TWC - Together We Check

A collection of tools for testing and validating LLM serving infrastructure.

Tools

Directory	Description	vLLM	SGLang
model-sanity/	Multi-model sanity checker — validates model serving across v1/chat/completions and v1/completions endpoints	✅	❌
n-run-stability/	Universal stress tester — repeatedly launches and tests a model to catch intermittent failures	✅	✅

Quick Start

# Model sanity check (vLLM)
cd model-sanity
export HF_TOKEN='hf_your_token_here'
./run_sanity_check.sh

# Stability stress test (Universal - supports SGLang and vLLM)
cd n-run-stability
export HF_TOKEN='hf_your_token_here'

# Default config (SGLang)
./stress_test.sh

# Use a preset
./stress_test.sh --config presets/vllm-llama-rocm.yaml

# Override options
./stress_test.sh --loops 50 --framework vllm --image vllm/vllm-openai:latest

See individual tool READMEs for detailed usage.