Text Generation
GGUF
laguna
Mixture of Experts
uncensored
abliterix
apex
quantization
poolside
imatrix
conversational
Instructions to use SC117/Laguna-S-2.1-Uncensored-APEX-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- llama-cpp-python
How to use SC117/Laguna-S-2.1-Uncensored-APEX-GGUF with llama-cpp-python:
# !pip install llama-cpp-python from llama_cpp import Llama llm = Llama.from_pretrained( repo_id="SC117/Laguna-S-2.1-Uncensored-APEX-GGUF", filename="Laguna-S-2.1-Uncensored-APEX-I-Balanced.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 SC117/Laguna-S-2.1-Uncensored-APEX-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 SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16 # Run inference directly in the terminal: llama cli -hf SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16 # Run inference directly in the terminal: llama cli -hf SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
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 SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16 # Run inference directly in the terminal: ./llama-cli -hf SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
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 SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16 # Run inference directly in the terminal: ./build/bin/llama-cli -hf SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
Use Docker
docker model run hf.co/SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
- LM Studio
- Jan
- vLLM
How to use SC117/Laguna-S-2.1-Uncensored-APEX-GGUF with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "SC117/Laguna-S-2.1-Uncensored-APEX-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": "SC117/Laguna-S-2.1-Uncensored-APEX-GGUF", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
- Ollama
How to use SC117/Laguna-S-2.1-Uncensored-APEX-GGUF with Ollama:
ollama run hf.co/SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
- Unsloth Studio
How to use SC117/Laguna-S-2.1-Uncensored-APEX-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 SC117/Laguna-S-2.1-Uncensored-APEX-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 SC117/Laguna-S-2.1-Uncensored-APEX-GGUF to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for SC117/Laguna-S-2.1-Uncensored-APEX-GGUF to start chatting
- Pi
How to use SC117/Laguna-S-2.1-Uncensored-APEX-GGUF with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
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": "SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Hermes Agent new
How to use SC117/Laguna-S-2.1-Uncensored-APEX-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 SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
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 SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
Run Hermes
hermes
- Atomic Chat new
- OpenClaw new
How to use SC117/Laguna-S-2.1-Uncensored-APEX-GGUF with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
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 "SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16" \ --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 SC117/Laguna-S-2.1-Uncensored-APEX-GGUF with Docker Model Runner:
docker model run hf.co/SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
- Lemonade
How to use SC117/Laguna-S-2.1-Uncensored-APEX-GGUF with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull SC117/Laguna-S-2.1-Uncensored-APEX-GGUF:BF16
Run and chat with the model
lemonade run user.Laguna-S-2.1-Uncensored-APEX-GGUF-BF16
List all available models
lemonade list
Upload README_zh.md
Browse files- README_zh.md +190 -0
README_zh.md
ADDED
|
@@ -0,0 +1,190 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: gguf
|
| 3 |
+
license: other
|
| 4 |
+
license_name: openmdw-1.1
|
| 5 |
+
license_link: https://openmdw.ai/
|
| 6 |
+
pipeline_tag: text-generation
|
| 7 |
+
tags:
|
| 8 |
+
- laguna
|
| 9 |
+
- moe
|
| 10 |
+
- uncensored
|
| 11 |
+
- abliterix
|
| 12 |
+
- apex
|
| 13 |
+
- quantization
|
| 14 |
+
- gguf
|
| 15 |
+
- poolside
|
| 16 |
+
base_model:
|
| 17 |
+
- poolside/Laguna-S-2.1
|
| 18 |
+
base_model_relation: quantized
|
| 19 |
+
---
|
| 20 |
+
|
| 21 |
+
<div style="font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif; margin-bottom: 24px;">
|
| 22 |
+
<div style="background: #f5f5f7; border-radius: 20px; padding: 36px 32px; margin-bottom: 20px; text-align: center; position: relative; overflow: hidden;">
|
| 23 |
+
<div style="position: absolute; top: -30px; right: -30px; width: 120px; height: 120px; background: #ddd6fe; border-radius: 50%;"></div>
|
| 24 |
+
<div style="position: absolute; bottom: -20px; left: 40px; width: 80px; height: 80px; background: #c4b5fd; border-radius: 50%;"></div>
|
| 25 |
+
<div style="position: absolute; top: 50%; left: -15px; width: 60px; height: 60px; background: #ddd6fe; border-radius: 50%;"></div>
|
| 26 |
+
<div style="display: inline-flex; gap: 8px; margin-bottom: 16px; position: relative; z-index: 1; flex-wrap: wrap; justify-content: center;">
|
| 27 |
+
<span style="background: #6366f1; color: white; font-size: 11px; font-weight: 600; padding: 5px 14px; border-radius: 20px;">ABLITERIX</span>
|
| 28 |
+
<span style="background: #8b5cf6; color: white; font-size: 11px; font-weight: 600; padding: 5px 14px; border-radius: 20px;">TRIAL 16</span>
|
| 29 |
+
<span style="background: #0ea5e9; color: white; font-size: 11px; font-weight: 600; padding: 5px 14px; border-radius: 20px;">APEX</span>
|
| 30 |
+
<span style="background: #64748b; color: white; font-size: 11px; font-weight: 600; padding: 5px 14px; border-radius: 20px;">OpenMDW-1.1</span>
|
| 31 |
+
</div>
|
| 32 |
+
<h1 style="margin: 0 0 8px 0; font-size: 28px; font-weight: 700; color: #1e1b4b; letter-spacing: -0.5px; border: none; position: relative; z-index: 1;">Laguna-S-2.1-Uncensored-APEX-GGUF</h1>
|
| 33 |
+
<p style="margin: 8px 0 0 0; font-size: 14px; position: relative; z-index: 1;"><a href="README.md" style="color: #6366f1; text-decoration: none;">English</a> | <span style="color: #6b7280;">📖 中文文档</span></p>
|
| 34 |
+
<p style="margin: 8px 0 0 0; font-size: 15px; color: #6b7280; position: relative; z-index: 1;">Uncensored 118B-A8B MoE · abliterix Trial 16 · APEX I 档 GGUF + imatrix</p>
|
| 35 |
+
</div>
|
| 36 |
+
</div>
|
| 37 |
+
|
| 38 |
+
<div style="font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif; display: flex; flex-direction: column; gap: 20px; margin-bottom: 30px;">
|
| 39 |
+
|
| 40 |
+
