Instructions to use bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct

SGLang

How to use bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct with Docker Model Runner:
```
docker model run hf.co/bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct
```

Bactrainus HotpotQA Question Decomposer — Llama 3 8B Instruct

Artifact identity

Status: metadata-only placeholder; no model weights are present
Intended base model: meta-llama/Meta-Llama-3-8B-Instruct
Audited Hub revision: a55a003d5f1b748598b428facd3ffec306674546
Public artifact date: August 2024
Intended role: generate ordered subquestions from a question and selected paragraphs

This repository cannot currently be loaded as a model. At the audited revision it contains only repository metadata and a README. The status is stated explicitly to prevent a placeholder from being mistaken for a released checkpoint.

The intended artifact generation is legacy Llama 3, not the Llama 3.1 question-decomposition experiment described in the revised manuscript.

Intended model behavior

The question decomposer is designed to receive:

the original multi-hop question;
paragraphs selected from the ten HotpotQA candidates;
an instruction to produce a concise ordered list of subquestions.

The subquestions are auxiliary inputs for a sentence selector. They do not replace the original question, do not constitute gold evidence, and must not introduce entities unsupported by the supplied input.

Availability

The following call is expected to fail because no configuration, tokenizer, or weight files are present:

# Not currently available:
# AutoModelForCausalLM.from_pretrained(
#     "bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct"
# )

Do not add loading or performance examples until a verified checkpoint, tokenizer, generation configuration, immutable revision, and run manifest are uploaded together.

Data and lineage boundary

HotpotQA is the source task. The matching canonical training view is question-decomposer-sft, pinned at revision 7f3a1d4d21f22aad7262d8ffd6520f31186b284d. It contains deterministic decomposition targets for all 90,447 training source IDs and joins to the original examples and every other view through source_id.

from datasets import load_dataset

train = load_dataset(
    "bactrianus/bactrainus-hotpotqa",
    "question-decomposer-sft",
    split="train",
    revision="7f3a1d4d21f22aad7262d8ffd6520f31186b284d",
)

This dataset release does not change the repository's availability boundary: no legacy model weights are present, and the canonical serialization is not claimed to be byte-identical to an unreleased 2024 training file.

The revised repository configuration question_decomposer.yaml documents the later Llama 3.1 manuscript setup. It must not be used to claim that a corresponding legacy Llama 3 checkpoint is present here.

Evaluation boundary

There is no loadable artifact and no evaluation result associated with this repository card. The paper discusses question decomposition as an optional module and characterizes its observed average contribution conservatively.

Limitations of the intended task

Generated subquestions can omit necessary reasoning steps or introduce unsupported assumptions.
Decomposition quality is not established by final-answer correctness alone.
Unnecessary subquestions can add latency and context noise.
The intended model operates within selected HotpotQA candidates, not over an open corpus.
A future upload must establish prompt, data, and checkpoint provenance before use.

License and attribution

Any future weights derived from Meta Llama 3 remain subject to the Meta Llama 3 Community License and Acceptable Use Policy.

Meta Llama 3 is licensed under the Meta Llama 3 Community License, Copyright Meta Platforms, Inc. All Rights Reserved.

Built with Meta Llama 3.

HotpotQA-derived data is licensed under CC BY-SA 4.0. Bactrainus code is Apache-2.0 licensed.

Citation

@article{barati2025bactrainus,
  title   = {Bactrainus: Optimizing Large Language Models for Multi-hop Complex Question Answering Tasks},
  author  = {Barati, Iman and Ghafouri, Arash and Minaei-Bidgoli, Behrouz},
  journal = {arXiv preprint arXiv:2501.06286},
  year    = {2025},
  doi     = {10.48550/arXiv.2501.06286},
  url     = {https://arxiv.org/abs/2501.06286}
}

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct

Base model

meta-llama/Meta-Llama-3-8B-Instruct

Finetuned

(1084)

this model

Dataset used to train bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct

Paper for bactrianus/HotpotQA-Question-Decomposition-Llama-3-8B-Instruct

Bactrainus: Optimizing Large Language Models for Multi-hop Complex Question Answering Tasks

Paper • 2501.06286 • Published Jan 10, 2025 • 1