Eunoia-4B-mini

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("shvgroups/Eunoia-4B-mini")
model = AutoModelForCausalLM.from_pretrained("shvgroups/Eunoia-4B-mini")

prompt = "Explain photosynthesis step by step."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---


## Training Details

### Training Data

Eunoia-4B-Mini is built on top of the base model’s training corpus and further refined through:

- Instruction-following supervision  
- Reasoning-structured prompts  
- Iterative evaluation and retry loops  
- Goal-decomposition templates  

No private or user data was used in training or refinement.


### Training Procedure

#### Training Hyperparameters

- **Training regime:** Mixed-precision fine-tuning (fp16 / bf16)  
- **Architecture:** Decoder-only transformer with an external reasoning controller  

---

## Evaluation

### Metrics

The model is evaluated primarily on the following qualitative and behavioral metrics:

- Long-horizon coherence  
- Instruction adherence over extended outputs  
- Multi-step reasoning stability  
- Retry and recovery behavior under failure  

Formal benchmark results will be released in future updates.

---

## Environmental Impact

- **Hardware:** NVIDIA GPUs  
- **Training setup:** Research-scale fine-tuning  
- **Carbon impact:** Not formally measured  

---

## Technical Specifications

### Model Architecture and Objective

- **Base transformer:** Qwen3-4B-Instruct  

- **External reasoning modules:**
  - Goal Tree  
  - Execution Gate  
  - Goal Evaluator  
  - Adaptive Goal Mutation Engine  

These components operate outside the core transformer and guide generation iteratively through structured goal management and adaptive control logic.

---

## Citation

If you use this model in academic work, please cite:

### BibTeX