Instructions to use webmp3/Sakura-Fara1.5-9B-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 webmp3/Sakura-Fara1.5-9B-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 webmp3/Sakura-Fara1.5-9B-GGUF:F16
# Run inference directly in the terminal:
llama cli -hf webmp3/Sakura-Fara1.5-9B-GGUF:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf webmp3/Sakura-Fara1.5-9B-GGUF:F16
# Run inference directly in the terminal:
llama cli -hf webmp3/Sakura-Fara1.5-9B-GGUF:F16

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 webmp3/Sakura-Fara1.5-9B-GGUF:F16
# Run inference directly in the terminal:
./llama-cli -hf webmp3/Sakura-Fara1.5-9B-GGUF:F16

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 webmp3/Sakura-Fara1.5-9B-GGUF:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf webmp3/Sakura-Fara1.5-9B-GGUF:F16

Use Docker

docker model run hf.co/webmp3/Sakura-Fara1.5-9B-GGUF:F16

LM Studio
Jan

vLLM

How to use webmp3/Sakura-Fara1.5-9B-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "webmp3/Sakura-Fara1.5-9B-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": "webmp3/Sakura-Fara1.5-9B-GGUF",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/webmp3/Sakura-Fara1.5-9B-GGUF:F16

Ollama
How to use webmp3/Sakura-Fara1.5-9B-GGUF with Ollama:
```
ollama run hf.co/webmp3/Sakura-Fara1.5-9B-GGUF:F16
```
Unsloth Desktop

How to use webmp3/Sakura-Fara1.5-9B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf webmp3/Sakura-Fara1.5-9B-GGUF:F16

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "webmp3/Sakura-Fara1.5-9B-GGUF:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use webmp3/Sakura-Fara1.5-9B-GGUF with Docker Model Runner:
```
docker model run hf.co/webmp3/Sakura-Fara1.5-9B-GGUF:F16
```

Lemonade

How to use webmp3/Sakura-Fara1.5-9B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull webmp3/Sakura-Fara1.5-9B-GGUF:F16

Run and chat with the model

lemonade run user.Sakura-Fara1.5-9B-GGUF-F16

List all available models

lemonade list

Hermes Agent

How to use webmp3/Sakura-Fara1.5-9B-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 webmp3/Sakura-Fara1.5-9B-GGUF:F16

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 webmp3/Sakura-Fara1.5-9B-GGUF:F16

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use webmp3/Sakura-Fara1.5-9B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf webmp3/Sakura-Fara1.5-9B-GGUF:F16

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 "webmp3/Sakura-Fara1.5-9B-GGUF:F16" \
  --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"

Sakura — Fara1.5-9B (GGUF, three sizes)

Fara1.5-9B is Microsoft's computer-use agent for web browsing (Qwen3.5-9B base, multimodal: it reads browser screenshots and emits structured tool calls). This repository holds three GGUF files of different size made from the BF16 weights with our measured mixed-codec method. Pick one on the Files page; the table below says what each file is. Independent community quantization, not an official Microsoft release. Part of the Sakura Mini line.

Quality at a glance

Same method, same texts, same machine for every file; lower KL divergence means closer to the BF16 original:

Sakura-Fara1.5-9B-3.91GiB.gguf vs bartowski IQ3_XXS (3.98 GiB): KL divergence to the BF16 original 36 % lower on average over all 3 texts (0.07 GiB smaller).
Sakura-Fara1.5-9B-4.84GiB.gguf vs bartowski IQ4_XS (4.88 GiB): KL divergence to the BF16 original 18 % lower on average over all 3 texts (0.04 GiB smaller).
Sakura-Fara1.5-9B-5.54GiB.gguf vs bartowski Q4_K_M (5.50 GiB): KL divergence to the BF16 original 40 % lower on average over all 3 texts (0.04 GiB larger).

Why: the bit budget per tensor is allocated from measured sensitivity (see How it was made), not from fixed rules. Full table below.

Which file should I take?

File	For
`Sakura-Fara1.5-9B-3.91GiB.gguf` (3.90 GiB, 3.74 bpw)	smallest: for tight memory budgets; expect visible quality loss
`Sakura-Fara1.5-9B-4.84GiB.gguf` (4.83 GiB, 4.64 bpw)	middle: best balance of size and quality for most machines
`Sakura-Fara1.5-9B-5.54GiB.gguf` (5.53 GiB, 5.31 bpw)	largest: closest to the BF16 original of the three

The three files

File	Size	bits/weight	KLD en	KLD dev	KLD wiki	same top token (avg)	Tensor types by size
`Sakura-Fara1.5-9B-3.91GiB.gguf`	3.90 GiB	3.74	0.0611	0.0600	0.1071	89.8 %	IQ3_S 41%, IQ4_XS 27%, Q3_K 21%, Q4_K 8%, Q5_K 2%
`Sakura-Fara1.5-9B-4.84GiB.gguf`	4.83 GiB	4.64	0.0158	0.0171	0.0282	94.6 %	Q4_K 40%, IQ4_XS 33%, Q5_K 26%
`Sakura-Fara1.5-9B-5.54GiB.gguf`	5.53 GiB	5.31	0.0082	0.0085	0.0114	96.4 %	Q5_K 83%, IQ4_XS 9%, Q4_K 6%

"Tensor types by size" lists the share of the file's bytes per ggml type (all are standard llama.cpp types, so the files run in stock llama.cpp). Quality columns: mean KL divergence of the quantized model's next-token distribution against the BF16 source on held-out text (lower is better) and how often the most likely token is unchanged (higher is better).

mmproj-Fara1.5-9B-f16.gguf: vision projector (unchanged from bartowski's F16 conversion).

Comparison at similar size

Same texts, same context, same chunks, measured by us with llama-perplexity (BF16 source = reference; PPL of the BF16 model: en: 1.901, dev: 2.044, wiki: 8.878). The reference quants are bartowski's imatrix quants of the same model, measured the same way. Sorted by size.

Quant	Source	Size	KLD en	KLD dev	KLD wiki	PPL en	same top token (avg)
`Sakura-Fara1.5-9B-3.91GiB.gguf`	this repository	3.91 GiB	0.0611	0.0600	0.1071	1.963	89.8 %
IQ3_XXS	bartowski (imatrix)	3.98 GiB	0.0937	0.0981	0.1662	1.993	88.0 %
`Sakura-Fara1.5-9B-4.84GiB.gguf`	this repository	4.84 GiB	0.0158	0.0171	0.0282	1.926	94.6 %
IQ4_XS	bartowski (imatrix)	4.88 GiB	0.0194	0.0224	0.0325	1.913	94.6 %
Q4_K_M	bartowski (imatrix)	5.50 GiB	0.0129	0.0131	0.0218	1.912	95.6 %
`Sakura-Fara1.5-9B-5.54GiB.gguf`	this repository	5.54 GiB	0.0082	0.0085	0.0114	1.918	96.4 %

Method: 12 chunks of 512 tokens per text; texts: English held-out text, developer-text held-out set, English encyclopedia held-out text (not used in any calibration). Single measurement on one machine; small KLD differences at similar size are not a quality ranking.

How it was made

BF16 GGUF of the original model and the importance matrix come from bartowski/Fara1.5-9B-GGUF; we used both unchanged (credit and thanks to bartowski).
Every large weight matrix was quantized once per candidate type (Q2_K, IQ2_S, IQ3_XXS, IQ3_S, Q3_K, IQ4_XS, Q4_K, Q5_K, Q6_K, Q8_0) with llama-quantize and the importance matrix; the error of each choice was estimated per matrix (importance-weighted) and scaled by a sensitivity factor measured with real KL divergence: every group of tensors (for example the FFN down projections, the attention value projections, the first and last layers) was moved alone to a lower-bit type, and the KL divergence it caused was compared with its quantization error.
An exact budget allocation (multiple-choice knapsack over bytes) picked one type per matrix for each target size; the final model was assembled from the stored tensors without re-quantizing. We do not publish per-tensor choices. Norms, small tensors and the multi-token-prediction layers stay at high precision.

Limits

Measured only with KL divergence and perplexity on short held-out texts and not with downstream benchmarks; do not read it as a task-quality claim.
Quantization always costs some quality; the smallest file costs the most.
Fara is a computer-use agent. Microsoft recommends running it only inside a sandbox with monitoring (MagenticLite) and describes safety behaviour (stopping and asking before payments, personal data and irreversible actions). Quantization can change behaviour; we did not test agent safety or screenshot grounding, only text perplexity/KLD.
The vision projector mmproj-Fara1.5-9B-f16.gguf is bartowski's conversion (unchanged, F16); it is needed for screenshots and is not quantized by us.

Credits

Original model: Fara1.5-9B (Microsoft, MIT), MIT license (included as LICENSE).
BF16 GGUF, importance matrix and the reference quants: bartowski.
GGUF format and tools: ggml-org/llama.cpp (MIT).

中文说明 · 樱花 (Simplified Chinese)

English above. 本节为上文的中文翻译(Sakura = 樱花 yīnghuā);完整的独立中文版见 README_zh.md。

Sakura — Fara1.5-9B(GGUF,三种大小)

Fara1.5-9B 是 Microsoft 面向网页浏览的计算机使用(computer-use)智能体(基于 Qwen3.5-9B,多模态:读取浏览器截图并输出结构化的工具调用)。本仓库包含 三个不同大小的 GGUF 文件,由 BF16 权重通过我们实测得出的混合编码方法制作。请在 Files 页面选择其中一个;下表说明了每个文件的用途。这是独立的社区量化版本,不是 Microsoft 的官方发布。属于 Sakura Mini 系列。

质量概览

每个文件使用相同的方法、相同的文本、同一台机器;KL 散度越低,越接近 BF16 原始模型:

Sakura-Fara1.5-9B-3.91GiB.gguf 对比 bartowski IQ3_XXS (3.98 GiB):相对于 BF16 原始模型,在全部 3 个文本上平均 KL 散度 低 36 %(小 0.07 GiB)。
Sakura-Fara1.5-9B-4.84GiB.gguf 对比 bartowski IQ4_XS (4.88 GiB):相对于 BF16 原始模型,在全部 3 个文本上平均 KL 散度 低 18 %(小 0.04 GiB)。
Sakura-Fara1.5-9B-5.54GiB.gguf 对比 bartowski Q4_K_M (5.50 GiB):相对于 BF16 原始模型,在全部 3 个文本上平均 KL 散度 低 40 %(大 0.04 GiB)。

原因:每个张量的比特预算是根据实测的敏感度分配的(见 制作方式),而不是按固定规则。完整表格见下文。

我该选哪个文件?

文件	适用场景
`Sakura-Fara1.5-9B-3.91GiB.gguf`(3.90 GiB,3.74 bpw)	最小:适合内存预算紧张的情况;预计质量会有明显损失
`Sakura-Fara1.5-9B-4.84GiB.gguf`(4.83 GiB,4.64 bpw)	居中:对大多数机器而言,大小与质量的最佳平衡
`Sakura-Fara1.5-9B-5.54GiB.gguf`(5.53 GiB,5.31 bpw)	最大:三者中最接近 BF16 原始模型

这三个文件

文件	大小	bits/weight	KLD en	KLD dev	KLD wiki	same top token (avg)	各张量类型(按大小)
`Sakura-Fara1.5-9B-3.91GiB.gguf`	3.90 GiB	3.74	0.0611	0.0600	0.1071	89.8 %	IQ3_S 41%, IQ4_XS 27%, Q3_K 21%, Q4_K 8%, Q5_K 2%
`Sakura-Fara1.5-9B-4.84GiB.gguf`	4.83 GiB	4.64	0.0158	0.0171	0.0282	94.6 %	Q4_K 40%, IQ4_XS 33%, Q5_K 26%
`Sakura-Fara1.5-9B-5.54GiB.gguf`	5.53 GiB	5.31	0.0082	0.0085	0.0114	96.4 %	Q5_K 83%, IQ4_XS 9%, Q4_K 6%

“各张量类型(按大小)”列出每个 ggml 类型占文件字节数的比例(全部是标准的 llama.cpp 类型,因此这些文件可在原版 llama.cpp 中运行)。质量列:量化模型的下一个 token 分布相对于 BF16 源文件在留出文本上的平均 KL 散度(越低越好),以及最可能的 token 保持不变的频率(越高越好)。

mmproj-Fara1.5-9B-f16.gguf:视觉投影器(与 bartowski 的 F16 转换版本相同,未改动)。

相近大小的比较

相同的文本、相同的上下文、相同的分块,由我们使用 llama-perplexity 测量(BF16 源文件 = 参照;BF16 模型的 PPL:en: 1.901,dev: 2.044,wiki: 8.878)。参照量化为 bartowski 对同一模型的 imatrix 量化,以相同方式测量。按大小排序。

量化	来源	大小	KLD en	KLD dev	KLD wiki	PPL en	same top token (avg)
`Sakura-Fara1.5-9B-3.91GiB.gguf`	本仓库	3.91 GiB	0.0611	0.0600	0.1071	1.963	89.8 %
IQ3_XXS	bartowski (imatrix)	3.98 GiB	0.0937	0.0981	0.1662	1.993	88.0 %
`Sakura-Fara1.5-9B-4.84GiB.gguf`	本仓库	4.84 GiB	0.0158	0.0171	0.0282	1.926	94.6 %
IQ4_XS	bartowski (imatrix)	4.88 GiB	0.0194	0.0224	0.0325	1.913	94.6 %
Q4_K_M	bartowski (imatrix)	5.50 GiB	0.0129	0.0131	0.0218	1.912	95.6 %
`Sakura-Fara1.5-9B-5.54GiB.gguf`	本仓库	5.54 GiB	0.0082	0.0085	0.0114	1.918	96.4 %

方法:每个文本取 12 个块,每块 512 个 token;文本为:英文留出文本、开发者文本留出集、英文百科留出文本(未用于任何校准)。这是在一台机器上的单次测量;相近大小下 KLD 的细小差异并不构成质量排名。

制作方式

原始模型的 BF16 GGUF 和重要性矩阵来自 bartowski/Fara1.5-9B-GGUF;我们原样使用了这两者(感谢并归功于 bartowski)。
每个大型权重矩阵针对每种候选类型(Q2_K、IQ2_S、IQ3_XXS、IQ3_S、Q3_K、IQ4_XS、Q4_K、Q5_K、Q6_K、Q8_0)使用 llama-quantize 和重要性矩阵各量化一次;每种选择的误差按矩阵估计(按重要性加权),并乘以一个 用真实 KL 散度测得的敏感度系数:每组张量(例如 FFN down 投影、注意力 value 投影、首尾几层)被单独移到较低比特的类型,并将它造成的 KL 散度与其量化误差进行比较。
精确的预算分配(以字节为单位的多选背包问题)为每个目标大小为每个矩阵挑选一种类型;最终模型由已存储的张量组装而成,无需重新量化。我们不公布逐张量的选择。归一化层、小张量和多 token 预测层保持高精度。

局限

仅用短的留出文本上的 KL 散度和困惑度测量,没有使用下游基准测试;请勿将其理解为任务质量的声明。
量化总会损失一些质量;最小的文件损失最大。
Fara 是一个 计算机使用智能体。Microsoft 建议仅在带监控的沙箱(MagenticLite)中运行它,并描述了其安全行为(在涉及付款、个人数据和不可逆操作之前停下并询问)。量化可能改变行为;我们没有测试智能体的安全性或截图定位能力,只测试了文本困惑度/KLD。
视觉投影器 mmproj-Fara1.5-9B-f16.gguf 是 bartowski 的转换版本(未改动,F16);截图需要它,我们没有对它进行量化。