Marcoro14-7B-slerp

This model is a merge of the following models made with mergekit:

πŸ† Evaluation

Marcoro14-7B-slerp is the best-performing 7B LLM on the Open LLM Leaderboard (rank 1 below is 9B):

I also evaluated it using Nous' benchmark suite and obtained the following results:

Model AGIEval GPT4ALL TruthfulQA Bigbench Average
Marcoro14-7B-slerp 44.66 76.24 64.15 45.64 57.67
OpenHermes-2.5-Mistral-7B 43.07 73.12 53.04 40.96 52.57
Change +1.59 +3.12 +11.11 +4.68 +5.1

AGIEval

Task Version Metric Value Stderr
agieval_aqua_rat 0 acc 26.38 Β± 2.77
acc_norm 24.41 Β± 2.70
agieval_logiqa_en 0 acc 38.25 Β± 1.91
acc_norm 39.32 Β± 1.92
agieval_lsat_ar 0 acc 24.35 Β± 2.84
acc_norm 25.22 Β± 2.87
agieval_lsat_lr 0 acc 50.00 Β± 2.22
acc_norm 50.59 Β± 2.22
agieval_lsat_rc 0 acc 62.83 Β± 2.95
acc_norm 62.08 Β± 2.96
agieval_sat_en 0 acc 79.61 Β± 2.81
acc_norm 79.61 Β± 2.81
agieval_sat_en_without_passage 0 acc 45.15 Β± 3.48
acc_norm 45.63 Β± 3.48
agieval_sat_math 0 acc 33.18 Β± 3.18
acc_norm 30.45 Β± 3.11

Average: 44.66%

GPT4ALL

Task Version Metric Value Stderr
arc_challenge 0 acc 63.91 Β± 1.40
acc_norm 64.93 Β± 1.39
arc_easy 0 acc 86.07 Β± 0.71
acc_norm 83.75 Β± 0.76
boolq 1 acc 88.56 Β± 0.56
hellaswag 0 acc 67.31 Β± 0.47
acc_norm 85.28 Β± 0.35
openbookqa 0 acc 36.40 Β± 2.15
acc_norm 48.20 Β± 2.24
piqa 0 acc 82.59 Β± 0.88
acc_norm 84.39 Β± 0.85
winogrande 0 acc 78.53 Β± 1.15

Average: 76.24%

TruthfulQA

Task Version Metric Value Stderr
truthfulqa_mc 1 mc1 46.88 Β± 1.75
mc2 64.15 Β± 1.52

Average: 64.15%

Bigbench

Task Version Metric Value Stderr
bigbench_causal_judgement 0 multiple_choice_grade 56.32 Β± 3.61
bigbench_date_understanding 0 multiple_choice_grade 66.40 Β± 2.46
bigbench_disambiguation_qa 0 multiple_choice_grade 45.35 Β± 3.11
bigbench_geometric_shapes 0 multiple_choice_grade 20.33 Β± 2.13
exact_str_match 4.74 Β± 1.12
bigbench_logical_deduction_five_objects 0 multiple_choice_grade 30.00 Β± 2.05
bigbench_logical_deduction_seven_objects 0 multiple_choice_grade 21.43 Β± 1.55
bigbench_logical_deduction_three_objects 0 multiple_choice_grade 52.33 Β± 2.89
bigbench_movie_recommendation 0 multiple_choice_grade 39.20 Β± 2.19
bigbench_navigate 0 multiple_choice_grade 53.90 Β± 1.58
bigbench_reasoning_about_colored_objects 0 multiple_choice_grade 72.15 Β± 1.00
bigbench_ruin_names 0 multiple_choice_grade 52.46 Β± 2.36
bigbench_salient_translation_error_detection 0 multiple_choice_grade 25.75 Β± 1.38
bigbench_snarks 0 multiple_choice_grade 72.38 Β± 3.33
bigbench_sports_understanding 0 multiple_choice_grade 73.63 Β± 1.40
bigbench_temporal_sequences 0 multiple_choice_grade 45.70 Β± 1.58
bigbench_tracking_shuffled_objects_five_objects 0 multiple_choice_grade 23.44 Β± 1.20
bigbench_tracking_shuffled_objects_seven_objects 0 multiple_choice_grade 18.51 Β± 0.93
bigbench_tracking_shuffled_objects_three_objects 0 multiple_choice_grade 52.33 Β± 2.89

Average: 45.64%

Average score: 57.67%

🧩 Configuration

slices:
  - sources:
      - model: AIDC-ai-business/Marcoroni-7B-v3
        layer_range: [0, 32]
      - model: EmbeddedLLM/Mistral-7B-Merge-14-v0.1
        layer_range: [0, 32]
merge_method: slerp
base_model: AIDC-ai-business/Marcoroni-7B-v3
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

πŸ’» Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "mlabonne/Marcoro14-7B-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"])

Output:

A large language model is a type of artificial intelligence (AI) system that has been trained on vast amounts of text data. It's designed to understand and generate human-like language, making predictions on what words or phrases might come next in a sentence or document. These models use complex algorithms and neural network architectures to learn from the data and improve their performance over time. Some well-known large language models include GPT-3 from OpenAI and BERT from Google.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 73.01
AI2 Reasoning Challenge (25-Shot) 69.80
HellaSwag (10-Shot) 87.13
MMLU (5-Shot) 65.11
TruthfulQA (0-shot) 63.54
Winogrande (5-shot) 81.61
GSM8k (5-shot) 70.89
Downloads last month
3,164
Safetensors
Model size
7.24B params
Tensor type
BF16
Β·
Inference Examples
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.

Model tree for mlabonne/Marcoro14-7B-slerp

Finetunes
16 models
Merges
5 models
Quantizations
2 models

Collection including mlabonne/Marcoro14-7B-slerp

Evaluation results