Celestria-MoE-8x10.7b
The Celestria Series, is the "Big Sister" of the Lumosia and Umbra Series. It is an experiment born from the collective wisdom of the AI community, a mosaic of the eight best-performing Solar models (By my prefrences)
its 3am.... again, I have a tendency to do this apparently so im not going to get to creative on this card.
With this model I have created positive and negative prompt sentances:
[Celestria Series] Based on prompt sentances.
[Umbra Series] based on prompt keywords.
[Lumosia Series] based on prompt topics.
Let me know what you think!
Template:
### System:
### USER:{prompt}
### Assistant:
Settings:
Temp: 1.0
min-p: 0.02-0.1
Evals:
To come
- Avg:
- ARC:
- HellaSwag:
- MMLU:
- T-QA:
- Winogrande:
- GSM8K:
Examples:
Example 1:
User:
Celestria:
Example 2:
User:
Celestria:
𧩠Configuration
yaml
experts:
- source_model: Fimbulvetr-10.7B-v1
- source_model: PiVoT-10.7B-Mistral-v0.2-RP
- source_model: UNA-POLAR-10.7B-InstructMath-v2
- source_model: LMCocktail-10.7B-v1
- source_model: CarbonBeagle-11B
- source_model: SOLARC-M-10.7B
- source_model: Nous-Hermes-2-SOLAR-10.7B-MISALIGNED
- source_model: CarbonVillain-en-10.7B-v4
π» Usage
python
!pip install -qU transformers bitsandbytes accelerate
from transformers import AutoTokenizer
import transformers
import torch
model = "Steelskull/Celestria-MoE-8x10.7b"
tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
"text-generation",
model=model,
model_kwargs={"torch_dtype": torch.bfloat16, "load_in_4bit": True},
)
messages = [{"role": "user", "content": "Explain what a Mixture of Experts is in less than 100 words."}]
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])
- Downloads last month
- 11
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.