|
--- |
|
language: |
|
- en |
|
license: apache-2.0 |
|
tags: |
|
- open-source |
|
- code |
|
- math |
|
- chemistry |
|
- biology |
|
- text-generation |
|
- question-answering |
|
pipeline_tag: text-generation |
|
--- |
|
|
|
# OpenCerebrum-2.0-7B |
|
|
|
OpenCerebrum-2.0-7B is an open-source language model fine-tuned from the alpindale/Mistral-7B-v0.2-hf base model on a diverse dataset aimed at replicating capabilities of Aether Research's proprietary Cerebrum model. |
|
|
|
The model was fine-tuned with SFT and DPO on approximately 7,000 examples across 15 data sources spanning coding, math, science, multi-turn conversation, RAG, reasoning, and general instruction-following. The goal was to assemble public datasets that could help the model achieve strong performance on benchmarks where Cerebrum excels. |
|
|
|
## Model Details |
|
|
|
- **Base Model:** alpindale/Mistral-7B-v0.2-hf |
|
- **Parameters:** 7 billion |
|
- **Fine-Tuning Dataset Size:** ~7,000 examples |
|
- **Fine-Tuning Data:** Advanced in-house curation techniques at Cognitive Computations, with 15 different data sources for DPO and SFT. |
|
- **Language:** English |
|
- **License:** Apache 2.0 |
|
|
|
## Quants |
|
|
|
### EXL2 [@bartowski](https://huggingface.co/bartowski/) |
|
|
|
- https://huggingface.co/bartowski/OpenCerebrum-2.0-7B-exl2 |
|
|
|
### GGUF [@bartowski](https://huggingface.co/bartowski/) |
|
|
|
- https://huggingface.co/bartowski/OpenCerebrum-2.0-7B-GGUF |
|
|
|
## Intended Use |
|
|
|
OpenCerebrum-2.0-7B is intended to be a powerful open-source model for coding, math, science, and general question-answering and text generation tasks. Its diverse fine-tuning data aims to equip it with broad knowledge and reasoning capabilities. |
|
|
|
However, as an open-source replica trained on a subset of data compared to the original Cerebrum, it may not match Cerebrum's full performance. Additionally, biases and limitations of the fine-tuning data may be reflected in the model's outputs. |
|
|
|
## Limitations and Biases |
|
|
|
- The model may have biases and limitations inherited from its fine-tuning datasets. Thorough testing is needed to characterize these. |
|
- As the model is based on a 7B parameter model, it has computational and memory constraints compared to larger models. |
|
|
|
## Evaluations |
|
|
|
| Tasks |Version|Filter|n-shot|Metric|Value | |Stderr| |
|
|--------------|------:|------|-----:|------|-----:|---|-----:| |
|
|truthfulqa_mc2| 2|none | 0|acc |0.5182|± |0.0152| |
|
|ai2_arc |N/A |none | 0|acc |0.7060|± |0.0073| |
|
| | |none | 0|acc_norm|0.7049|± |0.0074| |
|
| - arc_challenge | 1|none | 0|acc |0.5000|± |0.0146| |
|
| | |none | 0|acc_norm|0.5299|± |0.0146| |
|
| - arc_easy | 1|none | 0|acc |0.8077|± |0.0081| |
|
| | |none | 0|acc_norm|0.7912|± |0.0083| |
|
|agieval_nous |N/A |none | 0|acc |0.3778|± |0.0093| |
|
| | |none | 0|acc_norm|0.3574|± |0.0093| |
|
| - agieval_aqua_rat | 1|none | 0|acc |0.2402|± |0.0269| |
|
| | |none | 0|acc_norm|0.2205|± |0.0261| |
|
| - agieval_logiqa_en | 1|none | 0|acc |0.3164|± |0.0182| |
|
| | |none | 0|acc_norm|0.3656|± |0.0189| |
|
| - agieval_lsat_ar | 1|none | 0|acc |0.2130|± |0.0271| |
|
| | |none | 0|acc_norm|0.1913|± |0.0260| |
|
| - agieval_lsat_lr | 1|none | 0|acc |0.4078|± |0.0218| |
|
| | |none | 0|acc_norm|0.3647|± |0.0213| |
|
| - agieval_lsat_rc | 1|none | 0|acc |0.4981|± |0.0305| |
|
| | |none | 0|acc_norm|0.4498|± |0.0304| |
|
| - agieval_sat_en | 1|none | 0|acc |0.6650|± |0.0330| |
|
| | |none | 0|acc_norm|0.5922|± |0.0343| |
|
| - agieval_sat_en_without_passage| 1|none | 0|acc |0.4612|± |0.0348| |
|
| | |none | 0|acc_norm|0.3932|± |0.0341| |
|
| - agieval_sat_math | 1|none | 0|acc |0.3273|± |0.0317| |
|
| | |none | 0|acc_norm|0.2818|± |0.0304| |