BigCodeStarCoderStarcoder2ForCausalLM

Starcoder2 15B — Hardware Requirements & GPU Compatibility

ChatCode

Starcoder2 15B is a 16.0B-parameter open language model from BigCode in the StarCoder family. It supports a context window of up to 16,384 tokens. At Q4_K_M it needs about 10.04 GB of VRAM — see which GPUs and Macs can run it below.

16.0B params · dense16K contextbigcode-openrail-mFeb 20244.9K downloads675 likes5.1K quant downloads

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

S

Runs great on a 24 GB GPU

Runs from 12 GB (Q4_K_M · 16K)Best on 24 GB (Q4_K_M · ~65 tok/s · 16K)Macs from 16 GB (Q4_K_M · ~14 tok/s)

bigcode-openrail-m

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Starcoder2 15B

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Recommended

Ollama

one command · needs Ollama installed

$ollama run starcoder2:15b-q4_K_M

Q4_K_M fits fully on the RTX 4090 · 12.8 GB free · 16K context

Download: starcoder2-15b-Q4_K_M.gguf9.86 GBsecond-state

install, OS, other quants

Other ways

each opens step-by-step

Run in cloud

Fits on Tesla P40 (24 GB) (12 GB headroom) · Q4_K_M

Generation speed: ~22 tok/s; generation speed
Cost per 1M output tokens: $1.51; per 1M output tokens

Deployon Vast.aiBest optionfrom $0.12/hr

Compare GPUs →

or

How Much VRAM Does Starcoder2 15B Need?

Select a quantization level; the listed hardware adheres to it.

Quantization	VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
	7.3 GB	—	—	6.2 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Fits: 16K context · ~90 tok/s
	7.5 GB	—	—	7 GB	3-bit small — 3-bit small quantization	Fits: 16K context · ~88 tok/s
	8.3 GB	—	—	8.1 GB	3-bit medium — 3-bit medium quantization	Fits: 16K context · ~79 tok/s
	8.4 GB	—	—	9.1 GB	4-bit · legacy — 4-bit legacy quantization	Fits: 16K context · ~78 tok/s
	8.7 GB	—	—	9 GB	3-bit large — 3-bit large quantization	Fits: 16K context · ~76 tok/s
	9.4 GB	—	—	9.3 GB	4-bit small — 4-bit small quantization	Fits: 16K context · ~69 tok/s
	10.0 GB	—	—	9.9 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Fits: 16K context · ~65 tok/s
	10.4 GB	—	—	11 GB	5-bit · legacy — 5-bit legacy quantization	Fits: 16K context · ~63 tok/s
	11.4 GB	—	—	11 GB	5-bit small — 5-bit small quantization	Fits: 16K context · ~57 tok/s
	11.8 GB	—	—	11.4 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Fits: 16K context · ~55 tok/s
	13.6 GB	—	—	13.1 GB	6-bit, very good — 6-bit quantization, very good quality	Fits: 16K context · ~48 tok/s
	16.4 GB	—	—	17 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Fits: 16K context · ~40 tok/s
	32.4 GB	—	—	~31.9 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Does not fit: needs 33.6 GB

est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

11.2GBat 16K context

Weights: 9.6 GB
KV cache: 1.3 GB
Overhead: 0.3 GB

Fits RTX 4090 (reference) · 12.8 GB free · ~65 tok/s

≥50 tok/s needs ≥772 GB/s · RTX 5090 ~116, RTX 3090 Ti ~65

Which Hardware Can Run Starcoder2 15B?

