FITMYLLM · JULY 28, 2026 · OPEN METHODOLOGY · COMMUNITY BENCHMARKS
LIVE · UPDATED DAILY
Quantization Options Select your GPU for speed estimates Quant Bits VRAM @ 16K Quality Q3_K_M 4 9.4 GB
8.5 + 0.9 KV
low Q3_K_L 4.3 10.0 GB
9.1 + 0.9 KV
moderate IQ4_XS 4.46 10.3 GB
9.4 + 0.9 KV
moderate Q4_K_S 4.67 10.7 GB
9.8 + 0.9 KV
moderate Q4_K_M 4.89 11.2 GB
10.2 + 0.9 KV
good Q5_K_S 5.57 12.5 GB
11.6 + 0.9 KV
good Q5_K_M 5.7 12.8 GB
11.9 + 0.9 KV
good Q6_K 6.56 14.5 GB
13.6 + 0.9 KV
excellent Q8_0 8.5 18.4 GB
17.4 + 0.9 KV
lossless FP16 16 33.3 GB
32.4 + 0.9 KV
lossless
Select your GPU above to see speed estimates and compatibility for each quantization.
Deploying for a team or in production? Size GPUs, cost & scaling in Enterprise → ▸ READY TO RUN THIS? RENT BY THE HOUR
RENT A GPU AND RUN STARCODER2 15B NOW
Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.
Community Ratings Chat Coding Reasoning Creative Vision Roleplay Agentic
Loading ratings...
Run this model Q3_K_M — 8.5 GB VRAM Q3_K_L — 9.1 GB VRAM IQ4_XS — 9.4 GB VRAM Q4_K_S — 9.8 GB VRAM Q4_K_M — 10.2 GB VRAM Q5_K_S — 11.6 GB VRAM Q5_K_M — 11.9 GB VRAM Q6_K — 13.6 GB VRAM Q8_0 — 17.4 GB VRAM FP16 — 32.4 GB VRAM
Ollama llama.cpp vLLM LM Studio KoboldCpp Jan Docker
▸ Easiest way to get started · Beginners
DOCS ↗ curl -fsSL https://ollama.com/install.sh | shCOPY
$ ollama run starcoder2:15b-q4_K_MCOPY
Downloads and runs automatically. Add --verbose for speed stats.
▸ SETUP GUIDE >_
Auto-setup with fitmyllm CLI Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.
Auto-detect GPU Live tok/s in chat Speed benchmarks 9 inference engines
GPUs that can run this model At Q4_K_M quantization. Sorted by minimum VRAM.
Find the best GPU for StarCoder2 15B
Build Hardware for StarCoder2 15B StarCoder2 15B — top open code model for its size, extensive language support.
Read full model card ▸ COLOPHON FITMYLLM · INDEPENDENT · DATA-DRIVEN
FITMYLLM · EST. 2025 · © 2026
RECOMMENDATIONS FROM PUBLISHED MATH, CORRECTED BY THE COMMUNITY — 30.
▸ SPEC SHEET
StarCoder2 15B — 15.96B Dense. ▸ SPECIFICATIONS
PARAMETERS 15.96B
ARCHITECTURE Dense Transformer
CONTEXT LENGTH 16K tokens
CAPABILITIES coding
RELEASE DATE 2024-02-28
PROVIDER BigCode
FAMILY starcoder ▸ VRAM REQUIREMENTS
QUANT BPW VRAM QUALITY Q3_K_M 4 8.5 GB 88% Q3_K_L 4.3 9.1 GB 90% IQ4_XS 4.46 9.4 GB 92% Q4_K_S 4.67 9.8 GB 93% Q4_K_M 4.89 10.2 GB 94% Q5_K_S 5.57 11.6 GB 96% Q5_K_M 5.7 11.9 GB 96% Q6_K 6.56 13.6 GB 97% Q8_0 8.5 17.4 GB 100% FP16 16 32.4 GB 100%
§ 01 BENCHMARK SCORES
HumanEval 60.4
MMLU-PRO 15.0
MATH 6.0
IFEval 27.8
BBH 20.4
GPQA 3.1
MUSR 2.9
MBPP 65.1
BigCodeBench 37.6
§ 02 RUN COMMAND
Run StarCoder2 15B locally with Ollama — needs 10.2 GB VRAM at Q4_K_M:
$ ollama run starcoder2:15b
§ 03 COMPATIBLE GPUs
30 @ Q4_K_M Feedback 