---
language:
- en
license: apache-2.0
base_model: LiquidAI/LFM2.5-1.2B-Instruct
tags:
- MAM
- memory-augmented
- parametric-memory
---

# MAM (Memory As a Model) Fine-tuned Model

This model was trained using the MAM (Memory As a Model) framework, which uses a small model as parametric memory instead of traditional RAG's non-parametric datastore.

## Model Details

- **Base Model**: LiquidAI/LFM2.5-1.2B-Instruct
- **Training Framework**: MAM (Memory As a Model)
- **Training Approach**: Online learning with sequential chunk processing

## Training Data

The model was trained on academic papers, learning to build connections between concepts across different chunks/papers.

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("yungisimon/lfm_offigiri_paper_1_epoch_10")
tokenizer = AutoTokenizer.from_pretrained("yungisimon/lfm_offigiri_paper_1_epoch_10")

# Example: Query the model's accumulated knowledge
prompt = "What is the relationship between attention mechanisms and memory?"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## Citation

If you use this model, please cite the MAM paper.