Instructions to use christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF",
	filename="ROCmFP4-FAST/MiMo-V2.5-ROCmFP4-FAST-00001-of-00004.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use christopher-kapic/MiMo-V2.5-ROCmFP4-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 christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF
# Run inference directly in the terminal:
llama cli -hf christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF
# Run inference directly in the terminal:
llama cli -hf christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

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 christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF
# Run inference directly in the terminal:
./llama-cli -hf christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

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 christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF
# Run inference directly in the terminal:
./build/bin/llama-cli -hf christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

Use Docker

docker model run hf.co/christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

LM Studio
Jan

vLLM

How to use christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

Ollama
How to use christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF with Ollama:
```
ollama run hf.co/christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF
```

Unsloth Studio

How to use christopher-kapic/MiMo-V2.5-ROCmFP4-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 christopher-kapic/MiMo-V2.5-ROCmFP4-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 christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF to start chatting

How to use christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

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": "christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use christopher-kapic/MiMo-V2.5-ROCmFP4-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 christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

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 christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

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 "christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF" \
  --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 christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF with Docker Model Runner:
```
docker model run hf.co/christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF
```

Lemonade

How to use christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

Run and chat with the model

lemonade run user.MiMo-V2.5-ROCmFP4-GGUF-{{QUANT_TAG}}

List all available models

lemonade list

MiMo-V2.5 — ROCmFP4 GGUF (Strix Halo / gfx1151)

ROCmFP4 quantizations of MiMo-V2.5 (310B total / 15B active MoE), built for AMD Strix Halo (Ryzen AI Max+ 395, gfx1151) with the ROCmFPX fork of llama.cpp.

These are made for people running MiMo-V2.5 across two 128 GB Strix Halo boxes over a USB4/Thunderbolt link using llama.cpp's RPC layer split. The model does not fit on one node at this precision.

Note on quality: these are not imatrix-calibrated. Unsloth's UD-Q4_K_XL is, and it is tensor-aware on top of that. The ROCmFP4 variants here win on speed and size, but if output quality at the margins matters more to you than tok/s, benchmark both before switching.

Variants

variant	effective bpw	size	shards	folder
`Q4_0_ROCMFP4_FAST` ← recommended	4.26	153.47 GiB (164,795,562,720 B)	4	`ROCmFP4-FAST/`
`Q4_0_ROCMFP4`	5.24	189.02 GiB (202,969,561,824 B)	5	`ROCmFP4/`

For reference, unsloth/MiMo-V2.5-GGUF UD-Q4_K_XL is 178.44 GiB at 4.95 bpw — ROCmFP4_FAST is 14% smaller.

ROCmFP4_FAST dominates ROCmFP4 on every axis we measured — smaller, faster prefill, faster decode. Unless you specifically want the higher-precision tensors, take FAST. ROCmFP4 is published for completeness.

A note on the bpw labels. llama-quantize advertises Q4_0_ROCMFP4 as "4.50 bpw", but measured against the 309.77 B parameter count the real figure is 5.24 bpw — the recipe promotes several tensor classes (e.g. ffn_gate → q5_K) rather than quantizing everything to ROCmFP4. Q4_0_ROCMFP4_FAST measures 4.26 bpw against its 4.25 label, so that one is honest. This is why the "non-fast" build ends up larger than UD-Q4_K_XL rather than smaller.

Both were quantized from the BF16 GGUF (unsloth/MiMo-V2.5-GGUF, 14 shards, 619,638,702,336 bytes, verified byte-exact) with:

llama-quantize MiMo-V2.5-BF16-00001-of-00014.gguf \
    MiMo-V2.5-ROCmFP4-FAST.gguf Q4_0_ROCMFP4_FAST 16

Measured performance

Test setup

Two Beelink GTR 9 Pro (Ryzen AI Max+ 395, gfx1151, 128 GB unified) linked by a single USB4 cable, layer-split with llama.cpp RPC. All numbers are single-stream.


engine	ROCmFPX (llama.cpp fork), ROCm 6.4, HIP backend
OS / kernel	Ubuntu 24.04, mainline 6.18.6
topology	node2 = head, node1 = `rpc-server` over Thunderbolt (`192.168.2.1:50052`)
GPU carve-out	126976 MiB via `amdgpu.gttsize` / `ttm.pages_limit` kernel cmdline
transport	TCP over `thunderbolt0` (RDMA measured, no difference)

Benchmark command:

llama-bench -m <model.gguf> \
    -rpc 192.168.2.1:50052 -dev ROCm0/RPC0 \
    -ngl 999 -fa 1 -mmp 0 -r 1 -p 512 -n 128

-mmp 0 (no mmap) is required — with mmap the working set thrashes against the 128 GB of RAM and never converges. Note llama-bench wants -dev entries separated by /, while llama-cli wants ,.

