base_model:
- Qwen/QwQ-32B-Preview
tags:
- text-generation-inference
- transformers
- unsloth
- qwen2
- trl
- Chain-of-thought
- Reasoning
license: apache-2.0
language:
- en
new_version: Daemontatox/CogitoZ
library_name: transformers
datasets:
- PJMixers/Math-Multiturn-100K-ShareGPT
model-index:
- name: CogitoZ
results:
- task:
type: text-generation
name: Text Generation
dataset:
name: IFEval (0-Shot)
type: wis-k/instruction-following-eval
split: train
args:
num_few_shot: 0
metrics:
- type: inst_level_strict_acc and prompt_level_strict_acc
value: 39.67
name: averaged accuracy
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FCogitoZ
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: BBH (3-Shot)
type: SaylorTwift/bbh
split: test
args:
num_few_shot: 3
metrics:
- type: acc_norm
value: 53.89
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FCogitoZ
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: MATH Lvl 5 (4-Shot)
type: lighteval/MATH-Hard
split: test
args:
num_few_shot: 4
metrics:
- type: exact_match
value: 46.3
name: exact match
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FCogitoZ
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: GPQA (0-shot)
type: Idavidrein/gpqa
split: train
args:
num_few_shot: 0
metrics:
- type: acc_norm
value: 19.35
name: acc_norm
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FCogitoZ
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: MuSR (0-shot)
type: TAUR-Lab/MuSR
args:
num_few_shot: 0
metrics:
- type: acc_norm
value: 19.94
name: acc_norm
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FCogitoZ
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: MMLU-PRO (5-shot)
type: TIGER-Lab/MMLU-Pro
config: main
split: test
args:
num_few_shot: 5
metrics:
- type: acc
value: 51.03
name: accuracy
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FCogitoZ
name: Open LLM Leaderboard
CogitoZ - 32B
Model Overview
CogitoZ - 32B is a state-of-the-art large language model fine-tuned to excel in advanced reasoning and real-time decision-making tasks. This enhanced version was trained using Unsloth, achieving a 2x faster training process. Leveraging Hugging Face's TRL (Transformers Reinforcement Learning) library, CogitoZ combines efficiency with exceptional reasoning performance.
- Developed by: Daemontatox
- License: Apache 2.0
- Base Model: Qwen/QwQ-32B-Preview
- Finetuned To: Daemontatox/CogitoZ
Key Features
- Fast Training: Optimized with Unsloth, achieving a 2x faster training cycle without compromising model quality.
- Enhanced Reasoning: Utilizes advanced chain-of-thought (CoT) reasoning for solving complex problems.
- Quantization Ready: Supports 8-bit and 4-bit quantization for deployment on resource-constrained devices.
- Scalable Inference: Seamless integration with text-generation-inference tools for real-time applications.
Intended Use
Primary Use Cases
- Education: Real-time assistance for complex problem-solving, especially in mathematics and logic.
- Business: Supports decision-making, financial modeling, and operational strategy.
- Healthcare: Enhances diagnostic accuracy and supports structured clinical reasoning.
- Legal Analysis: Simplifies complex legal documents and constructs logical arguments.
Limitations
- May produce biased outputs if the input prompts contain prejudicial or harmful content.
- Should not be used for real-time, high-stakes autonomous decisions (e.g., robotics or autonomous vehicles).
Technical Details
- Training Framework: Hugging Face's Transformers and TRL libraries.
- Optimization Framework: Unsloth for faster and efficient training.
- Language Support: English.
- Quantization: Compatible with 8-bit and 4-bit inference modes for deployment on edge devices.
Deployment Example
Using Hugging Face Transformers:
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "Daemontatox/CogitoZ"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)
prompt = "Explain the Pythagorean theorem step-by-step:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Optimized Inference:
Install the transformers and text-generation-inference libraries. Deploy on servers or edge devices using quantized models for optimal performance. Training Data The fine-tuning process utilized reasoning-specific datasets, including:
MATH Dataset: Focused on logical and mathematical problems.
Custom Corpora: Tailored datasets for multi-domain reasoning and structured problem-solving.
Ethical Considerations
Bias Awareness -> The model reflects biases present in the training data. Users should carefully evaluate outputs in sensitive contexts.
Safe Deployment -> Not recommended for generating harmful or unethical content.
Acknowledgments
This model was developed with contributions from Daemontatox and the Unsloth team, utilizing state-of-the-art techniques in fine-tuning and optimization.
Open LLM Leaderboard Evaluation Results
Detailed results can be found here! Summarized results can be found here!
| Metric | Value (%) |
|---|---|
| Average | 38.36 |
| IFEval (0-Shot) | 39.67 |
| BBH (3-Shot) | 53.89 |
| MATH Lvl 5 (4-Shot) | 46.30 |
| GPQA (0-shot) | 19.35 |
| MuSR (0-shot) | 19.94 |
| MMLU-PRO (5-shot) | 51.03 |