Skip to content

BigCode · 7B · transformer-decoder

2024-02-28 16K context 7B params

Use Cases

code chat

Quantization Options

QuantBitsVRAMQualityStatus
Q4_K_Mrec45.5 GBGood
Q8_089.0 GBExcellent
F161616.0 GBExcellent

About this model

StarCoder2 7B is the mid-sized model in the BigCode StarCoder2 family, offering a strong balance between code generation quality and resource efficiency. Trained on The Stack v2 dataset with over 600 programming languages, it provides noticeably better code quality than the 3B variant while remaining accessible on consumer GPUs. With 16K context support and fill-in-the-middle capabilities, StarCoder2 7B is well-suited for local coding assistance, code completion, and generation tasks. Its permissive license makes it a practical choice for both personal and commercial use.

Benchmarks

35.0
mmlu