Q4_K_M · 10.0 GB

Starcoder2 15B (Q4_K_M) requires 10.0 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 14+ GB is recommended. Using the full 16K context window can add up to 1.2 GB, bringing total usage to 11.2 GB. 38 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3080 Ti. 150 devices with unified memory can run Starcoder2 15B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Mini M2 Pro (16 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8 GB · RTX 4060Doesn't fit — smallest quant needs 8.4 GB8 GB tier →

12 GB · RTX 3060 12GBQ4_K_M · 11.2 GB · ~23 tok/s · 16K11.2~2316K12 GB tier →

16 GB · RTX 5060 Ti 16GBQ4_K_M · 11.2 GB · ~29 tok/s · 16K11.2~2916K16 GB tier →

24 GB · RTX 4090referenceQ4_K_M · 11.2 GB · ~65 tok/s · 16K11.2~6516K24 GB tier →

Mac 32 GB · Mac Mini M4Q4_K_M · 11.2 GB · ~8 tok/s · 16K11.2~816KMacs →

Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 11.2 GB · ~38 tok/s · 16K11.2~3816KMacs →

How good is Starcoder2 15B?

llmrun Score blends public benchmark boards into one number; 100 = the average open model.

Not enough data yet for an llmrun Score.

A Score needs results on enough independent boards. How the Composite Score is built

Benchmarks

MMLU
Epoch AI · Feb 2024Lab-reported
64.1%
#40/85 open
Lab-reported
GSM8K
Epoch AI · Feb 2024Lab-reported
57.7%
#27/62 open
Lab-reported
ARC Challenge
Epoch AI · Feb 2024Lab-reported
47.2%
#34/56 open
Lab-reported
WinoGrande
Epoch AI · Feb 2024Lab-reported
64.3%
#38/51 open
Lab-reported

All 4 Starcoder2 15B benchmarks

Alternatives to Starcoder2 15B

StarCoder family · 5 models →

Similar models

6tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
Apriel 1.6 15B ThinkerServiceNow-AI	—	13.2 GB	~67	same size
Kimi VL A3B InstructMoonshot AI	—	10.7 GB	~208	same size
LLaDA2.1 MiniInclusion AI	—	33.5 GB	—needs > 24 GB	same size
LLaDA2.0 MiniInclusion AI	—	10.7 GB	~406	same size
Moonlight 16B A3BMoonshot AI	—	10.1 GB	~229	same size
DeepSeek Coder v2 Lite BaseDeepSeek	—	10.2 GB	~243	same size

Fine-tunes & variants(1)community builds of Starcoder2 15B

Starcoder2 15B Instruct v0.1

Specifications

Publisher: BigCode
Family: StarCoder
Parameters: 16.0B
Architecture: Starcoder2ForCausalLM
Context Length: 16,384 tokens
Vocabulary Size: 49,152
Release Date: 2024-02-20
License: bigcode-openrail-m

Resources

Frequently Asked Questions

How much VRAM does Starcoder2 15B need?

Starcoder2 15B requires 10.0 GB of VRAM at Q4_K_M, or 32.4 GB at BF16. Full 16K context adds up to 1.2 GB (11.2 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 16.0B × 4.8 bits ÷ 8 = 9.6 GB

KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 1.6 GB (at full 16K context)

VRAM usage by quantization

Q4_K_M

10.0 GB

Q4_K_M + full context

11.2 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 4090 run Starcoder2 15B?

Yes, at Q8_0 (16.4 GB) or lower. Higher quantizations like BF16 (32.4 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.

What's the best quantization for Starcoder2 15B?

For Starcoder2 15B, Q4_K_M (10.0 GB) offers the best balance of quality and VRAM usage. Q5_0 (10.4 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 7.3 GB.

VRAM requirement by quantization

Q2_K

7.3 GB~75%

Q4_0

8.4 GB~85%

Q4_K_M ★

10.0 GB~89%

Q5_0

10.4 GB~90%

Q5_K_M

11.8 GB~92%

BF16

32.4 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Starcoder2 15B on a Mac?

Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run Starcoder2 15B. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 10.0 GB of usable unified memory (RAM minus macOS overhead).

Can I run Starcoder2 15B locally?

Yes — Starcoder2 15B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Starcoder2 15B?

