Instructions to use 169Pi/Alpie_Seed_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 169Pi/Alpie_Seed_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 169Pi/Alpie_Seed_GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf 169Pi/Alpie_Seed_GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf 169Pi/Alpie_Seed_GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf 169Pi/Alpie_Seed_GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf 169Pi/Alpie_Seed_GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf 169Pi/Alpie_Seed_GGUF:Q4_K_M

Use Docker

docker model run hf.co/169Pi/Alpie_Seed_GGUF:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use 169Pi/Alpie_Seed_GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "169Pi/Alpie_Seed_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": "169Pi/Alpie_Seed_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/169Pi/Alpie_Seed_GGUF:Q4_K_M

Ollama
How to use 169Pi/Alpie_Seed_GGUF with Ollama:
```
ollama run hf.co/169Pi/Alpie_Seed_GGUF:Q4_K_M
```
Unsloth Desktop

How to use 169Pi/Alpie_Seed_GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf 169Pi/Alpie_Seed_GGUF:Q4_K_M

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": "169Pi/Alpie_Seed_GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use 169Pi/Alpie_Seed_GGUF with Docker Model Runner:
```
docker model run hf.co/169Pi/Alpie_Seed_GGUF:Q4_K_M
```

Lemonade

How to use 169Pi/Alpie_Seed_GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull 169Pi/Alpie_Seed_GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Alpie_Seed_GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use 169Pi/Alpie_Seed_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 169Pi/Alpie_Seed_GGUF:Q4_K_M

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 169Pi/Alpie_Seed_GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use 169Pi/Alpie_Seed_GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf 169Pi/Alpie_Seed_GGUF:Q4_K_M

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 "169Pi/Alpie_Seed_GGUF:Q4_K_M" \
  --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"

