Cataloged from huihui-ai/Huihui-DeepSeek-V4-Flash-abliterated-ds4-GGUF
This is an uncensored version of deepseek-ai/DeepSeek-V4-Flash created with abliteration.
This quants are specific for the DS4(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
The Q2 version has a certain refusal rate. It should be fine for writing code, while the other versions are still under testing.
Choose the appropriate model based on the size of your GPU. All models can run under both Fringe210/llama.cpp-deepseek-v4-flash-cuda(supports multi-GPU) and ds4(supports multi-GPU).
ds4 now supports multi-GPU operation. For more information on how to use it, please refer to x.com/support_huihui
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
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
The Template FILE comes from antirez/deepseek-v4-gguf/DeepSeek-V4-Flash-IQ2XXS-w2Q2K-AProjQ8-SExpQ8-OutQ8-chat-v2.gguf.
| 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, norms / sinks / bias |
F32Huihui-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.
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
Use the 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
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.
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
git clone https://github.com/antirez/ds4
cd ds4
make
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."
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 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
}'
MIT. The base model copyright is held by DeepSeek; the GGUFs are redistributed under the base model's release terms.
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.
If you like it, please click 'like' and follow us for more updates.
You can follow x.com/support_huihui to get the latest model information from huihui.ai.
bc1qqnkhuchxw0zqjh2ku3lu4hq45hc6gy84uk70ge