This model requires you to accept the DeepSeek license terms before download.
Description
DeepSeek-Coder-V2-Lite-Instruct — 15.7B-A2.4B code specialist; MLA attention (kv_lora 512, qk 128+64, v 128) + YaRN-mscale, MoE 64+2 experts top-6
Tags
At a glance
| Kind | code language model |
|---|---|
| Precision | bfloat16 |
| Container size | 29.9 GiB |
| Weights inside | 29.3 GiB |
| Components | 2 |
| Licence | DeepSeek |
What is inside the container
A .nbx container packs every part of DeepSeek-Coder-V2-Lite-Instruct — 2 components with weights, graphs and runtime settings — so one file is all the runtime needs.
| Component | Blocks | Layers | Weights |
|---|---|---|---|
| model | 0 | 27 | 28.9 GiB |
| lm_head | 0 | — | 400 MiB |
Default settings
The values the container ships with. They apply when you do not pass your own.
- Max length
- 163840
- Temperature
- 0.3
- Top-p
- 0.95
- Top-k
- 50
Memory needed
The weights of DeepSeek-Coder-V2-Lite-Instruct take 29.3 GiB in bfloat16. That is the floor for the memory the runtime has to place, spread over the devices it finds; activations come on top and grow with the size and length you ask for. A measured peak per device class is not published for this model yet.
Verify what you downloaded
Every download carries its SHA-256, and the file endpoint sends it in the X-NBX-SHA256 header. The container is 29.9 GiB (32,080,456,081 bytes).
$ sha256sum DeepSeek-Coder-V2-Lite-Instruct.nbx
f8a976fe26b8dbc1c70dc276a9d0840cb9f3c49a4bd5aecbe14f77714daa1666
Questions
What is DeepSeek-Coder-V2-Lite-Instruct?
A code language model by deepseek-ai, distributed as a bfloat16 .nbx container and licensed DeepSeek.
How big is the download?
29.9 GiB, of which 29.3 GiB are weights.
How do I run it?
Install the runtime from the download page, then use the commands under “CLI Usage” below: neurobrix import deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct, then neurobrix serve --model DeepSeek-Coder-V2-Lite-Instruct.
CLI Usage
$ neurobrix import deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct
$ neurobrix serve --model DeepSeek-Coder-V2-Lite-Instruct
$ neurobrix run --prompt "your prompt here"
$ neurobrix stop