deepanshupillm commited on 3 days ago

Commit

e5282d7

verified ·

1 Parent(s): daea3a3

Update README.md

Browse files

Files changed (1) hide show

README.md +91 -335

README.md CHANGED Viewed

@@ -1,383 +1,139 @@
 ---
 tags:
-  - alpie
-  - 169pi
-  - gguf
-  - llama-cpp
-  - local-inference
-  - mobile-inference
-  - on-device-ai
-  - quantization
-  - wikitext-2
-  - qwen3-5
 ---
-# Alpie Seed GGUF
-### Your AI. Your Device. Your Control.
-**Developed by 169Pi | 8.95B Parameters | GGUF | Local & On-Device Inference**
-What if you could run an approximately 9-billion-parameter language model directly on your smartphone?
-No cloud inference API required. No mandatory external server. Just a model running locally on compatible hardware.
-Introducing **Alpie Seed**, developed by 169Pi and released in GGUF format to make local AI inference accessible across desktops and mobile devices.
-From an iPhone 17 Pro to desktop systems, Alpie Seed demonstrates that modern language models can move beyond traditional server-based deployments.
-With multiple quantization variants, competitive WikiText-2 perplexity results, and real-world inference measurements, Alpie Seed is designed for developers, researchers, and enthusiasts exploring locally deployed AI.
----
-## Highlights
-| Feature | Result |
-|---|---|
-| Model scale | 8.95B parameters |
-| Mobile inference | iPhone 17 Pro tested |
-| iPhone generation speed | **9.59 tokens/sec** |
-| Mac M2 generation speed | ~14.1 tokens/sec |
-| Lowest reported PPL | **8.8921 (Q4)** |
-| Quantization options | Q2, Q3, Q4 |
-| Model format | GGUF |
-| Inference ecosystem | llama.cpp |
-### What makes Alpie Seed interesting?
-- **AI on smartphones:** Successfully executed and benchmarked on an iPhone 17 Pro using Apple Metal.
-- **Local inference:** Run supported model configurations directly on your device after downloading the weights.
-- **Flexible quantization:** Select Q2, Q3, or Q4 according to your available memory.
-- **Competitive language modeling:** Q4 records approximately 1.43% lower mean perplexity than the Qwen Q4 reference in our reported WikiText-2 evaluation.
-- **Developer-friendly:** GGUF distribution for integration with compatible local inference environments.
----
-# 1. Model Information
-| Attribute | Details |
-|---|---|
-| Model name | Alpie Seed |
-| Organization | 169Pi |
-| Model family | Alpie |
-| Parameter count | Approximately 8.95B |
-| Base model | Qwen3.5-9B |
-| Model architecture | Qwen3.5-based |
-| Distribution format | GGUF |
-| Quantization variants | Q2_K, Q3_K_M, Q4_K_M |
-| Primary inference engine | llama.cpp |
-| Primary evaluation dataset | WikiText-2 |
-| Primary evaluation metric | Mean perplexity |
-| Intended deployment | Desktop and compatible mobile hardware |
-# 2. Run AI Locally: Desktop to Smartphone
-One of the primary objectives of Alpie Seed is to make language-model inference accessible on consumer hardware.
-We evaluated the Q4_K_M configuration across three different hardware environments, including an iPhone 17 Pro.
-## Hardware performance
-| Device | Backend | Quantization | Prompt processing | Generation |
-|---|---|---|---:|---:|
-| iPhone 17 Pro | Apple Metal | Q4_K_M | 230.65 tok/s | **9.59 tok/s** |
-| Mac M2 (16 GB) | Apple Metal | Q4_K_M | ~130 tok/s | **~14.1 tok/s** |
-| Windows CPU | CPU | Q4_K_M | 6.40 tok/s | 2.66 tok/s |
-### iPhone 17 Pro: On-device inference
-**A model with approximately 8.95 billion parameters, running locally on a smartphone.**
-Reported benchmark configuration:
-| Parameter | Value |
-|---|---|
-| Device | iPhone 17 Pro |
-| Backend | Apple Metal |
-| Quantization | Q4_K_M |
-| Model artifact size | 5.62 GB (reported) |
-| Peak memory | 7 GB |
-| GPU layers | 99 |
-| Flash Attention | Enabled |
-| Prompt tokens | 512 |
-| Generation tokens | 128 |
-| Repetitions | 3 |
-| Prompt processing | 230.65 tokens/sec |
-| Token generation | 9.59 tokens/sec |
-| Total test time | 47 seconds |
-This demonstrates practical local inference on modern mobile hardware, using hardware acceleration rather than relying on a remote inference service.
-### Mac M2: Apple Silicon inference
-| Parameter | Value |
-|---|---|
-| Device | Mac M2, 16 GB |
-| Backend | Apple Metal |
-| Quantization | Q4_K_M |
-| Model size | Approximately 5.23 GiB |
-| Memory usage | Approximately 5.5 GB |
-| Prompt processing | Approximately 130 tokens/sec |
-| Generation | Approximately 14.1 tokens/sec |
-### Windows: CPU inference
