DeepSeek-Coder-V2-Lite-Instruct GGUF size and VRAM requirements
bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF is a large code-focused language model with 15.71 billion parameters, built on the deepseek2 architecture. It is released under the other license and has been downloaded 169,643 times.
Under the hood it uses 27 transformer layers, a hidden size of 2,048, 16 attention heads.
To run bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF locally at a 4,096-token context, its quantized versions need between 7.2 GB (IQ2_XS, lowest quality) and 60.16 GB (F32, highest quality) of memory, weights plus KV cache and a system margin included. Context length drives that memory directly: at 4,096 tokens the KV cache for DeepSeek-Coder-V2-Lite-Instruct-GGUF is about 0.84 GB, rising to roughly 33.75 GB at its full 163,840-token context. Shorter prompts free up memory for a higher-quality quantization.
For most users the best balance is Q2_K, needing about 7.63 GB. That means bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF fits entirely in the VRAM of an 8 GB GPU or larger, running fully on the GPU.
Available GGUF quantizations for bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF include IQ2_XS, IQ2_S, IQ2_M, Q2_K, IQ3_XXS, IQ3_XS, Q3_K_S, IQ3_M, Q3_K_M, Q3_K_L, IQ4_XS, Q4_K_S, Q4_K_M, Q4_K_L, Q5_K_S, Q5_K_M, Q5_K_L, Q6_K, Q6_K_L, Q8_0, Q8_0_L, F32. The model supports a native context length of up to 163,840 tokens; a longer context grows the KV cache and the memory needed.
GGUF file size and memory by quantization
Compare real GGUF weight sizes, estimated KV cache and total memory for Q4, Q5, Q8 and every quantization published in this repository.
| Quant. | Bits | Quality | Weights | KV | Total | Speed~ | Verdict |
|---|---|---|---|---|---|---|---|
| IQ2_XS | 3.04 | Low | 5.56 GB | 0.84 GB | 7.2 GB | 72.0 t/s | Fits in VRAM |
| IQ2_S | 3.06 | Low | 5.59 GB | 0.84 GB | 7.24 GB | 71.5 t/s | Fits in VRAM |
| IQ2_M | 3.22 | Low | 5.89 GB | 0.84 GB | 7.54 GB | 67.9 t/s | Fits in VRAM |
| Q2_K | 3.28 | Low | 5.99 GB | 0.84 GB | 7.63 GB | 66.8 t/s | Fits in VRAM |
| IQ3_XXS | 3.55 | Fair | 6.49 GB | 0.84 GB | 8.13 GB | 7.7 t/s | Offload |
| IQ3_XS | 3.63 | Fair | 6.63 GB | 0.84 GB | 8.28 GB | 7.5 t/s | Offload |
| Q3_K_S | 3.81 | Fair | 6.97 GB | 0.84 GB | 8.62 GB | 7.2 t/s | Offload |
| IQ3_M | 3.85 | Fair | 7.03 GB | 0.84 GB | 8.68 GB | 7.1 t/s | Offload |
| Q3_K_M | 4.14 | Fair | 7.57 GB | 0.84 GB | 9.21 GB | 6.6 t/s | Offload |
| Q3_K_L | 4.31 | Good | 7.88 GB | 0.84 GB | 9.52 GB | 6.3 t/s | Offload |
| IQ4_XS | 4.37 | Good | 7.98 GB | 0.84 GB | 9.63 GB | 6.3 t/s | Offload |
| Q4_K_S | 4.86 | Good | 8.88 GB | 0.84 GB | 10.52 GB | 5.6 t/s | Offload |
| Q4_K_M | 5.28 | Very good | 9.65 GB | 0.84 GB | 11.3 GB | 5.2 t/s | Offload |
| Q4_K_L | 5.56 | Very good | 10.16 GB | 0.84 GB | 11.81 GB | 4.9 t/s | Offload |
| Q5_K_S | 5.68 | Very good | 10.38 GB | 0.84 GB | 12.02 GB | 4.8 t/s | Offload |
| Q5_K_M | 6.04 | Very good | 11.04 GB | 0.84 GB | 12.68 GB | 4.5 t/s | Offload |
| Q5_K_L | 6.3 | Very good | 11.52 GB | 0.84 GB | 13.17 GB | 4.3 t/s | Offload |
| Q6_K | 7.16 | Excellent | 13.1 GB | 0.84 GB | 14.74 GB | 3.8 t/s | Offload |
| Q6_K_L | 7.42 | Excellent | 13.56 GB | 0.84 GB | 15.21 GB | 3.7 t/s | Offload |
| Q8_0 | 8.51 | Excellent | 15.56 GB | 0.84 GB | 17.2 GB | 3.2 t/s | Offload |
| Q8_0_L | 8.71 | Excellent | 15.92 GB | 0.84 GB | 17.57 GB | 3.1 t/s | Offload |
| F32 | 32.0 | Excellent | 58.51 GB | 0.84 GB | 60.16 GB | — | Insufficient |
KV cache computed from the model's exact architecture. Speed is a rough estimate bounded by memory bandwidth.
Frequently asked questions
What kind of model is bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF?
bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF is a code-focused language model with 15.71 billion parameters, based on the deepseek2 architecture. It is released under the other license and distributed as GGUF files for local inference.
How much VRAM do you need to run bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF?
You need about 7.63 GB of VRAM to run bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF entirely on the GPU using the Q2_K quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.
Can I run bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF on an 8 GB GPU?
Yes. With 8 GB of VRAM you can run bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF fully on the GPU using Q2_K (about 7.63 GB).
Can I run bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF on a 16 GB GPU?
Yes. With 16 GB of VRAM you can run bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF fully on the GPU using Q6_K_L (about 15.21 GB).
Can I run bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF on a 24 GB GPU?
Yes. With 24 GB of VRAM you can run bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF fully on the GPU using Q8_0_L (about 17.57 GB).
What context length does bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF support?
bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF supports a native context length of up to 163,840 tokens. A longer context grows the KV cache, so it increases the memory needed to run the model.
What is the best quantization for bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF?
For bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF, a strong default is Q4_K_M, which needs about 11.3 GB and keeps most of the quality while roughly halving the memory versus 8-bit. With VRAM to spare, Q5_K_M or Q6_K add a little more quality; if you are tight on memory, a smaller quantization still runs. Pick the highest quantization that fits your VRAM.