Update README.md
Browse files
README.md
CHANGED
@@ -15,42 +15,126 @@ tags:
|
|
15 |
This model was converted to GGUF format from [`ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B`](https://huggingface.co/ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B) using llama.cpp via the ggml.ai's [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space.
|
16 |
Refer to the [original model card](https://huggingface.co/ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B) for more details on the model.
|
17 |
|
18 |
-
## Use with llama.cpp
|
19 |
-
Install
|
|
|
|
|
20 |
|
21 |
-
|
22 |
-
brew install llama.cpp
|
23 |
|
24 |
-
|
25 |
-
Invoke the llama.cpp server or the CLI.
|
26 |
|
27 |
-
|
28 |
-
```bash
|
29 |
-
llama-cli --hf-repo ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B-Q4_0-GGUF --hf-file llama3.1-hermes3-supernova-8b-l3.1-purosani-2-8b-q4_0.gguf -p "The meaning to life and the universe is"
|
30 |
-
```
|
31 |
|
32 |
-
|
33 |
-
```bash
|
34 |
-
llama-server --hf-repo ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B-Q4_0-GGUF --hf-file llama3.1-hermes3-supernova-8b-l3.1-purosani-2-8b-q4_0.gguf -c 2048
|
35 |
-
```
|
36 |
|
37 |
-
|
|
|
38 |
|
39 |
-
|
40 |
-
```
|
41 |
-
git clone https://github.com/ggerganov/llama.cpp
|
42 |
-
```
|
43 |
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
48 |
|
49 |
-
Step 3: Run inference through the main binary.
|
50 |
-
```
|
51 |
-
./llama-cli --hf-repo ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B-Q4_0-GGUF --hf-file llama3.1-hermes3-supernova-8b-l3.1-purosani-2-8b-q4_0.gguf -p "The meaning to life and the universe is"
|
52 |
-
```
|
53 |
-
or
|
54 |
-
```
|
55 |
-
./llama-server --hf-repo ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B-Q4_0-GGUF --hf-file llama3.1-hermes3-supernova-8b-l3.1-purosani-2-8b-q4_0.gguf -c 2048
|
56 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
This model was converted to GGUF format from [`ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B`](https://huggingface.co/ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B) using llama.cpp via the ggml.ai's [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space.
|
16 |
Refer to the [original model card](https://huggingface.co/ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B) for more details on the model.
|
17 |
|
18 |
+
## Use with LMStudio or llama.cpp
|
19 |
+
Install LMStudio through their website: [`LM Studio`](https://lmstudio.ai/)
|
20 |
+
or
|
21 |
+
Install llama.cpp through their instructions on Github (works on Mac and Linux) [`Llama.cpp Usage`](https://github.com/ggerganov/llama.cpp/blob/master/README.md#usage)
|
22 |
|
23 |
+
# Original Model Card:
|
|
|
24 |
|
25 |
+
**ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B** is a cutting-edge merged model that blends the best features of two highly optimized architectures to create an **advanced**, **adaptive**, and **powerful** model. Whether for scientific research, complex instruction-following, or immersive roleplay scenarios, this model excels at every task itβs thrown into.
|
|
|
26 |
|
27 |
+
## π Family Tree
|
|
|
|
|
|
|
28 |
|
29 |
+
This model is a merger of the following:
|
|
|
|
|
|
|
30 |
|
31 |
+
- [**kotyKD/Llama3.1-Hermes3-SuperNovaLite-merged-with-base-8B**](https://huggingface.co/kotyKD/Llama3.1-Hermes3-SuperNovaLite-merged-with-base-8B)
|
32 |
+
- [**djuna/L3.1-Purosani-2-8B**](https://huggingface.co/djuna/L3.1-Purosani-2-8B)
|
33 |
|
34 |
+
These parent models are themselves the result of **complex merges** of various high-performance models, making ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B a **super hybrid** capable of handling diverse tasks with efficiency and finesse.
|
|
|
|
|
|
|
35 |
|
36 |
+
## 𧬠Detailed Model Lineage
|
37 |
+
|
38 |
+
### **A: kotyKD/Llama3.1-Hermes3-SuperNovaLite-merged-with-base-8B**
|
39 |
+
|
40 |
+
Merged using the **TIES merge method**, this model utilizes **unsloth/Meta-Llama-3.1-8B** as its base, combining:
|
41 |
+
|
42 |
+
- **arcee-ai/Llama-3.1-SuperNova-Lite**: A distilled 8B parameter version of the **Llama-3.1-405B-Instruct** model, designed to maintain high performance while minimizing resource consumption. Its training, via **EvolKit**, offers instruction-following precision and domain-specific adaptability.
|
43 |
+
- **NousResearch/Hermes-3-Llama-3.1-8B**: Known for its robustness, this model enhances long-range contextual understanding, making it ideal for complex, multi-layered tasks.
|
44 |
+
|
45 |
+
### **B: djuna/L3.1-Purosani-2-8B**
|
46 |
+
|
47 |
+
This merge incorporates:
|
48 |
+
|
49 |
+
- **hf-100/Llama-3-Spellbound-Instruct-8B-0.3**
|
50 |
+
- **arcee-ai/Llama-3.1-SuperNova-Lite**
|
51 |
+
- **grimjim/Llama-3-Instruct-abliteration-LoRA-8B**
|
52 |
+
- **THUDM/LongWriter-llama3.1-8B**, capable of generating over **10,000 words** in one pass, making it perfect for long-form content generation.
