Instructions to use marki256/resilient_challenge_2026 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use marki256/resilient_challenge_2026 with Transformers:

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

pipe = pipeline("text-generation", model="marki256/resilient_challenge_2026", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("marki256/resilient_challenge_2026", trust_remote_code=True, dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use marki256/resilient_challenge_2026 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "marki256/resilient_challenge_2026"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "marki256/resilient_challenge_2026",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/marki256/resilient_challenge_2026

SGLang

How to use marki256/resilient_challenge_2026 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 "marki256/resilient_challenge_2026" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "marki256/resilient_challenge_2026",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "marki256/resilient_challenge_2026" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "marki256/resilient_challenge_2026",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use marki256/resilient_challenge_2026 with Docker Model Runner:
```
docker model run hf.co/marki256/resilient_challenge_2026
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

resilient_challenge_2026 (Merged LoRA, INT4)

This repository contains the merged model artifact used for the Resilient AI Challenge 2026 submission:

Base: Sarvam 30B MoE (100 experts, drop2mid pruning)
Fine-tuning: LoRA merged into base weights
Quantization: BitsAndBytes 4-bit (NF4) (load_in_4bit: true)

Runtime Requirements

Python 3.10+
vllm with bitsandbytes support
transformers, accelerate, bitsandbytes
4 GPUs recommended for this 30B MoE model

Quick Start (vLLM OpenAI server)

pip install -U vllm transformers accelerate bitsandbytes

# Use the included config file
python -m vllm.entrypoints.openai.api_server --config vllm_config.yaml

If needed, you can still override with explicit flags:

python -m vllm.entrypoints.openai.api_server \
  --model marki256/resilient_challenge_2026 \
  --trust-remote-code \
  --tensor-parallel-size 1 \
  --pipeline-parallel-size 4 \
  --max-model-len 1024 \
  --gpu-memory-utilization 0.90 \
  --quantization bitsandbytes \
  --dtype float16

Quick Test

curl http://127.0.0.1:8000/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "marki256/resilient_challenge_2026",
    "messages": [{"role": "user", "content": "Explain residual recovery in two bullet points."}],
    "temperature": 0.0,
    "max_tokens": 128
  }'

Notes

This is a merged model artifact (not adapter-only).
Quantized weights are intended to be loaded using bitsandbytes 4-bit.
If your deployment environment prefers adapter runtime loading, use a base+adapter packaging instead of this merged repo.

Downloads last month: -

Safetensors

Model size

23B params

Tensor type

F32