Update README.md
Browse files
README.md
CHANGED
|
@@ -16,74 +16,12 @@ TinyQwex-4x620M-MoE is a Mixure of Experts (MoE) made with the following models
|
|
| 16 |
* [Qwen/Qwen1.5-0.5B](https://huggingface.co/Qwen/Qwen1.5-0.5B)
|
| 17 |
* [Qwen/Qwen1.5-0.5B](https://huggingface.co/Qwen/Qwen1.5-0.5B)
|
| 18 |
|
| 19 |
-
## 🧩 Configuration
|
| 20 |
-
|
| 21 |
-
```yamlbase_model: Qwen/Qwen1.5-0.5B
|
| 22 |
-
experts:
|
| 23 |
-
- source_model: Qwen/Qwen1.5-0.5B
|
| 24 |
-
positive_prompts:
|
| 25 |
-
- "reasoning"
|
| 26 |
-
- "logic"
|
| 27 |
-
- "problem-solving"
|
| 28 |
-
- "critical thinking"
|
| 29 |
-
- "analysis"
|
| 30 |
-
- "synthesis"
|
| 31 |
-
- "evaluation"
|
| 32 |
-
- "decision-making"
|
| 33 |
-
- "judgment"
|
| 34 |
-
- "insight"
|
| 35 |
|
| 36 |
-
- source_model: Qwen/Qwen1.5-0.5B
|
| 37 |
-
positive_prompts:
|
| 38 |
-
- "program"
|
| 39 |
-
- "software"
|
| 40 |
-
- "develop"
|
| 41 |
-
- "build"
|
| 42 |
-
- "create"
|
| 43 |
-
- "design"
|
| 44 |
-
- "implement"
|
| 45 |
-
- "debug"
|
| 46 |
-
- "test"
|
| 47 |
-
- "code"
|
| 48 |
-
- "python"
|
| 49 |
-
- "programming"
|
| 50 |
-
- "algorithm"
|
| 51 |
-
- "function"
|
| 52 |
-
|
| 53 |
-
- source_model: Qwen/Qwen1.5-0.5B
|
| 54 |
-
positive_prompts:
|
| 55 |
-
- "storytelling"
|
| 56 |
-
- "narrative"
|
| 57 |
-
- "fiction"
|
| 58 |
-
- "creative writing"
|
| 59 |
-
- "plot"
|
| 60 |
-
- "characters"
|
| 61 |
-
- "dialogue"
|
| 62 |
-
- "setting"
|
| 63 |
-
- "emotion"
|
| 64 |
-
- "imagination"
|
| 65 |
-
- "scene"
|
| 66 |
-
- "story"
|
| 67 |
-
- "character"
|
| 68 |
-
|
| 69 |
-
- source_model: Qwen/Qwen1.5-0.5B
|
| 70 |
-
positive_prompts:
|
| 71 |
-
- "chat"
|
| 72 |
-
- "conversation"
|
| 73 |
-
- "dialogue"
|
| 74 |
-
- "discuss"
|
| 75 |
-
- "ask questions"
|
| 76 |
-
- "share thoughts"
|
| 77 |
-
- "explore ideas"
|
| 78 |
-
- "learn new things"
|
| 79 |
-
- "personal assistant"
|
| 80 |
-
- "friendly helper"
|
| 81 |
-
```
|
| 82 |
|
| 83 |
## 💻 Usage
|
| 84 |
|
| 85 |
```python
|
| 86 |
-
!pip install -qU transformers bitsandbytes accelerate
|
| 87 |
|
| 88 |
from transformers import AutoTokenizer
|
| 89 |
import transformers
|
|
@@ -91,15 +29,36 @@ import torch
|
|
| 91 |
|
| 92 |
model = "Isotonic/TinyQwex-4x620M-MoE"
|
| 93 |
|
| 94 |
-
tokenizer = AutoTokenizer.from_pretrained(
|
| 95 |
pipeline = transformers.pipeline(
|
| 96 |
"text-generation",
|
| 97 |
model=model,
|
| 98 |
-
model_kwargs={"torch_dtype": torch.
|
| 99 |
)
|
| 100 |
|
| 101 |
messages = [{"role": "user", "content": "Explain what a Mixture of Experts is in less than 100 words."}]
|
| 102 |
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
| 103 |
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
|
| 104 |
print(outputs[0]["generated_text"])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 105 |
```
|
|
|
|
| 16 |
* [Qwen/Qwen1.5-0.5B](https://huggingface.co/Qwen/Qwen1.5-0.5B)
|
| 17 |
* [Qwen/Qwen1.5-0.5B](https://huggingface.co/Qwen/Qwen1.5-0.5B)
|
| 18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
|
| 21 |
## 💻 Usage
|
| 22 |
|
| 23 |
```python
|
| 24 |
+
!pip install -qU transformers bitsandbytes accelerate eniops
|
| 25 |
|
| 26 |
from transformers import AutoTokenizer
|
| 27 |
import transformers
|
|
|
|
| 29 |
|
| 30 |
model = "Isotonic/TinyQwex-4x620M-MoE"
|
| 31 |
|
| 32 |
+
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-0.5B")
|
| 33 |
pipeline = transformers.pipeline(
|
| 34 |
"text-generation",
|
| 35 |
model=model,
|
| 36 |
+
model_kwargs={"torch_dtype": torch.bfloat16, "load_in_4bit": True},
|
| 37 |
)
|
| 38 |
|
| 39 |
messages = [{"role": "user", "content": "Explain what a Mixture of Experts is in less than 100 words."}]
|
| 40 |
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
| 41 |
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
|
| 42 |
print(outputs[0]["generated_text"])
|
| 43 |
+
```
|
| 44 |
+
|
| 45 |
+
## 🧩 Configuration
|
| 46 |
+
|
| 47 |
+
```yamlbase_model: Qwen/Qwen1.5-0.5B
|
| 48 |
+
experts:
|
| 49 |
+
- source_model: Qwen/Qwen1.5-0.5B
|
| 50 |
+
positive_prompts:
|
| 51 |
+
- "reasoning"
|
| 52 |
+
|
| 53 |
+
- source_model: Qwen/Qwen1.5-0.5B
|
| 54 |
+
positive_prompts:
|
| 55 |
+
- "program"
|
| 56 |
+
|
| 57 |
+
- source_model: Qwen/Qwen1.5-0.5B
|
| 58 |
+
positive_prompts:
|
| 59 |
+
- "storytelling"
|
| 60 |
+
|
| 61 |
+
- source_model: Qwen/Qwen1.5-0.5B
|
| 62 |
+
positive_prompts:
|
| 63 |
+
- "Instruction following assistant"
|
| 64 |
```
|