litmus-lab 0.3.3

A CLI benchmarking framework for LLM inference across FP16/INT8/INT4/HQQ/Quanto/AWQ/GPTQ quantization

⚗️ litmus-lab

Benchmark your LLM across quantization formats and backends on your own GPU.
Get one verdict — which precision, which backend, what it costs.

---

What it does

litmus-lab runs your model through HuggingFace (FP16, INT8, NF4, FP4, NF4+double-quant, HQQ, Quanto INT8/INT4, plus AWQ/GPTQ against a pre-quantized checkpoint) and vLLM (FP16, BitsAndBytes, FP8, AWQ, GPTQ) in isolated passes — measuring VRAM, throughput, latency and perplexity on your actual hardware — then outputs a single deployment recommendation.

┏━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Mode            ┃ VRAM (MB) ┃ TPS      ┃ TTFT     ┃ Perplexity ┃
┡━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━━┩
│ HF · FP16       │ 7297.83   │ 32.69    │ 0.030s   │ 5.64       │
│ HF · INT8       │ 3846.47   │ 14.26    │ 0.083s   │ 5.81       │
│ HF · NF4        │ 2334.96   │ 25.98    │ 0.069s   │ 7.34       │
│ vLLM · FP16     │ 12687.31  │ 111.68   │ 0.448s   │ 5.65       │
└─────────────────┴───────────┴──────────┴──────────┴────────────┘

  Recommendation  Deploy vLLM · FP16
                  3.4× faster than HF · PPL delta 0.01
                  Memory-constrained? HF · NF4 saves 4963 MB (PPL +1.70)

---

Installation

# HF Transformers backend — FP16, INT8, NF4, FP4, NF4+double-quant, HQQ, Quanto INT8/INT4
pip install litmus-lab

# + vLLM backend (FP16, BitsAndBytes, FP8)
pip install "litmus-lab[vllm]"

# + AI recommendations via Groq
pip install "litmus-lab[ai]"

# + AWQ / GPTQ pre-quantized checkpoint support (HF + vLLM)
# Note: gptqmodel builds from source and can fail depending on your CUDA toolchain —
# kept separate from `all` on purpose so a failed build here never breaks the base CLI.
pip install "litmus-lab[awq,gptq]"

# vLLM + AI recommendations (guaranteed to install cleanly — no source builds)
pip install "litmus-lab[all]"

Note: vLLM requires Linux or WSL2. It does not run on Windows natively.

---

Quick start

litmus-lab \
  --model microsoft/Phi-3-mini-4k-instruct \
  --prompt "Explain the transformer architecture" \
  --backend all

---

Usage

Flags

Flag	Description	Default
--model	HuggingFace model repo ID	required
--prompt	Input prompt	required
--backend	hf · vllm · all	hf
--token	HuggingFace token for gated models	None
--awq-model	Pre-quantized AWQ checkpoint repo (e.g. TheBloke/Mistral-7B-v0.1-AWQ). Enables the HF/vLLM AWQ rows	None
--gptq-model	Pre-quantized GPTQ checkpoint repo (e.g. TheBloke/Mistral-7B-v0.1-GPTQ). Enables the HF/vLLM GPTQ rows	None
--version	Print version and exit	—

Backend modes

Mode	What runs
hf	HF Transformers · FP16, INT8, NF4, FP4, NF4+double-quant, HQQ, Quanto INT8/INT4 — plus AWQ/GPTQ if --awq-model/--gptq-model are given
vllm	vLLM · FP16, BitsAndBytes, FP8 — plus AWQ/GPTQ if --awq-model/--gptq-model are given
all	Both of the above in sequence

Any mode that hits a missing optional dependency or an unsupported GPU (e.g. FP8 on pre-Hopper hardware) prints a yellow skip line and continues — it won't abort the whole run.

