v2.2 · 2026-06-12
StarCoder2 15B
starcoder2-15b
StarCoder2 15B is a 15B code model profile for local AI planning. This page records provider, license, context, quantization, hardware-fit estimates, setup hints, source links, and caveats so readers do not choose a model by name alone.
PROVIDER
BigCode / NVIDIA
FAMILY
StarCoder2
MODEL TYPE
code
PARAMETERS
15B
MODALITIES
text
ARCHITECTURE
Transformer decoder with grouped-query attention, sliding-window attention, and Fill-in-the-Middle objective
CONTEXT WINDOW
16,384 tokens
TRAINING TOKENS
4.0T
RELEASE DATE
2024-02-28
GGUF REPOSITORIES
second-state/StarCoder2-15B-GGUF community conversion
Community GGUF conversion/quantization; file table used for Q4_K_M, Q5_K_M, and Q8_0 sizes.
SETUP HINTS
OLLAMA
Use `ollama run starcoder2:15b` for local completion-style use.LM STUDIO
Search for `second-state/StarCoder2-15B-GGUF` and choose Q4_K_M for the first local run.
LLAMA.CPP
Use `llama-cli -hf second-state/StarCoder2-15B-GGUF:Q4_K_M -p '<code prefix>'`QUANTIZATION FILE SIZES (GGUF)
Q4_K_M
9.86 GB
Q5_K_M
11.4 GB
Q8_0
17 GB
Quantization note: Q4_K_M
RAM / VRAM ESTIMATES
MIN RAM
14 GB
COMFORTABLE RAM
24 GB
MIN VRAM
12 GB
COMFORTABLE VRAM
24 GB
These are conservative local-inference estimates, not official hardware requirements.
HARDWARE FIT
CPU only (no GPU)
possible_but_slow
8 GB RAM
not_recommended
16 GB RAM
possible
32 GB RAM
good
8 GB VRAM
not_recommended
12 GB VRAM
possible
24 GB VRAM
good
Apple Silicon (unified memory)
16GB_unified_memory_possible_32GB_recommended
Hardware fit values are conservative local-inference estimates based on GGUF size plus runtime and KV-cache overhead. Actual requirements depend on context length, quantization, runtime, GPU offload, and other running apps.
BEST FOR
·Local code completion and fill-in-the-middle workflows
AVOID IF
·You need an instruction/chat model; the official model card says the base model is not an instruction model.
CAVEATS
·The official StarCoder2 base checkpoints are code-focused but not instruction-tuned; use a separate instruct checkpoint only if your catalog adds it as a separate model.
·RAM and VRAM estimates are conservative estimates from GGUF file sizes plus local runtime and KV-cache overhead.
SOURCE URLS
FIELD EVIDENCE
parameterSizeBhuggingface.co/bigcode/starcoder2-15b
contextWindowTokenshuggingface.co/bigcode/starcoder2-15b
trainingTokenshuggingface.co/bigcode/starcoder2-15b
q4FileSizeGbhuggingface.co/second-state/StarCoder2-15B-GGUF
q5FileSizeGbhuggingface.co/second-state/StarCoder2-15B-GGUF
q8FileSizeGbhuggingface.co/second-state/StarCoder2-15B-GGUF
LANGUAGES
600+ programming languages
CAPABILITIES
code tuned