ttupper75035's picture
Add self-describing model card (mtp-sidecar)
431c682 verified
|
Raw
History Blame Contribute Delete
1.98 kB
---
base_model: deepseek-ai/DeepSeek-V3.2-Exp-Base
tags:
- mtp-sidecar
- skulk
- foxlight
license: mit
foxlight:
artifact_type: mtp-sidecar
source_repo: deepseek-ai/DeepSeek-V3.2-Exp-Base
source_revision: b21ef155036dfbbd6d49013a345d31d5f5e38c89
target_model: deepseek-ai/DeepSeek-V3.2-Exp-Base
extracted_with: skulk-weights-publisher 0.1.0
generated_at: '2026-06-04T04:25:45Z'
---
# FoxlightAI/deepseek-v3-2-exp-base-mtp
This repository contains an **MTP speculative-decoding sidecar** derived from [`deepseek-ai/DeepSeek-V3.2-Exp-Base`](https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp-Base). It is **not a standalone model** — it provides the multi-token-prediction heads used by [Skulk](https://github.com/Foxlight-Foundation/Skulk) to speculatively decode for the target model **`deepseek-ai/DeepSeek-V3.2-Exp-Base`**. The heads ship at full precision (**bf16, unquantized**) — they are the speculative drafter, where precision drives draft acceptance — so there is **one sidecar per base model** and it serves **every quantization** of the target.
## Provenance
| Field | Value |
| --- | --- |
| Artifact type | mtp-sidecar |
| Source model | `deepseek-ai/DeepSeek-V3.2-Exp-Base` |
| Source revision | `b21ef155036dfbbd6d49013a345d31d5f5e38c89` |
| Target model | `deepseek-ai/DeepSeek-V3.2-Exp-Base` |
| Extracted with | skulk-weights-publisher `0.1.0` |
| Generated | 2026-06-04T04:25:45Z |
## Usage
Skulk loads this sidecar (`mtp.safetensors`) alongside the target model to enable MTP speculative decoding. It is referenced from the Skulk Weights Publisher catalog and fetched automatically by the Skulk shard downloader; it is not intended to be loaded standalone.
## License
This artifact is derived from [`deepseek-ai/DeepSeek-V3.2-Exp-Base`](https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp-Base) and is published under that model's original license (**mit**), preserved unchanged. Refer to the source model's card for the full terms.