Instructions to use unsloth/DeepSeek-V4-Flash-0731-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use unsloth/DeepSeek-V4-Flash-0731-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./llama-cli -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./build/bin/llama-cli -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL

Use Docker

docker model run hf.co/unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL

LM Studio
Jan
Ollama
How to use unsloth/DeepSeek-V4-Flash-0731-GGUF with Ollama:
```
ollama run hf.co/unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL
```

Unsloth Studio

How to use unsloth/DeepSeek-V4-Flash-0731-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for unsloth/DeepSeek-V4-Flash-0731-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for unsloth/DeepSeek-V4-Flash-0731-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for unsloth/DeepSeek-V4-Flash-0731-GGUF to start chatting

How to use unsloth/DeepSeek-V4-Flash-0731-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use unsloth/DeepSeek-V4-Flash-0731-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use unsloth/DeepSeek-V4-Flash-0731-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Docker Model Runner
How to use unsloth/DeepSeek-V4-Flash-0731-GGUF with Docker Model Runner:
```
docker model run hf.co/unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL
```

Lemonade

How to use unsloth/DeepSeek-V4-Flash-0731-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q4_K_XL

Run and chat with the model

lemonade run user.DeepSeek-V4-Flash-0731-GGUF-UD-Q4_K_XL

List all available models

lemonade list

dspark-DeepSeek-V4-Flash-0731-Q8_0.gguf FAILS to load in the latest llama-server

#26

by Nerdsking - opened about 16 hours ago

Discussion

Nerdsking

about 16 hours ago

Using latest llama (https://github.com/ggml-org/llama.cpp/releases/tag/b10240), it gives this message:

"0.00.708.410 W srv load_model: [spec] failed to measure draft model memory: failed to create llama_context from model"

Hardware: Ryzen 3700x 128GB RAM + 20GB VRAM (modded RTX 3080) - Windows 11

Command: llama-server.exe --model "c:\AI_Models\DeepSeek-V4-Flash-0731-UD-IQ4_NL-00001-of-00004.gguf" --model-draft "c:\AI_Models\dspark-DeepSeek-V4-Flash-0731-Q8_0.gguf" --spec-type draft-dspark --spec-draft-n-max 2 --top-p 0.95 --temp 1.0 --seed 42 --threads 8 -Cr 0-7 --ctx-size 50000 --host 0.0.0.0 -np 1 --cache-type-k q8_0 --cache-type-v q8_0 --flash-attn on --parallel 1 --batch-size 512 --ubatch-size 128 --offline --verbosity 4 --load-mode none

danielhanchen

Unsloth AI org about 14 hours ago

Using latest llama (https://github.com/ggml-org/llama.cpp/releases/tag/b10240), it gives this message:

"0.00.708.410 W srv load_model: [spec] failed to measure draft model memory: failed to create llama_context from model"

Hardware: Ryzen 3700x 128GB RAM + 20GB VRAM (modded RTX 3080) - Windows 11

Command: llama-server.exe --model "c:\AI_Models\DeepSeek-V4-Flash-0731-UD-IQ4_NL-00001-of-00004.gguf" --model-draft "c:\AI_Models\dspark-DeepSeek-V4-Flash-0731-Q8_0.gguf" --spec-type draft-dspark --spec-draft-n-max 2 --top-p 0.95 --temp 1.0 --seed 42 --threads 8 -Cr 0-7 --ctx-size 50000 --host 0.0.0.0 -np 1 --cache-type-k q8_0 --cache-type-v q8_0 --flash-attn on --parallel 1 --batch-size 512 --ubatch-size 128 --offline --verbosity 4 --load-mode none

Still experimental, we're going to announce once everything is ready! Thanks for the error report btw we will investigate

LaikaFramework

about 14 hours ago

try: --flash-attn auto

Mikeczyk

about 10 hours ago

•

edited about 10 hours ago

It has been working ok for me. I was using the Q8_K_XL model but tried the IQ4_ML model and seems to be working there as well. Acceptance rate is a bit low so it actually doesn't speed up my generation but it works.
My llama-server was build from:

commit 3581ba0cf591b3f772fbb002de0f70e294bc0396 (HEAD -> master, real-upstream/master)
Author: Aman Gupta <amangupta052@gmail.com>
Date:   Sun Aug 2 23:16:31 2026 +0800

    convert: add option to create separate dspark GGUF (#26452)
    
