// DeepSeek · Text LLM
DeepSeek-Coder-V2-Instruct: VRAM & local deployment
236B total · 21B active (MoE) parameters. About 159 GB at 4-bit and 520 GB at BF16 with one 8K-token request.
released 2024-06-14licence: Commercial with conditionsmemory figures are estimates
01Specs
| Publisher | DeepSeek · deepseek-ai |
|---|---|
| Family / series | DeepSeek · DeepSeek-Coder |
| Type | Text LLM (text-generation) |
| Parameters | 236B total · 21B active (MoE) Source: official model card“DeepSeek-Coder-V2 with 16B and 236B parameters … actived parameters of only 2.4B and 21B” |
| Context length | 160K tokens (163,840)max_position_embeddings in config.json |
| Architecture | 60 layers · MLA · MoE |
| Released weights | BF16 · 471 GBdownload |
| Licence | deepseek-license · Commercial with conditionsCommercial use allowed under the publisher’s own licence terms (user caps, acceptable-use policy, attribution or region limits). Read it before shipping. |
| Released | 2024-06-14 (Hugging Face repo created) |
02How much VRAM it needs (estimate)
Total = weights + 10% + 0.5GB overhead + KV cache for one request (FP16 KV). More concurrent users multiply the KV column.
| Precision | Weights | Total @ 8K | Total @ 32K | Total @ 128K | Smallest GPU | Mac |
|---|---|---|---|---|---|---|
| BF16 / FP16 (original) | 472 GB | 520 GB | 522 GB | 529 GB | 8 × H100 (640GB) | — |
| FP8 / INT8 / Q8_0 | 251 GB | 277 GB | 279 GB | 285 GB | 4 × RTX PRO 6000 (384GB) | Mac 512GB |
| 4-bit (Q4_K_M / AWQ / GPTQ-Int4) | 143 GB | 158 GB | 160 GB | 167 GB | 4 × RTX PRO 6000 (384GB) | Mac 256GB |
KV cache per request: 8K 0.6 GB · 32K 2.3 GB · 128K 9.1 GB
Bits per weight: BF16 16, FP8 8.5, 4-bit 4.85. “Fits” = total ≤ 90% of memory; Mac GPU can use ~2/3 of unified memory (≤36GB) or ~3/4 (larger). For MoE, all experts must be in memory, so size by total parameters. Method →
03How to run it
Only runtimes the publisher or the runtime’s official library names for this model.
- vLLM — mentioned in the official model card
- Transformers — mentioned in the official model card
- Ollama — in the official Ollama library · deepseek-coder-v2
ollama run deepseek-coder-v204Official repositories & quantised builds
2 official repo(s) from deepseek-ai.
- DeepSeek-Coder-V2-Instruct Instruct
- DeepSeek-Coder-V2-Base Base
05Other DeepSeek models
All 34 →| Model | Params | 4-bit total | Released |
|---|---|---|---|
| DeepSeek-Coder-V2-Instruct-0724 · same series | 236B / 21B | 159 GB | 2024-09-05 |
| DeepSeek-V4.1-Flash-1017 | 748B / 16B | 499 GB | 2026-10-17 |
| DeepSeek-V4.1-Flash | 748B / 16B | 499 GB | 2026-09-10 |
| DeepSeek-V4-Pro-0813 | 1.65T | 1.10 TB | 2026-08-13 |
| DeepSeek-V4-Flash-0731 | 304.2B | 203 GB | 2026-07-31 |
| DeepSeek-V4-Flash | 284B / 13B | 190 GB | 2026-04-22 |
| DeepSeek-V4-Pro | 1.6T / 49B | 1.07 TB | 2026-04-22 |
| DeepSeek-OCR-2 | 3.39B | 3.3 GB | 2026-01-27 |
| DeepSeek-V3.2 | 685.4B | 458 GB | 2025-12-01 |
06Sources
- Hugging Face · deepseek-ai/DeepSeek-Coder-V2-Instruct
- config.json
- License
- Ollama library · deepseek-coder-v2
Data checked 2024-08-21. Parameters, context and licence are copied from the official repository; memory figures are our estimates.