L3-12B-Lunaris-v1

L3-12B-Lunaris-v1 is a self merge of the following model using LazyMergekit with --clone-tensors argument added:

• Sao10K/L3-8B-Lunaris-v1

Works best with lower temperature, between 0.8-0.9.

🧩 Configuration

dtype: bfloat16
merge_method: passthrough
slices:
- sources:
  - layer_range: [0, 8]
    model: Sao10K/L3-8B-Lunaris-v1
    parameters:
      scale_rules:
      - filter: value
        value: 0.8
- sources:
  - layer_range: [8, 16]
    model: Sao10K/L3-8B-Lunaris-v1
    parameters:
      scale_rules:
      - filter: value
        value: 0.8
- sources:
  - layer_range: [16, 24]
    model: Sao10K/L3-8B-Lunaris-v1
    parameters:
      scale_rules:
      - filter: value
        value: 1.0
- sources:
  - layer_range: [24, 32]
    model: Sao10K/L3-8B-Lunaris-v1
    parameters:
      scale_rules:
      - filter: value
        value: 1.0
- sources:
  - layer_range: [0, 8]
    model: Sao10K/L3-8B-Lunaris-v1
    parameters:
      scale_rules:
      - filter: value
        value: 0.7
- sources:
  - layer_range: [8, 16]
    model: Sao10K/L3-8B-Lunaris-v1
    parameters:
      scale_rules:
      - filter: value
        value: 0.7

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "Tremontaine/L3-12B-Lunaris-v1"
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"])

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	25.38
IFEval (0-Shot)	69.09
BBH (3-Shot)	32.18
MATH Lvl 5 (4-Shot)	8.16
GPQA (0-shot)	7.94
MuSR (0-shot)	4.05
MMLU-PRO (5-shot)	30.83