--- license: mit library_name: transformers base_model: - deepseek-ai/DeepSeek-V4-Flash base_model_relation: quantized quantized_by: - antirez - huihui.ai tags: - abliterated - uncensored - GGUF - quantized - deepseek - deepseek-v4 - deepseek-v4-flash - moe - mixture-of-experts - 2-bit - 4-bit - iq2_xxs - q2_k - q4_k - ds4 - apple-silicon - metal - llama.cpp extra_gated_prompt: >- **Usage Warnings** “**Risk of Sensitive or Controversial Outputs**“: This model’s safety filtering has been significantly reduced, potentially generating sensitive, controversial, or inappropriate content. Users should exercise caution and rigorously review generated outputs. “**Not Suitable for All Audiences**:“ Due to limited content filtering, the model’s outputs may be inappropriate for public settings, underage users, or applications requiring high security. “**Legal and Ethical Responsibilities**“: Users must ensure their usage complies with local laws and ethical standards. Generated content may carry legal or ethical risks, and users are solely responsible for any consequences. “**Research and Experimental Use**“: It is recommended to use this model for research, testing, or controlled environments, avoiding direct use in production or public-facing commercial applications. “**Monitoring and Review Recommendations**“: Users are strongly advised to monitor model outputs in real-time and conduct manual reviews when necessary to prevent the dissemination of inappropriate content. “**No Default Safety Guarantees**“: Unlike standard models, this model has not undergone rigorous safety optimization. huihui.ai bears no responsibility for any consequences arising from its use. --- # huihui-ai/Huihui-DeepSeek-V4-Flash-abliterated-ds4-GGUF This is an uncensored version of [deepseek-ai/DeepSeek-V4-Flash](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash) created with abliteration. This quants are specific for the DS4([antirez/ds4](https://github.com/antirez/ds4)) and llama.cpp inference engine. They may work with other inference engines or not (they should, but not the MTP model which requires a specific loader). **Note** 1. The Q2 version has a certain refusal rate. It should be fine for writing code, while the other versions are still under testing. 2. Choose the appropriate model based on the size of your GPU. All models can run under both **[Fringe210/llama.cpp-deepseek-v4-flash-cuda](https://github.com/Fringe210/llama.cpp-deepseek-v4-flash-cuda)**(supports multi-GPU) and **[ds4](https://github.com/antirez/ds4)**(supports multi-GPU). 3. ds4 now supports multi-GPU operation. For more information on how to use it, please refer to [x.com/support_huihui](https://x.com/support_huihui) ## DS4 Unix Domain Socket (UDS) Acceleration Patch Dramatically accelerate multi-GPU layer-splitting inference **on the same machine** (coordinator + worker mode) by replacing TCP loopback with Unix Domain Sockets. open source 👉 [huihui-support/ds4/tree/uds](https://github.com/huihui-support/ds4/tree/uds) ## DS4 Tensor-Parallel Acceleration Patch Dramatically speed up multi-GPU layer-splitting inference on a single machine using a single process, with full support for consumer-grade graphics cards. open source 👉 [huihui-support/ds4/tree/tp](https://github.com/huihui-support/ds4/tree/tp) ## Files The Template FILE comes from [antirez/deepseek-v4-gguf/DeepSeek-V4-Flash-IQ2XXS-w2Q2K-AProjQ8-SExpQ8-OutQ8-chat-v2.gguf](https://huggingface.co/antirez/deepseek-v4-gguf/tree/main). | File | Size | Routed experts (`ffn_{gate,up,down}_exps`) | Everything else | |---|---:|---|---| | `Huihui-DeepSeek-V4-Flash-BF16-abliterated-ds4-Q2.gguf` | 80.8 GiB | `IQ2_XXS` (gate, up) + `Q2_K` (down)| `Q8_0` attn proj / shared experts / output, `F16` router + embed + indexer + compressor + HC, `F32` norms / sinks / bias | | `Huihui-DeepSeek-V4-Flash-BF16-abliterated-ds4-IQ2_XXS.gguf` | 74.7 GiB | `IQ2_XXS` (gate, up, down)|same as above | | `Huihui-DeepSeek-V4-Flash-BF16-abliterated-ds4-Q2_K.gguf` | 