Instructions to use jabbatheduck/DeepSeek-v4-flash-mini with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use jabbatheduck/DeepSeek-v4-flash-mini 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 jabbatheduck/DeepSeek-v4-flash-mini
# Run inference directly in the terminal:
llama cli -hf jabbatheduck/DeepSeek-v4-flash-mini

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf jabbatheduck/DeepSeek-v4-flash-mini
# Run inference directly in the terminal:
llama cli -hf jabbatheduck/DeepSeek-v4-flash-mini

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 jabbatheduck/DeepSeek-v4-flash-mini
# Run inference directly in the terminal:
./llama-cli -hf jabbatheduck/DeepSeek-v4-flash-mini

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 jabbatheduck/DeepSeek-v4-flash-mini
# Run inference directly in the terminal:
./build/bin/llama-cli -hf jabbatheduck/DeepSeek-v4-flash-mini

Use Docker

docker model run hf.co/jabbatheduck/DeepSeek-v4-flash-mini

LM Studio
Jan

vLLM

How to use jabbatheduck/DeepSeek-v4-flash-mini with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jabbatheduck/DeepSeek-v4-flash-mini"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jabbatheduck/DeepSeek-v4-flash-mini",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/jabbatheduck/DeepSeek-v4-flash-mini

Ollama
How to use jabbatheduck/DeepSeek-v4-flash-mini with Ollama:
```
ollama run hf.co/jabbatheduck/DeepSeek-v4-flash-mini
```

Unsloth Studio

How to use jabbatheduck/DeepSeek-v4-flash-mini 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 jabbatheduck/DeepSeek-v4-flash-mini 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 jabbatheduck/DeepSeek-v4-flash-mini to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for jabbatheduck/DeepSeek-v4-flash-mini to start chatting

How to use jabbatheduck/DeepSeek-v4-flash-mini with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf jabbatheduck/DeepSeek-v4-flash-mini

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": "jabbatheduck/DeepSeek-v4-flash-mini"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use jabbatheduck/DeepSeek-v4-flash-mini with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf jabbatheduck/DeepSeek-v4-flash-mini

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 jabbatheduck/DeepSeek-v4-flash-mini

Run Hermes

hermes

OpenClaw new

How to use jabbatheduck/DeepSeek-v4-flash-mini with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf jabbatheduck/DeepSeek-v4-flash-mini

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 "jabbatheduck/DeepSeek-v4-flash-mini" \
  --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 jabbatheduck/DeepSeek-v4-flash-mini with Docker Model Runner:
```
docker model run hf.co/jabbatheduck/DeepSeek-v4-flash-mini
```

Lemonade

How to use jabbatheduck/DeepSeek-v4-flash-mini with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull jabbatheduck/DeepSeek-v4-flash-mini

Run and chat with the model

lemonade run user.DeepSeek-v4-flash-mini-{{QUANT_TAG}}

List all available models

lemonade list

Atomic Chat

DeepSeek V4 Flash REAP K160 — GGUF

Quantized GGUF of the REAP expert-pruned DeepSeek-V4-Flash-0731 checkpoint. This is an aggressively compressed variant intended for memory-constrained inference on high-end consumer and workstation GPUs.

Source model: 0xSero/DeepSeek-V4-Flash-0731-REAP
Quantization inspiration: antirez/deepseek-v4-gguf

File

File	Size	Routed experts	Everything else
`DeepSeek-V4-Flash-REAP-IQ2XXS-w2Q2K-AProjQ8-OutQ8-chat-v2.gguf`	54 GiB	`IQ2_XXS` (gate, up) + `Q2_K` (down) — 160 of 256 retained	`Q8_0` attention / shared experts / output head, `F16` router + embed + indexer + compressor + HC, `F32` norms / sinks / bias

Quantization recipe

Tensor class	Quant	Notes
`blk..ffn_gate_exps`, `blk..ffn_up_exps`	`IQ2_XXS`	routed-expert up/gate
`blk.*.ffn_down_exps`	`Q2_K`	routed-expert down (K-quant for stability)
`blk.*.ffn_{gate,up,down}_shexp`	`Q8_0`	shared experts
`blk.*.attn_q_a`, `attn_q_b`, `attn_kv`, `attn_output_a`, `attn_output_b`	`Q8_0`	all attention projections
`output.weight`	`Q8_0`	output head
`token_embd.weight`	`F16`	input embedding
`blk.*.ffn_gate_inp`	`F16`	learned router
`blk..exp_probs_b`, `blk..attn_sinks`, all `*_norm.weight`	`F32`
`blk.*.ffn_gate_tid2eid`	`I32`	hash-routing tables (first 3 layers)
`blk..attn_compressor_`, `blk..indexer_`, `blk..hc_`, `blk..output_hc_`	`F16` / `F32`	DeepSeek-V4 auxiliary blocks

Why the asymmetry: routed experts dominate parameter count but fire on only a fraction of tokens. Aggressive quantization on them costs less in average quality than treating router, projections, or shared experts the same way. Keeping decision-making tensors at Q8_0 preserves behavior; crushing the experts buys the size.

Base-model context

This GGUF was derived from a K160 REAP checkpoint — 160 of 256 routed experts retained per MoE scope, top-6 routing preserved, router and indexer remapped to the retained expert IDs. Structural validation passed on the source checkpoint with 46/46 MoE scopes verified.

Requirements

Use a DeepSeek-V4-capable runtime. Generic transformers pipelines will not handle the custom MXFP4 / sparse-attention / tokenizer components.

Tested with the ds4 inference engine: https://github.com/antirez/ds4

License

MIT. The base model copyright is held by DeepSeek; this GGUF is redistributed under the base model's release terms.

Downloads last month: -

GGUF

Model size

180B params

Architecture

deepseek4

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Model tree for jabbatheduck/DeepSeek-v4-flash-mini

Base model

deepseek-ai/DeepSeek-V4-Flash-0731

Quantized

0xSero/DeepSeek-V4-Flash-0731-REAP

Quantized

(2)

this model