-| Parameter | Value |
-|---|---|
-| Backend | CPU |
-| Quantization | Q4_K_M |
-| Model size | 5.24 GB |
-| Memory usage | Approximately 7 GB |
-| GPU layers | 0 |
-| Prompt processing | 6.40 tokens/sec |
-| Generation | 2.66 tokens/sec |
-| Generation tokens | 128 |
-| Repetitions | 1 |
-### Benchmark interpretation
-These results demonstrate that Alpie Seed can operate across different hardware environments, including:
-- Apple Silicon with Metal acceleration.
-- Modern iPhone hardware with Metal acceleration.
-- CPU-only Windows environments.
-The reported speeds are hardware-specific measurements and should not be interpreted as guaranteed performance on other devices.
-Prompt processing and token generation are separate measurements. They should not be directly compared as equivalent throughput metrics.
 ---
-# 3. Choose Your Quantization
-Different devices have different memory capacities. Alpie Seed provides multiple quantization options to support different deployment requirements.
-| Variant | Recommended memory tier | Primary objective |
-|---|---|---|
-| Q4_K_M | More than 8 GB RAM | Higher-precision quantized inference |
-| Q3_K_M | Around 6 GB RAM | Balanced memory usage |
-| Q2_K | Lower-memory configurations | Reduced memory requirements |
-### Which version should you choose?
-**Q4_K_M — Recommended starting point**
-Choose Q4 when your device has sufficient available memory and you want a higher-precision quantized representation.
-**Q3_K_M — Memory-conscious deployment**
-Choose Q3 when you need to reduce memory requirements while retaining a more substantial representation than Q2.
-**Q2_K — Lower-memory deployment**
-Choose Q2 when memory constraints are your primary consideration.
-These are indicative configuration recommendations, not guaranteed minimum RAM requirements. Actual requirements depend on model file size, context length, KV cache, inference backend, operating-system overhead, and available memory.
-The iPhone 17 Pro result specifically demonstrates Q4_K_M execution. The Q3 and Q2 mobile memory tiers are recommendations rather than independently verified device benchmarks.
----
-# 4. Language Modeling Evaluation: WikiText-2
-We evaluated Alpie Seed using the WikiText-2 dataset, using mean perplexity (PPL) as the primary language-modeling metric.
-Perplexity measures how well a language model predicts a sequence of tokens.
-A lower perplexity indicates better next-token predictive performance under the evaluation setup.
-## Benchmark results
-| Model | Precision | Mean PPL |
-|---|---|---:|
-| Qwen reference | BF16 | 9.0251 |
-| Alpie Seed | BF16 | **9.0065** |
-| Qwen reference | Q4 | 9.0215 |
-| Alpie Seed | Q4 | **8.8921** |
-### Q4 comparison
-| Metric | Value |
-|---|---:|
-| Qwen Q4 PPL | 9.0215 |
-| Alpie Seed Q4 PPL | 8.8921 |
-| Absolute PPL difference | 0.1294 |
-| Relative PPL reduction | **1.43%** |
-Alpie Seed Q4 records approximately 1.43% lower mean perplexity than the Qwen Q4 reference.
-### BF16 comparison
-| Metric | Value |
-|---|---:|
-| Qwen BF16 PPL | 9.0251 |
-| Alpie Seed BF16 PPL | 9.0065 |
-| Absolute PPL difference | 0.0186 |
-| Relative PPL reduction | **0.21%** |
-Alpie Seed BF16 also records slightly lower mean perplexity than the Qwen BF16 reference.
-### What do these results indicate?
-The results show competitive language-modeling performance on WikiText-2.
-Notably, the Alpie Seed Q4 configuration achieves the lowest mean perplexity among the four reported evaluation artifacts.
-This is an interesting result because quantization reduces the numerical precision used to represent model weights. However, quantization does not necessarily produce a monotonically increasing perplexity across different model artifacts.
-**Important evaluation considerations:**
-- The Alpie BF16 reference checkpoint may not be identical to the source checkpoint used to generate the Q4 GGUF.
-- Cross-model perplexity comparisons should account for tokenizer differences, preprocessing, context length, and evaluation methodology.
-- These results represent the reported WikiText-2 evaluation and do not establish universal superiority.
-- Perplexity does not independently measure reasoning, instruction following, coding, factual accuracy, or general task performance.
-The results should be interpreted as evidence of competitive language-modeling performance under the reported evaluation conditions.
----
-# 5. Getting Started
-Alpie Seed GGUF is intended to work with compatible versions of llama.cpp.