Examples

# Compare quantization formats
litmus-lab \
  --model Qwen/Qwen2.5-7B-Instruct \
  --prompt "Explain quantum gravity" \
  --backend hf

# vLLM only
litmus-lab \
  --model mistralai/Mistral-7B-Instruct-v0.3 \
  --prompt "Write a Linux shell script" \
  --backend vllm

# Full benchmark — HF + vLLM side by side
litmus-lab \
  --model meta-llama/Llama-3.1-8B-Instruct \
  --token hf_xxxxxxx \
  --prompt "Explain TCP congestion control" \
  --backend all

# Include AWQ / GPTQ rows against pre-quantized checkpoints
litmus-lab \
  --model teknium/OpenHermes-2.5-Mistral-7B \
  --prompt "Explain the theory of relativity in simple terms" \
  --backend all \
  --awq-model TheBloke/OpenHermes-2.5-Mistral-7B-AWQ \
  --gptq-model TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ

---

AI Recommendations

Set a Groq API key (free tier) to get an AI-powered verdict from llama-3.3-70b-versatile instead of the built-in heuristic:

export GROQ_API_KEY=gsk_...
litmus-lab --model Qwen/Qwen2.5-7B --prompt "Hello" --backend all

No extra flags needed. Falls back to the offline heuristic automatically if the key is missing, rate limit is hit, or there is no internet.

---

Metrics explained

Metric	What it measures
VRAM (MB)	Peak GPU memory — model weights + KV cache, measured as a before/after delta
TPS	Tokens generated per second — generation throughput
TTFT	Time to first token — how quickly the model starts responding
Perplexity	Language quality on WikiText-2. Lower = better. Delta >2.0 vs FP16 signals degradation

---

Supported models

✅ HF backend

Any HuggingFace causal language model loadable via AutoModelForCausalLM:

Family	Models
Meta	Llama 3, 3.1, 3.2, 3.3
Microsoft	Phi-3, Phi-3.5, Phi-4
Alibaba	Qwen2, Qwen2.5
Mistral AI	Mistral 7B, Mixtral 8x7B
Google	Gemma, Gemma 2
DeepSeek	DeepSeek-V2, DeepSeek-V3, DeepSeek-R1
Community	Falcon, Yi, TinyLlama, Vicuna, Zephyr, WizardLM, and most fine-tunes

Gated models (Llama, Gemma) require --token.

Quantization modes run against the base repo on the fly — no separate checkpoint needed — except AWQ/GPTQ:

Mode	Notes
FP16	native precision, no quantization
INT8	bitsandbytes load_in_8bit
NF4	bitsandbytes 4-bit, NormalFloat4
FP4	bitsandbytes 4-bit, FloatingPoint4
NF4 + double quant	NF4 with nested quantization of the quant constants for extra VRAM savings
HQQ	Half-Quadratic Quantization — requires hqq (bundled by default). Skips cleanly if your transformers version hasn't finished migrating HQQ's loading path yet (NotImplementedError)
Quanto INT8 / INT4	optimum-quanto weight-only quantization (bundled by default)
AWQ / GPTQ	requires --awq-model / --gptq-model pointing at a pre-quantized checkpoint, plus pip install litmus-lab[awq,gptq]

✅ vLLM backend

Same model coverage as HF. VRAM is generally higher than HF at the same precision due to the KV cache pool, but throughput is significantly better under concurrent load.

Mode	Notes
FP16	native precision with PagedAttention
BitsAndBytes	on-the-fly quantization of the base checkpoint
FP8	on-the-fly W8A8 — requires a Hopper/Ada-class GPU (H100, L4, RTX 4090+); skips cleanly on older GPUs
AWQ / GPTQ	requires --awq-model / --gptq-model pointing at a pre-quantized checkpoint

❌ Not supported

	Reason
BERT, RoBERTa, encoder-only models	No autoregressive generation
T5, BART, mT5, seq2seq models	Different generation API
Models that exceed GPU VRAM at FP16	No CPU offloading yet
vLLM on Windows	Use Linux or WSL2
Multi-GPU / tensor parallel	Not yet implemented