|
53 |
+
|
54 |
+
Further contributors include **ResplendentAI/Smarts_Llama3** and **djuna/L3.1-Suze-Vume-2-calc**, making this model highly adaptable to a broad range of applications.
|
55 |
+
|
56 |
+
## π οΈ Merge Details
|
57 |
+
|
58 |
+
The model was merged using the **della merge method** with **kromeurus/L3.1-Aglow-Vulca-v0.1-8B** as the base. This method, combined with the following models, ensures both **precision** and **adaptability**:
|
59 |
+
|
60 |
+
- **djuna/L3.1-Noraian**
|
61 |
+
- **Casual-Autopsy/L3-Super-Nova-RP-8B**
|
62 |
+
- **TheDrummer/Llama-3SOME-8B-v2**
|
63 |
+
- **djuna/L3.1-ForStHS**
|
64 |
+
- **Blackroot/Llama-3-8B-Abomination-LORA**
|
65 |
+
|
66 |
+
## π§ Technical Configuration
|
67 |
+
|
68 |
+
The merging process used advanced methods to ensure smooth integration and consistent performance across various tasks:
|
69 |
+
|
70 |
+
```yaml
|
71 |
+
slices:
|
72 |
+
- sources:
|
73 |
+
- model: kotyKD/Llama3.1-Hermes3-SuperNovaLite-merged-with-base-8B
|
74 |
+
layer_range: [0, 32]
|
75 |
+
- model: djuna/L3.1-Purosani-2-8B
|
76 |
+
layer_range: [0, 32]
|
77 |
+
merge_method: slerp
|
78 |
+
base_model: djuna/L3.1-Purosani-2-8B
|
79 |
+
parameters:
|
80 |
+
t:
|
81 |
+
- filter: self_attn
|
82 |
+
value: [0, 0.5, 0.3, 0.7, 1]
|
83 |
+
- filter: mlp
|
84 |
+
value: [1, 0.5, 0.7, 0.3, 0]
|
85 |
+
- value: 0.5
|
86 |
+
dtype: bfloat16
|
87 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
88 |
```
|
89 |
+
|
90 |
+
## π― Extended Support for Roleplay & Immersive Storytelling
|
91 |
+
|
92 |
+
**ZeroXClem/Llama3.1-Hermes3-SuperNova-8B-L3.1-Purosani-2-8B** has been optimized for **extended roleplay support**, making it an exceptional choice for **interactive storytelling** and **deep character development**. With its ability to understand long-form context and generate cohesive responses over extensive interactions, this model excels in:
|
93 |
+
|
94 |
+
- **Character-driven interactions**: Develop rich, nuanced personalities that respond in believable and engaging ways.
|
95 |
+
- **World-building & Lore creation**: Create vast, interconnected universes with intricate lore, all generated in real-time.
|
96 |
+
- **Dynamic NPC dialogues**: Use the model to generate complex, reactive conversations for game NPCs, offering a fluid, immersive experience for players.
|
97 |
+
|
98 |
+
## π Key Features & Capabilities
|
99 |
+
|
100 |
+
### **Advanced Roleplay and Long-Form Content Generation**
|
101 |
+
|
102 |
+
With models like **THUDM/LongWriter-llama3.1-8B** contributing their expertise, this model is perfect for generating **long-form narratives** while maintaining coherence and creativity.
|
103 |
+
|
104 |
+
### **Instruction Following & Task Adaptability**
|
105 |
+
|
106 |
+
Combining the capabilities of **Hermes** and **SuperNovaLite**, this model can efficiently follow detailed instructions, making it ideal for:
|
107 |
+
|
108 |
+
- **Task automation**
|
109 |
+
- **Virtual assistants**
|
110 |
+
- **Research generation**
|
111 |
+
|
112 |
+
### **Efficiency Without Compromise**
|
113 |
+
|
114 |
+
Distilled models like **SuperNovaLite** ensure that this model delivers high performance without the extensive resource requirements of larger models.
|
115 |
+
|
116 |
+
## π― Use Case & Applications
|
117 |
+
|
118 |
+
- **Roleplay & Interactive Storytelling**: The perfect companion for storytellers, RPG enthusiasts, and game developers. Whether crafting dynamic NPC interactions or generating deep, immersive worlds, this model can handle it all.
|
119 |
+
- **Instruction-based AI**: With enhanced instruction-following abilities, this model is ideal for developing intelligent assistants or chatbots that require high accuracy and quick adaptability.
|
120 |
+
- **Long-Form Writing**: From novels to research papers, this model can generate lengthy, well-structured content with ease, thanks to its extensive training on long-form data.
|
121 |
+
|
122 |
+
## π License
|
123 |
+
|
124 |
+
This model is open-sourced under the **Apache-2.0 License**, allowing others to use and modify it freely, as long as they give proper attribution.
|
125 |
+
|
126 |
+
## π‘ Tags
|
127 |
+
|
128 |
+
- merge
|
129 |
+
- mergekit
|
130 |
+
- lazymergekit
|
131 |
+
- Hermes3
|
132 |
+
- SuperNovaLite
|
133 |
+
- Purosani
|
134 |
+
- Llama3.1
|
135 |
+
- kotyKD/Llama3.1-Hermes3-SuperNovaLite-merged-with-base-8B
|
136 |
+
- djuna/L3.1-Purosani-2-8B
|
137 |
+
- instruction-following
|
138 |
+
- long-form-generation
|
139 |
+
- roleplay
|
140 |
+
- storytelling
|