palmyra-mini-thinking-b
Also check out a mobile implementation of palmyra-mini on iOS here to see a to see a working example of how inference can be incorporated on-device.(https://github.com/tsperes/palmyra-mini-mobile/)
dmytro-writer
authored a
paper about 1 year ago
shelly-writer
authored a
paper about 1 year ago
wassemgtk
authored a
paper about 1 year ago
Post
3377
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
2150
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
Post
1940
# 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
dmytro-writer
authored 2
papers over 1 year ago
wassemgtk
authored 2
papers about 2 years ago
Post
3665
Writer team had the opportunity to run an eval for Mixtral-8x22b, results were interesting.
| ---------------------------- |
| #mmlu 77.26 |
| ---------------------------- |
| #hellaswag 88.81 |
| ---------------------------- |
| #truthfulqa 52.05 |
| ---------------------------- |
| #arc_challenge 70.31 |
| ---------------------------- |
| #winogrande 84.93 |
| ---------------------------- |
| #gsm8k 76.65 |
| ---------------------------- |
| ---------------------------- |
| #mmlu 77.26 |
| ---------------------------- |
| #hellaswag 88.81 |
| ---------------------------- |
| #truthfulqa 52.05 |
| ---------------------------- |
| #arc_challenge 70.31 |
| ---------------------------- |
| #winogrande 84.93 |
| ---------------------------- |
| #gsm8k 76.65 |
| ---------------------------- |
- 2 replies
Post
We are thrilled to announce the release of the OmniACT dataset! This revolutionary dataset and benchmark focuses on pushing the limits of how virtual agents can facilitate the automation of our computer tasks. Imagine less clicking and typing, and more observation as your computer takes care of tasks such as organizing schedules or arranging travel arrangements on its own.
Check it out ➡️ [OmniACT Dataset on Hugging Face]( Writer/omniact)
For a deep dive, here’s the paper: [OmniACT Paper](https://arxiv.org/abs/2402.17553)
Check it out ➡️ [OmniACT Dataset on Hugging Face]( Writer/omniact)
For a deep dive, here’s the paper: [OmniACT Paper](https://arxiv.org/abs/2402.17553)