92.8 GiB | `Q2_K` (gate, up, down)|same as above | | `Huihui-DeepSeek-V4-Flash-BF16-abliterated-ds4-Q4_K.gguf` | 153 GiB | `Q4_K` (gate, up, down)| same as above| | `DeepSeek-V4-Flash-MTP-Q4K-Q8_0-F32.gguf` | 3.6 GiB | MTP / speculative-decoding support (optional, not standalone). | | Use **q2** on 128 GB Mac machines, **q4** on machines with ≥ 256 GB RAM, pair either with **MTP** for optional speculative decoding. ## Download ``` hf download huihui-ai/Huihui-DeepSeek-V4-Flash-abliterated-ds4-GGUF \ --local-dir ./huihui-ai/Huihui-DeepSeek-V4-Flash-abliterated-ds4-GGUF \ --token hf_xxx ``` ## llama.cpp Use the [Fringe210/llama.cpp-deepseek-v4-flash-cuda](https://github.com/Fringe210/llama.cpp-deepseek-v4-flash-cuda) program (llama-cli needs to be compiled) ``` llama-cli -m huihui-ai/Huihui-DeepSeek-V4-Flash-BF16-abliterated-ds4-Q2.gguf -n 40960 ``` ## DS4 ### Test environment Windows, WSL2, Ubuntu 24.04, RTX 6000 Pro (96GB), CUDA 13.0 In this environment, inference can reach more than 35 tokens per second. Not tested in the Apple environment. ### Supported Hardware Only the RTX 6000 Pro has been tested; other hardware has not been tested. **Metal** : MacBook with 96GB of RAM. Mac Studio class machines **NVIDIA CUDA** : DGX Spark. RTX 6000 Pro ### Install ```bash git clone https://github.com/antirez/ds4 cd ds4 make ``` ### CLI ``` export CUDA_VISIBLE_DEVICES=0 ./ds4 -m ./huihui-ai/Huihui-DeepSeek-V4-Flash-abliterated-ds4-GGUF/Huihui-DeepSeek-V4-Flash-BF16-abliterated-ds4-Q2.gguf \ -p "Explain Redis streams in one paragraph." ``` ### Server ``` export CUDA_VISIBLE_DEVICES=0 ./ds4-server \ --cuda \ -m ././huihui-ai/Huihui-DeepSeek-V4-Flash-abliterated-ds4-GGUF/Huihui-DeepSeek-V4-Flash-BF16-abliterated-ds4-Q2.gguf \ --ctx 131072 \ --kv-disk-dir ./ds4-kv-cache \ --kv-disk-space-mb 32768 \ --power 75 \ --warm-weights ``` #### curl test ``` curl http://127.0.0.1:8000/v1/models curl http://127.0.0.1:8000/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{ "model": "deepseek-v4-flash", "messages": [ {"role": "user", "content": "hello"} ], "temperature": 0.7, "max_tokens": 512, "stream": false }' ``` ## License MIT. The base model copyright is held by DeepSeek; the GGUFs are redistributed under the base model's release terms. ## Usage Warnings - **Risk of Sensitive or Controversial Outputs**: This model’s safety filtering has been significantly reduced, potentially generating sensitive, controversial, or inappropriate content. Users should exercise caution and rigorously review generated outputs. - **Not Suitable for All Audiences**: Due to limited content filtering, the model’s outputs may be inappropriate for public settings, underage users, or applications requiring high security. - **Legal and Ethical Responsibilities**: Users must ensure their usage complies with local laws and ethical standards. Generated content may carry legal or ethical risks, and users are solely responsible for any consequences. - **Research and Experimental Use**: It is recommended to use this model for research, testing, or controlled environments, avoiding direct use in production or public-facing commercial applications. - **Monitoring and Review Recommendations**: Users are strongly advised to monitor model outputs in real-time and conduct manual reviews when necessary to prevent the dissemination of inappropriate content. - **No Default Safety Guarantees**: Unlike standard models, this model has not undergone rigorous safety optimization. huihui.ai bears no responsibility for any consequences arising from its use. ## Donation If you like it, please click 'like' and follow us for more updates. You can follow [x.com/support_huihui](https://x.com/support_huihui) to get the latest model information from huihui.ai. ### Your donation helps us continue our further development and improvement, a cup of coffee can do it. - bitcoin（BTC): ``` bc1qqnkhuchxw0zqjh2ku3lu4hq45hc6gy84uk70ge ``` - Support our work on Ko-fi (https://ko-fi.com/huihuiai)!