-## Step 1: Install the Hugging Face CLI
-```bash
 pip install -U huggingface_hub
 ```
-## Step 2: Download the model
-Download the repository's GGUF artifacts:
-```bash
-hf download 169Pi/Alpie_Seed_GGUF \
-    --include "*.gguf" \
-    --local-dir ./Alpie_Seed
-```
-Alternatively, select an individual quantization file from the repository.
-## Step 3: Install llama.cpp
-Follow the official installation instructions:
-https://github.com/ggml-org/llama.cpp
-## Step 4: Run Alpie Seed
-```bash
 llama-cli \
-    -m ./Alpie_Seed/YOUR_MODEL_FILE.gguf \
-    -cnv
 ```
-Replace `YOUR_MODEL_FILE.gguf` with the downloaded model filename.
-You can now interact with Alpie Seed locally through the terminal.
-## Mobile deployment
-Alpie Seed can also be used in compatible mobile inference applications built around llama.cpp or equivalent GGUF-compatible runtimes.
-For mobile devices:
-- Select a quantization appropriate for available memory.
-- Ensure that the inference application supports the model architecture.
-- Use hardware acceleration where supported.
-- Adjust context length according to device memory.
-- For multimodal configurations, ensure that the matching projector is loaded.
-Mobile application compatibility and performance depend on the specific runtime and device.
----
-# 6. Multimodal Model Components
-Where provided in the repository, multimodal configurations may include separate projector artifacts.
-For vision-language inference, the model and its corresponding multimodal projector must be compatible and loaded together.
-A standalone text GGUF should not be assumed to support image or video input.
-Refer to the available repository files and the corresponding inference runtime documentation when configuring multimodal inference.
----
-# 7. Intended Use Cases
-Alpie Seed is suitable for experimentation and development involving:
-- Local AI assistants.
-- On-device language-model inference.
-- Offline-capable AI applications.
-- Mobile AI experimentation.
-- Desktop-based local inference.
-- Quantized model evaluation.
-- Private local text-processing workflows.
-- Research into model quantization and deployment.
-Whether an application operates fully offline depends on its runtime, downloaded dependencies, and integration.
----
-# 8. Limitations and Considerations
-- Quantization can introduce numerical approximation and may affect model quality.
-- Actual memory consumption depends on context length, KV cache, backend, and runtime configuration.
-- Performance varies across processors, GPUs, memory bandwidth, and inference engines.
-- Mobile operating systems may restrict available memory.
-- Recommended Q2/Q3/Q4 memory tiers are estimates, not device-specific guarantees.
-- The reported perplexity evaluation is limited to WikiText-2 and should not be treated as a comprehensive model capability assessment.
-- Multimodal functionality requires compatible projector components and runtime support.
----
-# 9. About 169Pi
-**169Pi is developing AI systems across multiple layers of the AI technology stack.**
-Our work spans:
-- Data and knowledge engineering.
-- Language-model development and evaluation.
-- Reasoning and agentic capabilities.
-- AI orchestration and execution.
-- Application-level AI systems.
-Alpie is part of our effort to develop and deploy practical AI models and systems.
-Alpie Seed GGUF represents our focus on making language-model inference more accessible across different hardware environments, including personal computers and mobile devices.
-We believe AI should not be limited to large-scale cloud infrastructure. Local inference opens opportunities for developers to experiment, build, and deploy AI applications directly on their own hardware.
----
-# 10. Acknowledgements
-We acknowledge:
-- The Qwen team for the underlying model family.
-- The llama.cpp community for its contributions to efficient GGUF inference.
-- The Hugging Face community for providing model distribution and developer infrastructure.
----
-# 11. License and Usage
-Please refer to the repository's `LICENSE` file for the applicable usage and redistribution terms.
-The underlying base model's licensing conditions must also be respected.
----
-## Explore Alpie Seed
-**Built by 169Pi. Designed for local inference. Ready for developers to explore.**
-Your AI. Your Device. Your Control.
-**Hugging Face:** https://huggingface.co/169Pi/Alpie_Seed_GGUF
-**Inference ecosystem:** https://github.com/ggml-org/llama.cpp

 ---
+license: apache-2.0
+base_model: 169Pi/Alpie_seed
+library_name: gguf
+pipeline_tag: image-text-to-text
 tags:
+- alpie
+- 169pi
+- gguf
+- quantized
+- local-inference
+- vision-language
+- multimodal
+- reasoning
+- coding
+- llama.cpp
+- 9b
+- apache-2.0
 ---
+<div align="center">
+<h1>🌱 Alpie Seed 9B GGUF</h1>
+<p><b>Run Alpie Seed locally.</b></p>
+<p>
+<a href="https://huggingface.co/169Pi/Alpie_seed"><img src="https://img.shields.io/badge/🤗_Primary_Model-Alpie_Seed_9B-FFD21E?style=for-the-badge&labelColor=1a1a2e" alt="Primary model"></a>
+<a href="https://huggingface.co/169Pi/Alpie_Seed_GGUF"><img src="https://img.shields.io/badge/📦_GGUF-2/3/4--bit-7C3AED?style=for-the-badge&labelColor=1a1a2e" alt="GGUF"></a>
+<a href="https://ollama.com/169pi/alpie-core"><img src="https://img.shields.io/badge/🦙_Ollama-169pi/alpie--core-00C2A8?style=for-the-badge&labelColor=1a1a2e" alt="Ollama"></a>
+</p>
+<p>
+<img src="https://img.shields.io/badge/License-Apache_2.0-22C55E?style=flat-square" alt="license">
+<img src="https://img.shields.io/badge/Params-9B-6366F1?style=flat-square" alt="params">
+<img src="https://img.shields.io/badge/Quantization-2--bit_·_3--bit_·_4--bit-F59E0B?style=flat-square" alt="quantization">
+<img src="https://img.shields.io/badge/Runtime-llama.cpp-0EA5E9?style=flat-square" alt="runtime">
+<img src="https://img.shields.io/badge/Modality-Text_·_Image-EC4899?style=flat-square" alt="modality">
+</p>
+<p>
+<a href="https://github.com/169Pi"><img src="https://img.shields.io/badge/GitHub-169Pi-181717?style=flat-square&logo=github" alt="GitHub"></a>
+<a href="https://www.linkedin.com/company/169pi"><img src="https://img.shields.io/badge/LinkedIn-169pi-0A66C2?style=flat-square&logo=linkedin&logoColor=white" alt="LinkedIn"></a>
+<a href="https://www.instagram.com/169pi.ai/"><img src="https://img.shields.io/badge/Instagram-169pi.ai-E4405F?style=flat-square&logo=instagram&logoColor=white" alt="Instagram"></a>
+<a href="https://www.youtube.com/@169pi"><img src="https://img.shields.io/badge/YouTube-@169pi-FF0000?style=flat-square&logo=youtube&logoColor=white" alt="YouTube"></a>
+<a href="https://x.com/169Pi_ai"><img src="https://img.shields.io/badge/X-@169Pi__ai-000000?style=flat-square&logo=x&logoColor=white" alt="X"></a>
+</p>
+</div>
 ---
+This repository provides GGUF quantized versions of **Alpie Seed 9B**, the open-weight vision-language model developed by **169Pi AI**. Built for local deployment, these releases make Alpie Seed available in multiple quantization levels, allowing developers and researchers to explore different trade-offs between model size, memory requirements, and inference quality.
+## Alpie Seed GGUF Highlights
+- Based on [Alpie Seed 9B](https://huggingface.co/169Pi/Alpie_seed), approximately 9B parameters.
+- **2-bit, 3-bit, and 4-bit** quantization options.
+- Designed for local inference, including desktop, workstation, and mobile hardware.
+- Reasoning, coding, and multimodal capabilities inherited from the base model.
+- Apache 2.0 license.
+## Quantization Formats
+Different quantization levels provide different trade-offs. The best choice depends on available memory, hardware, context length, and the requirements of the application.
+- **2-bit**: lowest memory footprint among the provided formats.
+- **3-bit**: intermediate memory and quality trade-off.
+- **4-bit**: higher-fidelity quantized option.
+Actual file sizes and memory requirements depend on the specific GGUF variant and runtime configuration. Refer to the repository's file list for the available artifacts.
+## Capabilities
+Alpie Seed's capabilities include multi-step reasoning, mathematical problem solving, code generation, explanation and debugging, image understanding and visual reasoning, OCR, chart and document tasks, and long-context use (262K native context in the base model, subject to runtime and memory).
