Skip to content
Other

Caracal Instruct v2

an instruction-tuned LoRA adapter for an African LLM

2 min readupdated

Caracal Instruct v2 is a PEFT LoRA adapter (a lightweight add-on for fine-tuning without modifying the entire model's weights) by theophilusowiti built on top of the Caracal_GPT base language model. It turns the base model into a conversational assistant capable of following instructions in the ChatML format. It generates responses across 9 languages out of the box: English, Swahili, Amharic, Luganda, Kinyarwanda, Hausa, Yoruba, Igbo, and Lingala. Ideal for developers building multilingual chatbots and researchers working with low-resource African languages.

Key features

  • Supports 9 languages: English, Swahili, Amharic, Luganda, Kinyarwanda, Hausa, Yoruba, Igbo, and Lingala.
  • Optimized for dialogue and instruction following using the ChatML template.
  • Trained with SFT (Supervised Fine-Tuning) via the trl library on the Caracal_Instruct_Dataset dataset.
  • Released under the open Apache 2.0 license, allowing free and commercial use.

What you can do

  • Integrate it into a chatbot or helpdesk service to communicate with users in African languages and English.
  • Use it in NLP research as a ready-to-go baseline adapter for evaluating generation quality on low-resource languages.
  • Further fine-tune it for specialized downstream tasks on top of the Caracal_GPT and Caracal_instruct_v2 stack via the peft library.

Example

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

tok = AutoTokenizer.from_pretrained("theophilusowiti/Caracal_GPT")
base = AutoModelForCausalLM.from_pretrained("theophilusowiti/Caracal_GPT", torch_dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(base, "theophilusowiti/Caracal_instruct_v2")

inputs = tok.apply_chat_template([{"role": "user", "content": "Who is the president of Kenya?"}], return_tensors="pt", add_generation_prompt=True).to(model.device)
print(tok.decode(model.generate(inputs, max_new_tokens=100)[0], skip_special_tokens=True))
Read moreDetails
Architecture: Causal Language Model with a PEFT LoRA adapter based on theophilusowiti/Caracal_GPT.
Training: SFTTrainer from trl on the theophilusowiti/Caracal_Instruct_Dataset dataset.
Metrics and VRAM requirements are not specified in the model card (training logs were tracked in Weights & Biases).
Discuss Telegram 0 30

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.7 GB · ready files2.4 GB
  • FP8 / INT8recommendedweights 1.1 GB1.7 GB
  • 4-bitweights 0.8 GB1.4 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model Caracal_GPT.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # theophilusowiti/Caracal_instruct_v2
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download theophilusowiti/Caracal_GPT --local-dir ~/models/Caracal_GPT
    hf download theophilusowiti/Caracal_instruct_v2 --local-dir ~/models/Caracal_instruct_v2
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Caracal_GPT
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

Requires the base model

theophilusowiti/Caracal_GPT

This model is a LoRA / adapter. Download the base model first, then load this one on top.

  1. InkubaLM-0.4Bfamily root
  2. Caracal_GPT
  3. Caracal_instruct_v2

New posts about versions of InkubaLM-0.4B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 2All on Hugging Face
Fine-tunes 11All on Hugging Face
LoRAs and adapters 3All on Hugging Face

About the model

theophilusowiti/Caracal_instruct_v2
Author
theophilusowiti
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

300

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Other

DAVID-Net

DAVID-Net
Other

ec-esm

ec-esm