Dansxpantheon RP Engine V1.2 24b Small Instruct Ties Merge

🚀 DansXPantheon-RP-Engine-V1.2-24b-Small-Instruct-Ties-Merge

DansXPantheon-RP-Engine-V1.2-24b-Small-Instruct-Ties-Merge是一個經過優化的模型，它修復了EOS令牌問題，實現了更好的合併效果，減少了冗餘輸出，甚至還能使用表情符號，為角色扮演和故事創作等場景帶來了更豐富的體驗。

✨ 主要特性

• 修復了EOS令牌問題。
• 更好的合併效果，減少冗餘輸出。
• 不知為何能夠使用表情符號，有點像Gemma - 3，雖然不清楚原理，但有時很可愛。

📚 詳細文檔

設置

"temperature": 0.8,
"top_k": 40,
"top_p": 0.95,
"min_p": 0.05,
"repeat_last_n": 40,
"repeat_penalty": 1.2,

在Ollama上運行

這些是非imatrix版本。我稍後會發布imatrix版本。

GGUF 3位Q3_K_M版本，約需27GB的顯存/內存：

ollama run hf.co/h34v7/DansXPantheon-RP-Engine-V1.2-24b-Small-Instruct-Ties-Merge-GGUF:Q3_K_M

GGUF 4位Q4_K_M版本，約需30GB的顯存/內存：

ollama run hf.co/h34v7/DansXPantheon-RP-Engine-V1.2-24b-Small-Instruct-Ties-Merge-GGUF:Q4_K_M

GGUF 5位Q5_K_M版本，約需33GB的顯存/內存：

ollama run hf.co/h34v7/DansXPantheon-RP-Engine-V1.2-24b-Small-Instruct-Ties-Merge-GGUF:Q5_K_M

輸出示例

使用GGUF Q5 - K - M默認的Ollama設置。FP - 16版本在此。

User: hi
Assistance: Hello! It's nice to meet you. How are you doing today? üòä

User: alright tell me some joke
Assistance: Sure, here's a lighthearted joke for you:
What do you call a fake noodle?
An impasta! 🍝🤪
Hope that made you chuckle. If you'd like another one or need any other assistance, just let me know!

User: Give me 5 non-formal ways to say 'See you later' in French.
Assistance: Certainly! Here are five casual ways to say "see you later" in French:
1. À plus tard - This is a very common and informal way to bid farewell.
2. Salut à plus - A more relaxed version that combines "hello/goodbye" (salut) with "see you later."
3. Ciao à plus - Using the Italian "ciao" gives it an even more casual feel.
4. Bisous à plus tard - This one adds a friendly touch by including "kisses" before saying see you later.
5. À la prochaine - Literally means "until next time," but is often used casually as well.
Remember that French people generally use "tu" (informal) with friends and family, while "vous" is more formal for strangers or in professional settings. These phrases work best when using the informal "tu." 🇫🇷✨

🔧 技術細節

合併詳情

此模型是使用mergekit對預訓練語言模型進行合併得到的。

合併方法

該模型使用TIES合併方法，以Gryphe/Pantheon - RP - 1.8 - 24b - Small - 3.1為基礎進行合併。

合併的模型

合併中包含以下模型：

• PocketDoc/Dans - PersonalityEngine - V1.2.0 - 24b
• Gryphe/Pantheon - RP - 1.8 - 24b - Small - 3.1

配置

以下是用於生成此模型的YAML配置：

models:
  - model: Gryphe/Pantheon-RP-1.8-24b-Small-3.1
    parameters:
      density: 0.7
      weight: 0.7
  - model: PocketDoc/Dans-PersonalityEngine-V1.2.0-24b
    parameters:
      density: 0.5
      weight: 0.5
      
merge_method: ties
base_model: Gryphe/Pantheon-RP-1.8-24b-Small-3.1
parameters:
  normalize: false
  int8_mask: true
dtype: float16
tokenizer: 
 source: union

我的系統配置

Motherboard: Asus Maximus Z790
CPU: Intel i9-13900KS
RAM: 96 GB 6400mhz
GPU: RTX-4090 OC 24GB

信息表格