<div style="border: 1px solid #cbd5e1; border-radius: 12px; overflow: hidden; background: #ffffff; box-shadow: 0 2px 4px rgba(0,0,0,0.02);">
|
| 41 |
+
<div style="background: linear-gradient(135deg, #6366f1 0%, #8b5cf6 100%); padding: 12px 16px; color: white; font-weight: 700; font-size: 14px; display: flex; align-items: center; gap: 8px;"><span>🌊</span> 关于此版本</div>
|
| 42 |
+
<div style="padding: 16px; font-size: 13px; color: #334155; line-height: 1.7;">
|
| 43 |
+
<p style="margin: 0 0 12px 0;"><a href="https://huggingface.co/poolside/Laguna-S-2.1" target="_blank" style="color: #4f46e5; text-decoration: none; font-weight: 700;">Laguna S 2.1</a> 是 poolside 发布的 <b>约 118B 总参数、每 token 激活约 8B</b> 的 Mixture-of-Experts 模型,面向 agentic coding 与长程任务。架构要点:token-choice 路由与 softplus 门控、<b>256 个路由专家 + 1 个共享专家(top-10)</b>、GQA、<b>1:3 全局/滑窗注意力</b>(48 层:12 全局 + 36 滑窗,窗口 512)、约 <b>100 万 token 上下文</b>,并支持按请求开关的原生思考(thinking)与工具调用间交错推理。</p>
|
| 44 |
+
<p style="margin: 0 0 12px 0;">本发布版在官方权重之上做了两件事:</p>
|
| 45 |
+
<ol style="margin: 0 0 12px 0; padding-left: 20px;">
|
| 46 |
+
<li style="margin-bottom: 6px;"><b>Uncensored 行为调整</b>:使用 <a href="https://github.com/wuwangzhang1216/abliterix" target="_blank" style="color: #4f46e5; text-decoration: none; font-weight: 700;">abliterix</a>(ROCm + bitsandbytes 4-bit 搜索路径),选定 <b>Trial 16</b> LoRA,stream-merge 回 BF16,并对 MoE 安全相关专家做 router 侧 bake-in。</li>
|
| 47 |
+
<li style="margin-bottom: 6px;"><b>APEX 混合精度 GGUF</b>:在 poolside <a href="https://github.com/poolsideai/llama.cpp/tree/laguna" target="_blank" style="color: #4f46e5; text-decoration: none; font-weight: 700;">llama.cpp(laguna 分支)</a> 的 BF16 GGUF 上,用 <a href="https://github.com/mudler/apex-quant" target="_blank" style="color: #4f46e5; text-decoration: none; font-weight: 700;">APEX</a> 式张量类型配置 + <b>imatrix</b>(token embedding / output 保持 <b>BF16</b>)。</li>
|
| 48 |
+
</ol>
|
| 49 |
+
<p style="margin: 0;"><b>许可证:OpenMDW-1.1</b>(与基座同系)。详见 <a href="https://openmdw.ai/" target="_blank" style="color: #4f46e5; text-decoration: none; font-weight: 700;">openmdw.ai</a> 与 poolside 条款。</p>
|
| 50 |
+
</div>
|
| 51 |
+
</div>
|
| 52 |
+
|
| 53 |
+
<div style="border: 1px solid #fdba74; border-radius: 12px; overflow: hidden; background: #fff7ed; box-shadow: 0 2px 4px rgba(0,0,0,0.02);">
|
| 54 |
+
<div style="background: linear-gradient(135deg, #f97316 0%, #ea580c 100%); padding: 12px 16px; color: white; font-weight: 700; font-size: 14px; display: flex; align-items: center; gap: 8px;"><span>⚠️</span> Uncensored 说明</div>
|
| 55 |
+
<div style="padding: 16px; font-size: 13px; color: #334155; line-height: 1.7;">
|
| 56 |
+
<p style="margin: 0 0 12px 0;">合并 abliterix <b>Trial 16</b> 后,本模型<strong>拒答倾向明显降低</strong>,输出行为可能与官方 Laguna-S-2.1 有实质差异。请在目标场景下自行评估合规与安全策略,并做好访问控制与审计。</p>
|
| 57 |
+
<table style="width: 100%; border-collapse: collapse; font-size: 13px;"><tbody>
|
| 58 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">拒答(harmful eval)</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;"><b>8 / 100</b>(基线约 97 / 100)</td></tr>
|
| 59 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">KL divergence</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;">约 0.0042</td></tr>
|
| 60 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">Length deviation</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;">约 0.07 σ</td></tr>
|
| 61 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">Generation health</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;">PASSED</td></tr>
|
| 62 |
+
<tr><td style="padding: 6px 10px; font-weight: bold; color: #334155; background: white;">选用 Trial</td><td style="padding: 6px 10px; color: #4b5563; background: white;">abliterix Trial 16</td></tr>
|
| 63 |
+
</tbody></table>
|
| 64 |
+
<p style="margin: 12px 0 0 0; font-size: 12px; color: #9a3412;">实现摘要:LoRA 合并 <code>W += (B @ A) * (alpha / r)</code>(本 trial alpha = r = 1);每层 top 约 20 个安全专家,router 行缩放约 0.74(见合并元数据)。</p>
|
| 65 |
+
</div>
|
| 66 |
+
</div>
|
| 67 |
+
|
| 68 |
+
<div style="border: 1px solid #c4b5fd; border-radius: 12px; overflow: hidden; background: #f5f3ff; box-shadow: 0 2px 4px rgba(0,0,0,0.02);">
|
| 69 |
+
<div style="background: linear-gradient(135deg, #7c3aed 0%, #6d28d9 100%); padding: 12px 16px; color: white; font-weight: 700; font-size: 14px; display: flex; align-items: center; gap: 8px;"><span>📜</span> 声明(身份认知与文本质量)</div>
|
| 70 |
+
<div style="padding: 16px; font-size: 13px; color: #334155; line-height: 1.7;">
|
| 71 |
+
<p style="margin: 0 0 12px 0;">以下现象在<strong>官方原版 Laguna-S-2.1</strong>(含在线服务等高精度部署)上亦可能出现,<strong>并非由本仓库的 Uncensored 调整所引入</strong>,请勿归因于 abliterix / 去拒答合并:</p>
|
| 72 |
+
<ul style="margin: 0 0 12px 0; padding-left: 20px;">
|
| 73 |
+
<li style="margin-bottom: 6px;"><b>模型身份认知</b>异常或与官方人设不一致的自我描述;</li>
|
| 74 |
+
<li style="margin-bottom: 6px;">中文(及其他语言)中的<strong>异常字词、错字、生硬或偶发不通顺</strong>等文本质量问题。</li>
|
| 75 |
+
</ul>
|
| 76 |
+
<p style="margin: 0;">本仓库的 Uncensored 流程主要改变<strong>拒答与安全相关行为</strong>。量化档位(尤其更激进的 Compact 等)可能<strong>放大</strong>原有文本噪声,但根源问题在原模型与/或量化推理链路中已存在。若需对照,建议在同等推理设置下比较官方原版与本 Uncensored 版本。</p>
|
| 77 |
+
</div>
|
| 78 |
+
</div>
|
| 79 |
+
|
| 80 |
+
<div style="border: 1px solid #cbd5e1; border-radius: 12px; overflow: hidden; background: #ffffff; box-shadow: 0 2px 4px rgba(0,0,0,0.02);">
|
| 81 |
+
<div style="background: linear-gradient(135deg, #6366f1 0%, #8b5cf6 100%); padding: 12px 16px; color: white; font-weight: 700; font-size: 14px; display: flex; align-items: center; gap: 8px;"><span>🧠</span> 模型规格</div>
|
| 82 |
+
<div style="padding: 16px;">
|
| 83 |
+
<table style="width: 100%; border-collapse: collapse; font-size: 13px;"><tbody>
|
| 84 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">架构</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;">Laguna MoE(poolside <code>laguna</code>)</td></tr>
|
| 85 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">参数量</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;">约 118B 总计,每 token 激活约 8B</td></tr>
|
| 86 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">层数</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;">48(L0 为 dense FFN,L1–L47 为 MoE)</td></tr>
|
| 87 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">专家</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;">256 路由 + 1 共享,top-10</td></tr>
|
| 88 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">注意力</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;">GQA,8 KV heads,head dim 128</td></tr>
|
| 89 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">上下文</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;">至约 1,048,576 tokens(实际受显存与 <code>-c</code> 限制)</td></tr>
|
| 90 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">词表</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;">100,352(Laguna 族 tokenizer)</td></tr>
|
| 91 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">模态</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white;">文本 → 文本(本包无 mmproj)</td></tr>
|
| 92 |
+
<tr><td style="padding: 6px 10px; font-weight: bold; color: #334155; background: white;">本仓库</td><td style="padding: 6px 10px; color: #4b5563; background: white;">APEX I-Quality / I-Balanced / I-Compact GGUF</td></tr>
|
| 93 |
+
</tbody></table>
|
| 94 |
+
<p style="margin: 12px 0 0 0; font-size: 12px; color: #64748b;">编程与工具能力大体保留;拒答与对齐行为已改变。本衍生版未重新跑完整官方 bench 表。</p>
|
| 95 |
+
</div>
|
| 96 |
+
</div>
|
| 97 |
+
|
| 98 |
+
<div style="border: 1px solid #cbd5e1; border-radius: 12px; overflow: hidden; background: #ffffff; box-shadow: 0 2px 4px rgba(0,0,0,0.02);">
|
| 99 |
+
<div style="background: linear-gradient(135deg, #6366f1 0%, #8b5cf6 100%); padding: 12px 16px; color: white; font-weight: 700; font-size: 14px; display: flex; align-items: center; gap: 8px;"><span>💡</span> 什么是 APEX?</div>
|
| 100 |
+
<div style="padding: 16px; font-size: 13px; color: #334155; line-height: 1.7;">
|
| 101 |
+
<p style="margin: 0 0 12px 0;">本仓库 GGUF 采用 <a href="https://github.com/mudler/apex-quant" target="_blank" style="color: #4f46e5; text-decoration: none; font-weight: 700;">APEX</a> 式 MoE 感知混合精度:按<strong>张量角色 + 层位置</strong>分配精度(边缘更高、中间更激进),并配合 <b>imatrix</b> 做 <code>I-</code> 档。</p>
|
| 102 |
+
<p style="margin: 0;">共性设置:源为 <code>Laguna-S-2.1-Uncensored</code> BF16 GGUF;校准 <code>laguna-s-2.1.imatrix</code>;<b>token embedding / output = BF16</b>;router、norm 等小张量多为高精度默认。配置针对 Laguna 48 层命名(<code>ffn_*_exps</code> / <code>ffn_*_shexp</code> / <code>attn_*</code>,L0 dense)。</p>
|
| 103 |
+
</div>
|
| 104 |
+
</div>
|
| 105 |
+
|
| 106 |
+
<div style="border: 1px solid #cbd5e1; border-radius: 12px; overflow: hidden; background: #ffffff; box-shadow: 0 2px 4px rgba(0,0,0,0.02);">
|
| 107 |
+
<div style="background: linear-gradient(135deg, #6366f1 0%, #8b5cf6 100%); padding: 12px 16px; color: white; font-weight: 700; font-size: 14px; display: flex; align-items: center; gap: 8px;"><span>📦</span> APEX 量化档位</div>
|
| 108 |
+
<div style="padding: 16px;">
|
| 109 |
+
<table style="width: 100%; border-collapse: collapse; font-size: 13px;"><thead><tr style="background: rgba(99,102,241,0.06);">
|
| 110 |
+
<th style="padding: 8px 10px; border-bottom: 2px solid #6366f1; text-align: left; color: #4f46e5;">文件</th>
|
| 111 |
+
<th style="padding: 8px 10px; border-bottom: 2px solid #6366f1; text-align: left; color: #4f46e5;">体积</th>
|
| 112 |
+
<th style="padding: 8px 10px; border-bottom: 2px solid #6366f1; text-align: left; color: #4f46e5;">中间 expert</th>
|
| 113 |
+
<th style="padding: 8px 10px; border-bottom: 2px solid #6366f1; text-align: left; color: #4f46e5;">适用</th>
|
| 114 |
+
</tr></thead><tbody>
|
| 115 |
+
<tr><td style="padding: 8px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); background: white;"><code>*-I-Quality.gguf</code></td><td style="padding: 8px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); background: white;">约 70 GB</td><td style="padding: 8px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); background: white;">边缘 Q6_K / 近缘 Q5_K / 中间 <b>iq4_xs</b>;shared Q8_0;attn Q6_K</td><td style="padding: 8px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); background: white;">更抠体积时的高质量尝试(中间层为 IQ)</td></tr>