    * convert: add option to create separate dspark GGUF
    
    * add --no-nextn
    
    * fix convert bug

root@llama:/opt/llama.cpp# ./llama-server --rpc pve.lan:8083 -m  /AI/DeepSeek-V4-Flash-0731-UD-IQ4_NL-00001-of-00004.gguf --host 0.0.0.0 --port 8081 -ngl 99 --cache-type-k q8_0 --cache-type-v q8_0 -c 131072 --threads 2 --threads-batch 4 --top-p 0.95 --temp .9 --repeat-penalty 1 --jinja -np 1 -fa on --dynatemp-range 0.2 --reasoning-budget 3072 --load-mode none --reasoning-format deepseek --reasoning-budget-message "Therefore, the final concise conclusion is:"  --ubatch-size 1024 --batch-size 2048 --cache-prompt --tensor-split 2.4,0,1.8,3,3,3,3,3 --spec-type draft-dspark  --spec-draft-n-max 1 --spec-draft-model /AI2/DeepseekV4-Flash-20260731-DSpark.gguf -devd CUDA0,CUDA6 -ctvd q8_0 -ctkd q8_0
0.00.078.299 I cmn  common_param: common_params_print_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
0.01.357.090 W srv  llama_server: -----------------
0.01.357.094 W srv  llama_server: CORS is set to allow all origins ('*') and no API key is set
0.01.357.095 W srv  llama_server: this can be a security risk (cross-origin attacks)
0.01.357.095 W srv  llama_server: more info: https://github.com/ggml-org/llama.cpp/pull/25655
0.01.357.095 W srv  llama_server: -----------------
0.01.358.576 I srv    load_model: loading model '/AI/DeepSeek-V4-Flash-0731-UD-IQ4_NL-00001-of-00004.gguf'
0.01.746.194 E llama_init_from_model: failed to initialize the context: dflash requires ctx_other to be set (this warning is normal during memory fitting)
0.01.775.476 W srv    load_model: [spec] failed to measure draft model memory: failed to create llama_context from model
13.55.481.933 I common_speculative_init_result: loading draft model '/AI2/DeepseekV4-Flash-20260731-DSpark.gguf'
14.52.491.668 I srv    load_model: initializing, n_slots = 1, n_ctx_slot = 131072, kv_unified = 'false'
14.52.491.717 I common_speculative_impl_draft_dflash: adding speculative implementation 'draft-dspark'
14.52.491.726 I common_speculative_impl_draft_dflash: - n_max=1, n_min=0, p_min=0.00
14.52.491.727 I common_speculative_impl_draft_dflash: - block_size=5, mask_token_id=128799, n_extract=3
14.52.668.641 I srv  llama_server: model loaded
14.52.668.649 I srv  llama_server: listening on http://0.0.0.0:8081
15.22.652.057 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
17.26.798.328 I slot launch_slot_: id  0 | task 415 | processing task, is_child = 0
17.33.759.155 I slot print_timing: id  0 | task 415 | prompt processing, n_tokens =    441, progress = 0.99, t =   6.96 s / 63.35 tokens per second
17.49.723.415 I slot print_timing: id  0 | task 415 | n_decoded =    100, tg =   6.44 t/s, tg_3s =   6.44 t/s
17.52.964.167 I slot print_timing: id  0 | task 415 | n_decoded =    123, tg =   6.55 t/s, tg_3s =   7.10 t/s
17.56.203.860 I slot print_timing: id  0 | task 415 | n_decoded =    145, tg =   6.59 t/s, tg_3s =   6.79 t/s
17.59.443.249 I slot print_timing: id  0 | task 415 | n_decoded =    165, tg =   6.53 t/s, tg_3s =   6.17 t/s
18.02.704.079 I slot print_timing: id  0 | task 415 | n_decoded =    188, tg =   6.59 t/s, tg_3s =   7.05 t/s
18.05.813.277 I slot print_timing: id  0 | task 415 | n_decoded =    208, tg =   6.58 t/s, tg_3s =   6.43 t/s
18.09.052.295 I slot print_timing: id  0 | task 415 | n_decoded =    229, tg =   6.57 t/s, tg_3s =   6.48 t/s
18.12.293.780 I slot print_timing: id  0 | task 415 | n_decoded =    252, tg =   6.61 t/s, tg_3s =   7.10 t/s
18.15.532.153 I slot print_timing: id  0 | task 415 | n_decoded =    273, tg =   6.60 t/s, tg_3s =   6.48 t/s
18.17.421.771 I slot print_timing: id  0 | task 415 | prompt eval time =    7392.84 ms /   445 tokens (   16.61 ms per token,    60.19 tokens per second)
18.17.421.782 I slot print_timing: id  0 | task 415 |        eval time =   43230.29 ms /   286 tokens (  151.15 ms per token,     6.62 tokens per second)
18.17.421.784 I slot print_timing: id  0 | task 415 |       total time =   50623.13 ms /   731 tokens
18.17.421.785 I slot print_timing: id  0 | task 415 |    graphs reused =        556
18.17.421.788 I slot print_timing: id  0 | task 415 | draft acceptance = 0.79874 (  127 accepted /   159 generated), mean len =  1.80
18.17.421.886 I slot      release: id  0 | task 415 | stop processing: n_tokens = 731, truncated = 0

gopi87

about 2 hours ago

Using latest llama (https://github.com/ggml-org/llama.cpp/releases/tag/b10240), it gives this message:

"0.00.708.410 W srv load_model: [spec] failed to measure draft model memory: failed to create llama_context from model"

Hardware: Ryzen 3700x 128GB RAM + 20GB VRAM (modded RTX 3080) - Windows 11

Command: llama-server.exe --model "c:\AI_Models\DeepSeek-V4-Flash-0731-UD-IQ4_NL-00001-of-00004.gguf" --model-draft "c:\AI_Models\dspark-DeepSeek-V4-Flash-0731-Q8_0.gguf" --spec-type draft-dspark --spec-draft-n-max 2 --top-p 0.95 --temp 1.0 --seed 42 --threads 8 -Cr 0-7 --ctx-size 50000 --host 0.0.0.0 -np 1 --cache-type-k q8_0 --cache-type-v q8_0 --flash-attn on --parallel 1 --batch-size 512 --ubatch-size 128 --offline --verbosity 4 --load-mode none

Still experimental, we're going to announce once everything is ready! Thanks for the error report btw we will investigate

hi sir why pp is really slow when using cpu moe even ub and b is set at 7300 is this model draw back or some works need to be done ?

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment