LLaMa-3-8b-Spanish-slerp
LLaMa-3-8b-Spanish-slerp is a merge of the following models using LazyMergekit:
🧩 Configuration
slices:
- sources:
- model: Kukedlc/LLaMa-3-8b-en-es-v1
layer_range: [0, 32]
- model: Kukedlc/LLaMa-3-8b-Spanish-RAG-v1
layer_range: [0, 32]
merge_method: slerp
base_model: Kukedlc/LLaMa-3-8b-Spanish-RAG-v1
parameters:
t:
- filter: self_attn
value: [0, 0.5, 0.3, 0.7, 1]
- filter: mlp
value: [1, 0.5, 0.7, 0.3, 0]
- value: 0.5
dtype: bfloat16
💻 Usage
!pip install -qU transformers accelerate
from transformers import AutoTokenizer
import transformers
import torch
model = "Kukedlc/LLaMa-3-8b-Spanish-slerp"
messages = [{"role": "user", "content": "What is a large language model?"}]
tokenizer = AutoTokenizer.from_pretrained(model)
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
pipeline = transformers.pipeline(
"text-generation",
model=model,
torch_dtype=torch.float16,
device_map="auto",
)
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"])