+Multimodal and extended-context functionality depends on the compatibility of the selected runtime and the required model assets. A GGUF language-model file alone should not be assumed to provide every vision or video capability.
+## Quickstart
+Install the Hugging Face CLI and download the repository:
+```shell
 pip install -U huggingface_hub
+hf download 169Pi/Alpie_Seed_GGUF --local-dir ./Alpie_Seed_GGUF
 ```
+Select the quantized file that matches your available memory and intended runtime. For a compatible GGUF runtime such as llama.cpp, a basic text-generation example is:
+```shell
 llama-cli \
+  -m <path-to-model.gguf> \
+  -p "Explain the difference between supervised fine-tuning and reinforcement learning."
 ```
+For multimodal inference, use a compatible runtime build and the corresponding vision components or companion files required by the model. Check the repository files and runtime documentation for the supported configuration.
+<br>
+<div align="center">
+<img src="https://img.shields.io/badge/⚡_LOCAL_INFERENCE-7C3AED?style=for-the-badge" alt="Local inference">
+</div>
+## Local Inference
+Alpie Seed has been tested in quantized offline inference environments across desktop, workstation, and mobile hardware. The following are representative hardware measurements, not universal performance guarantees. 🏆 marks the fastest result in each column.
+| Hardware | Prompt processing | Generation |
+|:---|:---:|:---:|
+| 🍎 Mac M2, 16 GB | ![~130](https://img.shields.io/badge/~130_tok%2Fs-7C3AED?style=flat-square) | ![14.1](https://img.shields.io/badge/14.1_tok%2Fs-06B6D4?style=flat-square) |
+| 🎮 NVIDIA RTX PRO 4500 | ![3,057.16](https://img.shields.io/badge/3%2C057.16_tok%2Fs-F59E0B?style=flat-square) 🏆 | ![110.66](https://img.shields.io/badge/110.66_tok%2Fs-F59E0B?style=flat-square) 🏆 |
+| 📱 iPhone 17 Pro | ![230.65](https://img.shields.io/badge/230.65_tok%2Fs-7C3AED?style=flat-square) | ![9.59](https://img.shields.io/badge/9.59_tok%2Fs-06B6D4?style=flat-square) |
+| 📱 Samsung Galaxy Z Flip8 | ![24.31](https://img.shields.io/badge/24.31_tok%2Fs-7C3AED?style=flat-square) | ![5.90](https://img.shields.io/badge/5.90_tok%2Fs-06B6D4?style=flat-square) |
+| 💻 Windows CPU-only | ![16.74](https://img.shields.io/badge/16.74_tok%2Fs-7C3AED?style=flat-square) | ![1.95](https://img.shields.io/badge/1.95_tok%2Fs-06B6D4?style=flat-square) |
+* Results were obtained under different device and runtime configurations. They should not be compared directly without accounting for differences in prompt length, generation length, context settings, memory reporting, and inference implementation.
+* For benchmark accuracy and the full model evaluation, refer to the [primary Alpie Seed model card](https://huggingface.co/169Pi/Alpie_seed).
+## Intended Use and Limitations
+The GGUF releases are intended for local experimentation and research, offline and private on-device workflows, coding and reasoning assistants, multimodal application development where supported by the runtime, and evaluation of quantization and deployment trade-offs.
+- Quantization can affect model quality and output consistency.
+- Performance varies significantly with hardware and runtime configuration.
+- Long-context capability depends on runtime support and available memory.
+- Vision and video capabilities require compatible multimodal runtime support and the necessary model assets.
+- Local deployment does not remove the need to validate outputs, especially in high-impact applications.
+## License
+Released under the Apache License 2.0, consistent with the primary Alpie Seed model.
+<div align="center">
+**Developed by 169Pi AI** 🌱 · [Primary model](https://huggingface.co/169Pi/Alpie_seed) · [GGUF repository](https://huggingface.co/169Pi/Alpie_Seed_GGUF) · [GitHub](https://github.com/169Pi) · [LinkedIn](https://www.linkedin.com/company/169pi) · [Instagram](https://www.instagram.com/169pi.ai/) · [YouTube](https://www.youtube.com/@169pi) · [X](https://x.com/169Pi_ai) · [Ollama](https://ollama.com/169pi/alpie-core)
+</div>