File size: 6,470 Bytes
4eab9ea 38aba7d 4eab9ea 777b2c7 4eab9ea 38aba7d 8f4b077 38aba7d 8f4b077 38aba7d 8f4b077 38aba7d 8f4b077 38aba7d 8f4b077 38aba7d 8f4b077 38aba7d 8f4b077 38aba7d 8f4b077 38aba7d 4eab9ea 38aba7d 5aa70ec 38aba7d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 |
---
license: apache-2.0
tags:
- merge
- mergekit
---
# Marcoro14-7B-slerp
This model is a merge of the following models made with [mergekit](https://github.com/cg123/mergekit):
* [AIDC-ai-business/Marcoroni-7B-v3](https://huggingface.co/AIDC-ai-business/Marcoroni-7B-v3)
* [EmbeddedLLM/Mistral-7B-Merge-14-v0.1](https://huggingface.co/EmbeddedLLM/Mistral-7B-Merge-14-v0.1)
## 🏆 Evaluation
Marcoro14-7B-slerp is the second best-performing 7B LLM on the Open LLM Leaderboard:
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|± | 2.77|
| | |acc_norm|24.41|± | 2.70|
|agieval_logiqa_en | 0|acc |38.25|± | 1.91|
| | |acc_norm|39.32|± | 1.92|
|agieval_lsat_ar | 0|acc |24.35|± | 2.84|
| | |acc_norm|25.22|± | 2.87|
|agieval_lsat_lr | 0|acc |50.00|± | 2.22|
| | |acc_norm|50.59|± | 2.22|
|agieval_lsat_rc | 0|acc |62.83|± | 2.95|
| | |acc_norm|62.08|± | 2.96|
|agieval_sat_en | 0|acc |79.61|± | 2.81|
| | |acc_norm|79.61|± | 2.81|
|agieval_sat_en_without_passage| 0|acc |45.15|± | 3.48|
| | |acc_norm|45.63|± | 3.48|
|agieval_sat_math | 0|acc |33.18|± | 3.18|
| | |acc_norm|30.45|± | 3.11|
Average: 44.66%
### GPT4ALL
| Task |Version| Metric |Value| |Stderr|
|-------------|------:|--------|----:|---|-----:|
|arc_challenge| 0|acc |63.91|± | 1.40|
| | |acc_norm|64.93|± | 1.39|
|arc_easy | 0|acc |86.07|± | 0.71|
| | |acc_norm|83.75|± | 0.76|
|boolq | 1|acc |88.56|± | 0.56|
|hellaswag | 0|acc |67.31|± | 0.47|
| | |acc_norm|85.28|± | 0.35|
|openbookqa | 0|acc |36.40|± | 2.15|
| | |acc_norm|48.20|± | 2.24|
|piqa | 0|acc |82.59|± | 0.88|
| | |acc_norm|84.39|± | 0.85|
|winogrande | 0|acc |78.53|± | 1.15|
Average: 76.24%
### TruthfulQA
| Task |Version|Metric|Value| |Stderr|
|-------------|------:|------|----:|---|-----:|
|truthfulqa_mc| 1|mc1 |46.88|± | 1.75|
| | |mc2 |64.15|± | 1.52|
Average: 64.15%
### Bigbench
| Task |Version| Metric |Value| |Stderr|
|------------------------------------------------|------:|---------------------|----:|---|-----:|
|bigbench_causal_judgement | 0|multiple_choice_grade|56.32|± | 3.61|
|bigbench_date_understanding | 0|multiple_choice_grade|66.40|± | 2.46|
|bigbench_disambiguation_qa | 0|multiple_choice_grade|45.35|± | 3.11|
|bigbench_geometric_shapes | 0|multiple_choice_grade|20.33|± | 2.13|
| | |exact_str_match | 4.74|± | 1.12|
|bigbench_logical_deduction_five_objects | 0|multiple_choice_grade|30.00|± | 2.05|
|bigbench_logical_deduction_seven_objects | 0|multiple_choice_grade|21.43|± | 1.55|
|bigbench_logical_deduction_three_objects | 0|multiple_choice_grade|52.33|± | 2.89|
|bigbench_movie_recommendation | 0|multiple_choice_grade|39.20|± | 2.19|
|bigbench_navigate | 0|multiple_choice_grade|53.90|± | 1.58|
|bigbench_reasoning_about_colored_objects | 0|multiple_choice_grade|72.15|± | 1.00|
|bigbench_ruin_names | 0|multiple_choice_grade|52.46|± | 2.36|
|bigbench_salient_translation_error_detection | 0|multiple_choice_grade|25.75|± | 1.38|
|bigbench_snarks | 0|multiple_choice_grade|72.38|± | 3.33|
|bigbench_sports_understanding | 0|multiple_choice_grade|73.63|± | 1.40|
|bigbench_temporal_sequences | 0|multiple_choice_grade|45.70|± | 1.58|
|bigbench_tracking_shuffled_objects_five_objects | 0|multiple_choice_grade|23.44|± | 1.20|
|bigbench_tracking_shuffled_objects_seven_objects| 0|multiple_choice_grade|18.51|± | 0.93|
|bigbench_tracking_shuffled_objects_three_objects| 0|multiple_choice_grade|52.33|± | 2.89|
Average: 45.64%
Average score: 57.67%
## 🧩 Configuration
```yaml
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
```
## 💻 Usage
```python
!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"])
```
|