Single-stream results

variant	size	prefill (pp512)	decode (tg128)
`Q4_0_ROCMFP4_FAST`	153.47 GiB	322.87 t/s	20.94 t/s
`Q4_0_ROCMFP4`	189.02 GiB	248.10 t/s	16.80 t/s
(reference) `UD-Q4_K_XL`	178.44 GiB	326.36 t/s	15.73 t/s

ROCmFP4_FAST is +33% decode over UD-Q4_K_XL at −1% prefill. The gain is larger than the 14% size reduction alone predicts — the single-scale layout also dequantizes more cheaply on gfx1151.

ROCmFP4 (non-fast) is −24% prefill against UD-Q4_K_XL for +7% decode, while being 10.6 GiB larger. It is the weakest of the three on throughput.

ROCmFP4_FAST was measured twice (a standalone headline run and the depth sweep's zero-depth point) as a reproducibility check: pp512 322.87 vs 315.53 (2.3% spread), tg128 20.94 vs 21.07 (0.6% spread).

Decode vs. context depth

depth	`ROCmFP4_FAST` pp / tg	`ROCmFP4` pp / tg	`UD-Q4_K_XL` pp / tg
0	315.5 / 21.07	248.5 / 17.15	324.7 / 15.69
8192	267.1 / 20.68	217.2 / 16.41	273.0 / 15.40
32768	197.7 / 19.53	167.9 / 16.07	200.8 / 14.75

Decode is remarkably flat with context on all three — a consequence of the 9-full/39-SWA attention split, where only 9 layers grow with depth. Prefill decays normally.

Aggregate throughput under concurrency

Total tokens/s across all streams, 16 distinct prompts so that slots cannot share a prefix-cache hit and inflate the result (--parallel 8, -c 32768).

variant	C4	C6	C8
`Q4_0_ROCMFP4_FAST`	31.93 t/s	32.61 t/s	33.92 t/s
`Q4_0_ROCMFP4`	24.96 t/s	26.37 t/s	28.32 t/s
(reference) `UD-Q4_K_XL`	25.81 t/s	29.19 t/s	31.72 t/s

ROCmFP4_FAST is fastest at every concurrency level, but its margin shrinks as concurrency rises — +24% over UD-Q4_K_XL at C4, +7% at C8, against +33% for a single stream. The FAST layout's advantage is in memory bandwidth and dequant cost, which dominate single-stream decode; as batch size grows the workload shifts toward expert scatter and compute, where the two are much closer.

MiMo-V2.5 is a 256-expert top-8 MoE, the least favourable case for batch amortization: the number of distinct experts touched at batch B grows as 256·(1−(1−8/256)^B), so more expert weights must be read as concurrency rises instead of being amortized across the batch. This is why aggregate throughput scales so weakly — ROCmFP4_FAST gains only 6% going from 4 streams to 8, and per-stream latency roughly halves over that range (7.98 → 4.24 t/s).

Usage

Serve across two nodes (run rpc-server on the second box first):

# node1
rpc-server -H 0.0.0.0 -p 50052

# node2
llama-server -m ROCmFP4-FAST/MiMo-V2.5-ROCmFP4-FAST-00001-of-00004.gguf \
    --rpc 192.168.2.1:50052 -dev ROCm0,RPC0 \
    -ngl 999 -fa 1 --no-mmap -c 32768 --parallel 8 --host 0.0.0.0

Point llama.cpp at the first shard; it pulls in the rest automatically.

Mind the flag spelling — llama-server takes --rpc and comma-separated -dev, while llama-bench takes -rpc and slash-separated -dev (ROCm0/RPC0). Passing the wrong one makes the server exit immediately with error: invalid argument.

Things worth knowing

Keep KV cache at f16. On gfx1151 every KV quantization we tested is slower at every context depth — the ROCm dequant overhead costs more than the bandwidth it saves — and f16 still fits a very long context here.
Speculative decoding does not pay off on this model. DFlash, MTP and EAGLE3 were all evaluated; on a 256-expert MoE, batch-verifying a draft block costs roughly 3.5 normal decode steps, so break-even needs ~3.5 accepted tokens per step. Best measured DFlash config was 10.4 t/s against a 12.8 t/s no-speculation baseline on an earlier build.
KV geometry: MiMo-V2.5 has 9 full-attention + 39 sliding-window layers, so only the 9 full layers scale with context — about 22.5 KiB/token.

License

Inherits the license of the base model, XiaomiMiMo/MiMo-V2.5. Quantization adds no additional restrictions.

Credits

XiaomiMiMo — MiMo-V2.5
unsloth — BF16 GGUF conversion used as the source
ROCmFPX — the ROCmFP4 quant formats

Downloads last month: -

GGUF

Model size

310B params

Architecture

mimo2

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Model tree for christopher-kapic/MiMo-V2.5-ROCmFP4-GGUF

Base model

XiaomiMiMo/MiMo-V2.5

Quantized

(30)

this model