automerger's picture
Update README.md
76972ed verified
metadata
license: apache-2.0
tags:
  - merge
  - mergekit
  - lazymergekit
  - automerger
base_model:
  - automerger/YamShadow-7B
  - yam-peleg/Experiment28-7B

πŸ§ͺ YamshadowExperiment28-7B

image/jpeg

πŸŽ‰ YamshadowExperiment28-7B is currently the best-performing 7B model on the Open LLM Leaderboard (08 Apr 24). Use it with caution, as it is likely a sign of overfitting the benchmarks.

YamshadowExperiment28-7B is an automated merge created by Maxime Labonne using the following configuration.

πŸ” Applications

This model uses a context window of 8k. I recommend using it with the Alpaca chat template (works perfectly with LM Studio).

The model can sometimes break and output a lot of "INST". From my experience, its excellent results on the Open LLM Leaderboard are probably a sign of overfitting.

⚑ Quantized models

πŸ† Evaluation

Open LLM Leaderboard

YamshadowExperiment28-7B is currently the best-performing 7B model on the Open LLM Leaderboard (08 Apr 24).

image/png

EQ-bench

Thanks to Samuel J. Paech, who kindly ran the evaluation.

image/png

Nous

Evaluation performed using LLM AutoEval. See the entire leaderboard here.

image/png

🌳 Model Family Tree

image/png

🧩 Configuration

slices:
  - sources:
      - model: automerger/YamShadow-7B
        layer_range: [0, 32]
      - model: yam-peleg/Experiment28-7B
        layer_range: [0, 32]
merge_method: slerp
base_model: automerger/YamShadow-7B
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
random_seed: 0

πŸ’» Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "automerger/YamshadowExperiment28-7B"
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"])