RichardErkhov/dominguesm_-_mambarim-110m-4bits

Quantization made by Richard Erkhov.

Github

Discord

Request more models

mambarim-110m - bnb 4bits

• Model creator: https://huggingface.co/dominguesm/
• Original model: https://huggingface.co/dominguesm/mambarim-110m/

Original model description:

library_name: transformers language: - pt license: cc-by-4.0 tags: - text-generation - pytorch - LLM - Portuguese - mamba datasets: - nicholasKluge/Pt-Corpus-Instruct inference: parameters: repetition_penalty: 1.2 temperature: 0.8 top_k: 50 top_p: 0.85 max_new_tokens: 150 widget: - text: "O Natal é uma" example_title: Exemplo - text: "A muitos anos atrás, em uma galáxia muito distante, vivia uma raça de" example_title: Exemplo - text: "Em meio a um escândalo, a frente parlamentar pediu ao Senador Silva para" example_title: Exemplo pipeline_tag: text-generation

Mambarim-110M

Model Summary

Mambarim-110M is the first Portuguese language model based on a state-space model architecture (Mamba), not a transformer.

WIP

Details

• Architecture: a Mamba model pre-trained via causal language modeling
• Size: 119,930,880 parameters
• Context length: 2048 tokens
• Dataset: Pt-Corpus Instruct (6.2B tokens)
• Language: Portuguese
• Number of steps: 758,423

This repository has the source code used to train this model.

Intended Uses

WIP

Out-of-scope Use

WIP

Basic usage

You need to install transformers from main until transformers=4.39.0 is released.

pip install git+https://github.com/huggingface/transformers@main

We also recommend you to install both causal_conv_1d and mamba-ssm using:

pip install causal-conv1d>=1.2.0
pip install mamba-ssm

You can use the classic generate API:

>>> from transformers import MambaConfig, MambaForCausalLM, AutoTokenizer
>>> import torch
>>> tokenizer = AutoTokenizer.from_pretrained("dominguesm/mambarim-110m")
>>> model = MambaForCausalLM.from_pretrained("dominguesm/mambarim-110m")
>>> input_ids = tokenizer("O Natal é uma", return_tensors="pt")["input_ids"]
>>> out = model.generate(
    input_ids,
    repetition_penalty=1.2,
    temperature=0.8,
    top_k=50,
    top_p=0.85,
    do_sample=True,
    max_new_tokens=10
)
>>> print(tokenizer.batch_decode(out))
["<s> O Natal é uma data em que as pessoas passam horas de lazer e"]

Benchmarks

Evaluations on Brazilian Portuguese benchmarks were performed using a Portuguese implementation of the EleutherAI LM Evaluation Harness (created by Eduardo Garcia).

Detailed results can be found here

Model	Average	ENEM	BLUEX	OAB Exams	ASSIN2 RTE	ASSIN2 STS	FAQNAD NLI	HateBR	PT Hate Speech	tweetSentBR	Architecture
TeenyTinyLlama-460m	28.86	20.15	25.73	27.02	53.61	13	46.41	33.59	22.99	17.28	LlamaForCausalLM
TeenyTinyLlama-160m	28.2	19.24	23.09	22.37	53.97	0.24	43.97	36.92	42.63	11.39	LlamaForCausalLM
MulaBR/Mula-4x160-v0.1	26.24	21.34	25.17	25.06	33.57	11.35	43.97	41.5	22.99	11.24	MixtralForCausalLM
TeenyTinyLlama-460m-Chat	25.49	20.29	25.45	26.74	43.77	4.52	34	33.49	22.99	18.13	LlamaForCausalLM
manbarim-110m	14.16	18.4	10.57	21.87	16.09	1.89	9.29	15.75	17.77	15.79	MambaForCausalLM
GloriaTA-3B	4.09	1.89	3.2	5.19	0	2.32	0.26	0.28	23.52	0.19	GPTNeoForCausalLM