---

Requirements

• Python 3.10+
• NVIDIA GPU with CUDA support (CPU works but is very slow, and quantization gives smaller wins there)
• NVIDIA driver + CUDA 11.8+ / 12.x for most GPUs. Exception: Blackwell-generation GPUs (RTX 50-series, compute capability 12.x) need CUDA ≥ 12.9 and driver ≥ 575.51 — the vllm backend's FlashInfer kernels fail to initialize below that (RuntimeError: FlashInfer requires GPUs with sm75 or higher / SM 12.x requires CUDA >= 12.9). The hf backend is unaffected and works fine on any CUDA version.
• Disk space for whatever model you benchmark — 7B models are roughly 15-30GB on disk depending on dtype — plus the small WikiText-2 dataset used for perplexity
• --token required for gated models (Llama, Gemma, etc.)
• vLLM backend (--backend vllm / all): Linux or WSL2 only, install with pip install "litmus-lab[vllm]" — does not run natively on Windows
• AWQ/GPTQ pre-quantized modes (--awq-model / --gptq-model): need pip install "litmus-lab[awq,gptq]" and a separately pre-quantized checkpoint repo (e.g. TheBloke/*-AWQ) — these don't run against the base model repo

---

Troubleshooting

vLLM fails to initialize on Blackwell GPUs (RTX 50-series)

Symptoms — one or more of these during --backend vllm / all, often across every vLLM mode at once:

ImportError: vLLM import failed (libcudart.so.13: cannot open shared object file...)
RuntimeError: ... cudaHostGetDevicePointer failed: CUDA driver version is insufficient for CUDA runtime version
Failed to get device capability: SM 12.x requires CUDA >= 12.9.
RuntimeError: FlashInfer requires GPUs with sm75 or higher
RuntimeError: The NVIDIA driver on your system is too old (found version 12080)

Root cause: these are all symptoms of the same mismatch — Blackwell-generation GPUs (compute capability 12.x) need CUDA ≥ 12.9 and driver ≥ 575.51 for vLLM's FlashInfer kernels. Which exact error you see depends on which model/quantization mode happens to hit the mismatch first.

Diagnose:

nvidia-smi   # top-right "CUDA Version" is the max your *driver* supports
python -c "import torch; print(torch.version.cuda, torch.cuda.is_available())"

If nvidia-smi reports less than 12.9, that's confirmed.

Fix, if you control the host (your own machine, not a rented container):

sudo apt update && sudo apt install -y nvidia-driver-575   # or newer — check nvidia.com
sudo reboot
pip uninstall torch torchvision torchaudio vllm -y
pip install torch --index-url https://download.pytorch.org/whl/cu129
pip install vllm

If you're on a rented cloud pod (RunPod/Vast.ai/Lambda/etc.) — the driver belongs to the host, not your container, so nothing you pip/apt install inside it can change it. Options:

• Look for a pod template/image explicitly advertising a newer driver or "Blackwell-ready"/CUDA 12.9 support
• Ask the provider's support directly whether such nodes exist
• Provision a fresh pod — driver versions vary node to node

Careful: upgrading torch alone to a cu129 build without a matching driver makes things worse, not better — torch's own driver check then fails outright (RuntimeError: The NVIDIA driver on your system is too old), and torch.cuda.is_available() silently returns False, so litmus-lab falls back to CPU for everything, including the HF backend that was working fine before. If that happens, revert:

pip uninstall torch torchvision torchaudio -y
pip install torch --index-url https://download.pytorch.org/whl/cu128

In the meantime: --backend hf is unaffected by any of this and works fine on Blackwell GPUs, since it never touches vLLM/FlashInfer.

---

What's being built 🚧

Feature	Status
GGUF / llama.cpp backend (CPU + Mac M-series)	🔨 In progress
Cost prediction (--target-users, --gpu-cost)	🔨 In progress
Concurrency benchmarking (--concurrency 1,4,8,16,32)	📋 Planned
Multi-model comparison in one run	📋 Planned
JSON / HTML report export	📋 Planned
Multi-GPU / tensor parallel benchmarking	📋 Planned

Want to shape the roadmap? Open an issue or join the waitlist for the web platform.

---

License

MIT