Dense2MoE from scratch
Collection
7 items
β’
Updated
cosmo-8x334M-random-router-random_init is a Mixture of Experts (MoE) made with the following models using LazyMergekit:
gate_mode: random # one of "hidden", "cheap_embed", or "random"
dtype: bfloat16 # output dtype (float32, float16, or bfloat16)
experts_per_token: 2
base_model: yentinglin/cosmo-334M-random-1
experts:
- source_model: yentinglin/cosmo-334M-random-1
- source_model: yentinglin/cosmo-334M-random-2
- source_model: yentinglin/cosmo-334M-random-3
- source_model: yentinglin/cosmo-334M-random-4
- source_model: yentinglin/cosmo-334M-random-5
- source_model: yentinglin/cosmo-334M-random-6
- source_model: yentinglin/cosmo-334M-random-7
- source_model: yentinglin/cosmo-334M-random-8
dtype: bfloat16
!pip install -qU transformers bitsandbytes accelerate
from transformers import AutoTokenizer
import transformers
import torch
model = "yentinglin/cosmo-8x334M-random-router-random_init"
tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
"text-generation",
model=model,
model_kwargs={"torch_dtype": torch.float16, "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"])