|
| 116 |
+
<tr><td style="padding: 8px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); background: #eef2ff;"><code>*-I-Balanced.gguf</code></td><td style="padding: 8px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); background: #eef2ff;"><b>约 80 GB</b></td><td style="padding: 8px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); background: #eef2ff;">边缘 Q6_K / 近缘与中间 <b>Q5_K</b>;shared Q8_0;attn Q6_K</td><td style="padding: 8px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); background: #eef2ff;"><b>中文用户推荐默认</b>,质量与稳定性更均衡</td></tr>
|
| 117 |
+
<tr><td style="padding: 8px 10px; background: white;"><code>*-I-Compact.gguf</code></td><td style="padding: 8px 10px; background: white;">约 52 GB</td><td style="padding: 8px 10px; background: white;">边缘 Q4_K / 中间 Q3_K;shared Q6_K;attn Q4_K</td><td style="padding: 8px 10px; background: white;">显存更紧,可接受更多 quant 噪声</td></tr>
|
| 118 |
+
</tbody></table>
|
| 119 |
+
<p style="margin: 12px 0 0 0; font-size: 13px; color: #334155; line-height: 1.7;"><b>选用建议:</b></p>
|
| 120 |
+
<ul style="margin: 8px 0 0 0; padding-left: 20px; font-size: 13px; color: #334155; line-height: 1.7;">
|
| 121 |
+
<li style="margin-bottom: 6px;"><b>中文为主:</b>请优先使用 <b>I-Balanced</b>。中间层用 Q5_K 而非 iq4_xs,观感通常更稳。显存紧张再退到 I-Compact。</li>
|
| 122 |
+
<li style="margin-bottom: 6px;"><b>英文 / 代码为主:</b>各档可用差异通常不大,可按显存与速度在 I-Balanced 与 I-Compact 之间选择;I-Quality 中间层为 IQ,按需试用。</li>
|
| 123 |
+
</ul>
|
| 124 |
+
</div>
|
| 125 |
+
</div>
|
| 126 |
+
|
| 127 |
+
<div style="border: 1px solid #cbd5e1; border-radius: 12px; overflow: hidden; background: #ffffff; box-shadow: 0 2px 4px rgba(0,0,0,0.02);">
|
| 128 |
+
<div style="background: linear-gradient(135deg, #6366f1 0%, #8b5cf6 100%); padding: 12px 16px; color: white; font-weight: 700; font-size: 14px; display: flex; align-items: center; gap: 8px;"><span>🚀</span> 使用方式(llama.cpp)</div>
|
| 129 |
+
<div style="padding: 16px; font-size: 13px; color: #334155; line-height: 1.7;">
|
| 130 |
+
<p style="margin: 0 0 8px 0;">请使用<strong>支持 Laguna 架构</strong>的 llama.cpp(poolside 分支 <code>laguna</code> 或已合并相应支持的构建)。</p>
|
| 131 |
+
<p style="margin: 0 0 8px 0; font-weight: bold; color: #312e81;">示例(推荐 I-Balanced)</p>
|
| 132 |
+
<pre style="margin: 0 0 12px 0; font-family: monospace; background: #f8fafc; padding: 10px 14px; border-radius: 6px; border: 1px solid #e2e8f0; font-size: 12px; color: #1e293b; white-space: pre-wrap;">./llama-server \
|
| 133 |
+
-m ./Laguna-S-2.1-Uncensored-APEX-I-Balanced.gguf \
|
| 134 |
+
--port 8080 \
|
| 135 |
+
-sm none \
|
| 136 |
+
--device rocm0 \
|
| 137 |
+
--ctx-size 131072 \
|
| 138 |
+
--flash-attn on \
|
| 139 |
+
--no-mmap \
|
| 140 |
+
--fit on \
|
| 141 |
+
--jinja \
|
| 142 |
+
--host 0.0.0.0</pre>
|
| 143 |
+
<ul style="margin: 0; padding-left: 20px;">
|
| 144 |
+
<li style="margin-bottom: 6px;">必须能识别 <code>general.architecture = laguna</code>。</li>
|
| 145 |
+
<li style="margin-bottom: 6px;">思考与工具相关参数请按客户端 / 服务端文档配置(如 <code>enable_thinking</code>、reasoning 相关选项)。</li>
