capablecodingagentic-codingtool-uselong-context

DeepSeek Coder V2 Lite Instruct

Useful lightweight coding MoE; shown without a composite quality rank.

Model profile

Parameters16BDense or fallback
Maximum context131,072
LicenseDeepSeek Model License
Released2024-06-17

32K / one-agent memory trace

Using GGUF Q4_K_M. This is fit math, not a fixed “billions to GB” lookup.

Quantized weights8.38 GB+ 0.34 GB format overhead
KV cache9.4372 GBfallback
Total with workspace + safety21.34 GB48 GB recommended RAM
The model uses an MoE/MLA architecture; this catalog does not force it into the standard GQA KV-cache equation.

Architecture metadata

Fallback estimate: The model uses an MoE/MLA architecture; this catalog does not force it into the standard GQA KV-cache equation.

Evidence records

No cataloged benchmark is sufficiently comparable. The model remains unranked on unsupported dimensions.