At Q4_K_M, Starcoder2 15B can reach ~478 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~65 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 10.04 × 0.65 = ~518 tok/s

Estimated speed at Q4_K_M (10.0 GB)

NVIDIA B200

~518 tok/s

NVIDIA GeForce RTX 4090

~65 tok/s

NVIDIA B300

~518 tok/s

AMD Instinct MI350X

~478 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of Starcoder2 15B?

At Q4_K_M, the download is about 9.57 GB. The full-precision BF16 version is 31.92 GB. The smallest option (Q2_K) is 6.78 GB.

Which GPUs can run Starcoder2 15B?

38 consumer GPUs can run Starcoder2 15B at Q4_K_M. Top options include AMD Radeon RX 6800, AMD Radeon RX 6800 XT, AMD Radeon RX 6900 XT, AMD Radeon RX 6700 XT. 26 GPUs have plenty of headroom for comfortable inference.

Which devices can run Starcoder2 15B?

150 devices with unified memory can run Starcoder2 15B at Q4_K_M (10.0 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.

Run Starcoder2 15B

11.2 GB VRAM at 16K context · fits the RTX 4090 (12.8 GB free)

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run starcoder2:15b-q4_K_M
Official Ollama library build. For the exact Q4_K_M file from second-state/StarCoder2-15B-GGUF: ollama run hf.co/second-state/StarCoder2-15B-GGUF:Q4_K_M
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/second-state/StarCoder2-15B-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/second-state/StarCoder2-15B-GGUF@Q4_K_M
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf second-state/StarCoder2-15B-GGUF:Q4_K_M -c 16384 -ngl 99 --port 8080

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/second-state/StarCoder2-15B-GGUF:Q4_K_M
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf second-state/StarCoder2-15B-GGUF:Q4_K_M -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1.
$vllm serve bigcode/starcoder2-15b --max-model-len 16384
Runs the original weights, not the Q4_K_M GGUF; needs ~35 GB of GPU memory.

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$second-state/StarCoder2-15B-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
Tesla P40 (24 GB)CheapestCheapest12 GB free	~22 tok/s	$1.51	Deploy on Vast.aifrom $0.12/hr (opens in new tab)	24 GB · 12 GB free · ~22 tok/s · $1.51 / 1M tokens
RTX A5000 (24 GB)12 GB free	~50 tok/s	$0.89	Deploy on RunPodfrom $0.16/hr (opens in new tab)	24 GB · 12 GB free · ~50 tok/s · $0.89 / 1M tokens
RTX A4000 (16 GB)4 GB free	~29 tok/s	$1.63	Deploy on RunPodfrom $0.17/hr (opens in new tab)	16 GB · 4 GB free · ~29 tok/s · $1.63 / 1M tokens
RTX 5060 Ti 16GB4 GB free	~29 tok/s	$1.80	Deploy on Vast.aifrom $0.19/hr (opens in new tab)	4 GB free · ~29 tok/s · $1.80 / 1M tokens
RTX 4060 Ti 16GB4 GB free	~19 tok/s	$2.83	Deploy on Vast.aifrom $0.19/hr (opens in new tab)	4 GB free · ~19 tok/s · $2.83 / 1M tokens
RTX 4000 Ada (20 GB)8 GB free	~23 tok/s	$2.38	Deploy on RunPodfrom $0.20/hr (opens in new tab)	20 GB · 8 GB free · ~23 tok/s · $2.38 / 1M tokens
RTX A5000 (24 GB)12 GB free	~50 tok/s	$1.20	Deploy on Vast.aifrom $0.21/hr (opens in new tab)	24 GB · 12 GB free · ~50 tok/s · $1.20 / 1M tokens
RTX 3090 (24 GB)12 GB free	~61 tok/s	$1.01	Deploy on RunPodfrom $0.22/hr (opens in new tab)	24 GB · 12 GB free · ~61 tok/s · $1.01 / 1M tokens

Compare GPUs →