title: DeepSeek-V4-Flash-DSpark-Abliterated
emoji: 🔥
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
DeepSeek-V4-Flash-DSpark-Abliterated Demo
🔥 Abliterated (uncensored) DeepSeek-V4-Flash with 1M context window.
Model Specifications
| Metric | Value |
|---|---|
| Parameters | 284B MoE / ~13B active |
| Context | 1,048,576 tokens |
| C1 Decode | ~57 tok/s |
| Refusal Bypass | ~100% |
How to Use
- Enter your vLLM API endpoint in the settings
- Adjust temperature and max tokens as needed
- Start chatting!
Hardware Requirements
This model requires 2× NVIDIA DGX Spark (GB10) servers with vLLM and kv_cache_dtype=nvfp4_ds_mla.
Disclaimer
⚠️ This model has safety refusals removed. You are responsible for adding appropriate safety filtering, human review, and access controls for your deployment.