|
| 146 |
+
<li style="margin-bottom: 6px;">若出现 <code>special_eos_id is not in special_eog_ids</code> 一类警告,多为 tokenizer 元数据提示,模型仍可加载;停写异常时请检查 stop / max_tokens 与模板。</li>
|
| 147 |
+
<li style="margin-bottom: 0;">本包为纯文本 GGUF,无 mmproj。</li>
|
| 148 |
+
</ul>
|
| 149 |
+
</div>
|
| 150 |
+
</div>
|
| 151 |
+
|
| 152 |
+
<div style="border: 1px solid #cbd5e1; border-radius: 12px; overflow: hidden; background: #ffffff; box-shadow: 0 2px 4px rgba(0,0,0,0.02);">
|
| 153 |
+
<div style="background: linear-gradient(135deg, #6366f1 0%, #8b5cf6 100%); padding: 12px 16px; color: white; font-weight: 700; font-size: 14px; display: flex; align-items: center; gap: 8px;"><span>🎛️</span> 推荐采样</div>
|
| 154 |
+
<div style="padding: 16px;">
|
| 155 |
+
<table style="width: 100%; border-collapse: collapse; font-size: 13px;"><tbody>
|
| 156 |
+
<tr><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); font-weight: bold; color: #334155; background: white;">通用 / 编码</td><td style="padding: 6px 10px; box-shadow: 0 1px 0 0 rgba(128,128,128,0.15); color: #4b5563; background: white; font-family: monospace;">temperature 0.6–1.0, top_p 0.95, top_k 20</td></tr>
|
| 157 |
+
<tr><td style="padding: 6px 10px; font-weight: bold; color: #334155; background: white;">更稳输出</td><td style="padding: 6px 10px; color: #4b5563; background: white;">适当降低 temperature</td></tr>
|
| 158 |
+
</tbody></table>
|
| 159 |
+
</div>
|
| 160 |
+
</div>
|
| 161 |
+
|
| 162 |
+
<div style="border: 1px solid #cbd5e1; border-radius: 12px; overflow: hidden; background: #ffffff; box-shadow: 0 2px 4px rgba(0,0,0,0.02);">
|
| 163 |
+
<div style="background: linear-gradient(135deg, #6366f1 0%, #8b5cf6 100%); padding: 12px 16px; color: white; font-weight: 700; font-size: 14px; display: flex; align-items: center; gap: 8px;"><span>🔧</span> 构建链路(摘要)</div>
|
| 164 |
+
<div style="padding: 16px; font-size: 13px; color: #334155; line-height: 1.7;">
|
| 165 |
+
<ol style="margin: 0; padding-left: 20px;">
|
| 166 |
+
<li style="margin-bottom: 6px;">基座:<code>poolside/Laguna-S-2.1</code></li>
|
| 167 |
+
<li style="margin-bottom: 6px;">abliterix 搜索 → <b>Trial 16</b> LoRA + MoE router 调整</li>
|
| 168 |
+
<li style="margin-bottom: 6px;">BF16 stream-merge → <code>Laguna-S-2.1-Uncensored</code></li>
|
| 169 |
+
<li style="margin-bottom: 6px;">poolside llama.cpp → BF16 GGUF + imatrix</li>
|
| 170 |
+
<li style="margin-bottom: 0;">APEX tensor-type-file + imatrix → I-Quality / I-Balanced / I-Compact</li>
|
| 171 |
+
</ol>
|
| 172 |
+
</div>
|
| 173 |
+
</div>
|
| 174 |
+
|
| 175 |
+
</div>
|
| 176 |
+
|
| 177 |
+
## 链接
|
| 178 |
+
|
| 179 |
+
- **原始模型**:https://huggingface.co/poolside/Laguna-S-2.1
|
| 180 |
+
- **发布博文**:https://poolside.ai/blog/introducing-laguna-s-2-1
|
| 181 |
+
- **OpenRouter**:https://openrouter.ai/poolside/laguna-s-2.1
|
| 182 |
+
- **官方 GGUF**:https://huggingface.co/poolside/Laguna-S-2.1-GGUF
|
| 183 |
+
- **poolside llama.cpp(laguna)**:https://github.com/poolsideai/llama.cpp/tree/laguna
|
| 184 |
+
- **abliterix**:https://github.com/wuwangzhang1216/abliterix
|
| 185 |
+
- **APEX**:https://github.com/mudler/apex-quant
|
| 186 |
+
- **许可证**:https://openmdw.ai/
|
| 187 |
+
|
| 188 |
+
## 免责声明
|
| 189 |
+
|
| 190 |
+
本仓库为社区衍生(行为修改 + 量化)版本,**非 poolside 官方发布**。使用风险自负;请遵守当地法律与上游许可证。
|