Model library / menezesbruno
Manacá-1B is a 1.7 billion-parameter Llama-style decoder-only model pretrained for Brazilian Portuguese.
Parameters in total
Context length in tokens, as published
Weights at the released precision
Licence declared on the repository
Overview
Manacá-1B is a base, decoder-only transformer trained from scratch for Brazilian Portuguese. The publisher, menezesbruno, describes it as a research-grade language model for text modelling, continuation and as a starting point for fine-tuning or instruction-tuning, but it is not instruction-tuned or aligned for chat.
The model contains roughly 1.7 billion parameters, uses a 4,096-token context window and is stored in bfloat16 format. It is released under a CC-BY-4.0 licence, allowing commercial use with attribution. Deployment will require enough GPU memory to hold the model weights and context, typically a few gigabytes of VRAM.
Specification
| Repository | menezesbruno/manaca-1b-base |
| Publisher | menezesbruno |
| Published | 31 August 2026 |
| Architecture | LlamaForCausalLM |
| Model type | llama |
| Parameters | 1,722,951,680 |
| Layers | 24 |
| Hidden size | 2,048 |
| Attention heads | 32, 8 key/value heads |
| Context length | 4,096 tokens |
| Vocabulary | 64,128 tokens |
| Weight format | bfloat16 |
| Licence | cc-by-4.0 |
Memory
Parameter count multiplied by the bytes each format uses. The key-value cache comes on top.
| Precision | Weights | Notes |
|---|---|---|
| bf16 | 3.2 GiB | Full precision as released |
| fp8 | 1.6 GiB | 8-bit, near-lossless on most models |
| int4 | 0.8 GiB | 4-bit, smallest footprint |
The cache costs 48 KiB per token at 16-bit, so the full 4,096-token context of one request needs about 0.2 GiB. Concurrency multiplies that number, not the weights.
Hardware
At the released precision, with 90% of the card given to the server and 2.5 GiB kept for the runtime. The context column is what is left for the cache on that setup.
| GPU | Cards needed | Free for cache | Context it holds |
|---|---|---|---|
| RTX 4090 24GB | 1 | 15.9 GiB | the full 4,096 tokens |
| RTX 5090 32GB | 1 | 23.1 GiB | the full 4,096 tokens |
| L40S 48GB | 1 | 37.5 GiB | the full 4,096 tokens |
| A100 80GB | 1 | 66.3 GiB | the full 4,096 tokens |
| H100 80GB | 1 | 66.3 GiB | the full 4,096 tokens |
| RTX PRO 6000 96GB | 1 | 80.7 GiB | the full 4,096 tokens |
| H200 141GB | 1 | 121.2 GiB | the full 4,096 tokens |
Serving
A vLLM launch line for the shape above. Check the model card for a runtime the publisher recommends.
vllm serve menezesbruno/manaca-1b-base \ --max-model-len 4096 \ --gpu-memory-utilization 0.90
Questions
The weights take about 3 GiB at the released precision. That fits on one RTX 4090 24GB, which leaves roughly 16 GiB for the key-value cache.
The repository declares cc-by-4.0. Read the licence text before commercial use: the name alone does not tell you what is allowed.
The configuration allows 4,096 tokens. Whether the whole window is usable depends on the memory left for the cache, which the hardware table works out per card.
This page is a reference, and Aiqre does not serve this model today. We deploy open-weight models on dedicated EU hardware on request, so if you want an OpenAI-compatible endpoint for it, email [email protected] with the model name and roughly what volume you expect.
Also in the library
MiniCPM5-2B is a 2-billion-parameter dense transformer aimed at on-device and resource-constrained deployments.
K2-Horizon-0.9B is a 1.1 billion-parameter dense decoder-only model with a 128 k token context window.
MiniCPM5-2B is a 2.5-billion-parameter dense Llama-style causal language model released by openbmb.
Spark-X2.5-1.7B-Base is a 1.7 billion-parameter general-purpose language model with a native 1 M-token context window.
Hosting
Tell us the model and roughly what volume you expect. We reply with a price and an OpenAI-compatible endpoint, with your prompts never stored.