File size: 8,205 Bytes
d803e15 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 |
---
tags:
- merge
- mergekit
- lazymergekit
- flemmingmiguel/NeuDist-Ro-7B
- johannhartmann/Brezn3
- ResplendentAI/Flora_DPO_7B
base_model:
- flemmingmiguel/NeuDist-Ro-7B
- johannhartmann/Brezn3
- ResplendentAI/Flora_DPO_7B
language:
- de
- en
---
# Spaetzle-v8-7b
This model is supposed to show adequate performance in German and English on a number of tasks, while mostly behaving well, that is, without rambling on, intermixing tokens from different templates in training and adapting, etc.
It is mostly a quick test, and considerably weaker in German grammar and orthography than DiscoLM e.g., but for use cases where this is not too important, but e.g. instruction following, reasoning, etc, it might actually be a little bit preferable.
It is a merge of the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing):
* [flemmingmiguel/NeuDist-Ro-7B](https://huggingface.co/flemmingmiguel/NeuDist-Ro-7B)
* [johannhartmann/Brezn3](https://huggingface.co/johannhartmann/Brezn3)
* [ResplendentAI/Flora_DPO_7B](https://huggingface.co/ResplendentAI/Flora_DPO_7B)
* on the basis of [mayflowergmbh/Wiedervereinigung-7b-dpo-laser](https://huggingface.co/mayflowergmbh/Wiedervereinigung-7b-dpo-laser)
All credits are due to the creators of those original models and the training datasets involved.
For a suitable quantized version, try [cstr/Spaetzle-v8-7b-GGUF](https://huggingface.co/cstr/Spaetzle-v8-7b-GGUF)
## Evaluation
[Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_cstr__Spaetzle-v8-7b)
| Metric |Value|
|---------------------------------|----:|
|Avg. |72.27|
|AI2 Reasoning Challenge (25-Shot)|68.69|
|HellaSwag (10-Shot) |86.68|
|MMLU (5-Shot) |64.60|
|TruthfulQA (0-shot) |64.05|
|Winogrande (5-shot) |81.45|
|GSM8k (5-shot) |68.16|
EQ-Bench (v2_de): 61.04 / english (v2): 78.3
| Model |AGIEval|GPT4All|TruthfulQA|Bigbench|Average|
|------------------------------------------------------------|------:|------:|---------:|-------:|------:|
|[Spaetzle-v8-7b](https://huggingface.co/cstr/Spaetzle-v8-7b)| 45.31| 75.69| 63.94| 45.57| 57.63|
### AGIEval
| Task |Version| Metric |Value| |Stderr|
|------------------------------|------:|--------|----:|---|-----:|
|agieval_aqua_rat | 0|acc |25.59|± | 2.74|
| | |acc_norm|24.80|± | 2.72|
|agieval_logiqa_en | 0|acc |39.63|± | 1.92|
| | |acc_norm|39.78|± | 1.92|
|agieval_lsat_ar | 0|acc |23.48|± | 2.80|
| | |acc_norm|24.35|± | 2.84|
|agieval_lsat_lr | 0|acc |50.98|± | 2.22|
| | |acc_norm|51.96|± | 2.21|
|agieval_lsat_rc | 0|acc |62.08|± | 2.96|
| | |acc_norm|62.83|± | 2.95|
|agieval_sat_en | 0|acc |78.64|± | 2.86|
| | |acc_norm|79.13|± | 2.84|
|agieval_sat_en_without_passage| 0|acc |44.66|± | 3.47|
| | |acc_norm|44.66|± | 3.47|
|agieval_sat_math | 0|acc |37.27|± | 3.27|
| | |acc_norm|35.00|± | 3.22|
Average: 45.31%
### GPT4All
| Task |Version| Metric |Value| |Stderr|
|-------------|------:|--------|----:|---|-----:|
|arc_challenge| 0|acc |63.14|± | 1.41|
| | |acc_norm|64.51|± | 1.40|
|arc_easy | 0|acc |85.98|± | 0.71|
| | |acc_norm|82.49|± | 0.78|
|boolq | 1|acc |88.10|± | 0.57|
|hellaswag | 0|acc |66.31|± | 0.47|
| | |acc_norm|85.17|± | 0.35|
|openbookqa | 0|acc |38.00|± | 2.17|
| | |acc_norm|47.20|± | 2.23|
|piqa | 0|acc |83.35|± | 0.87|
| | |acc_norm|84.17|± | 0.85|
|winogrande | 0|acc |78.22|± | 1.16|
Average: 75.69%
### TruthfulQA
| Task |Version|Metric|Value| |Stderr|
|-------------|------:|------|----:|---|-----:|
|truthfulqa_mc| 1|mc1 |47.74|± | 1.75|
| | |mc2 |63.94|± | 1.53|
Average: 63.94%
### Bigbench
| Task |Version| Metric |Value| |Stderr|
|------------------------------------------------|------:|---------------------|----:|---|-----:|
|bigbench_causal_judgement | 0|multiple_choice_grade|56.84|± | 3.60|
|bigbench_date_understanding | 0|multiple_choice_grade|66.12|± | 2.47|
|bigbench_disambiguation_qa | 0|multiple_choice_grade|41.47|± | 3.07|
|bigbench_geometric_shapes | 0|multiple_choice_grade|22.01|± | 2.19|
| | |exact_str_match | 0.00|± | 0.00|
|bigbench_logical_deduction_five_objects | 0|multiple_choice_grade|31.40|± | 2.08|
|bigbench_logical_deduction_seven_objects | 0|multiple_choice_grade|23.14|± | 1.60|
|bigbench_logical_deduction_three_objects | 0|multiple_choice_grade|56.00|± | 2.87|
|bigbench_movie_recommendation | 0|multiple_choice_grade|45.00|± | 2.23|
|bigbench_navigate | 0|multiple_choice_grade|50.70|± | 1.58|
|bigbench_reasoning_about_colored_objects | 0|multiple_choice_grade|70.05|± | 1.02|
|bigbench_ruin_names | 0|multiple_choice_grade|45.54|± | 2.36|
|bigbench_salient_translation_error_detection | 0|multiple_choice_grade|26.05|± | 1.39|
|bigbench_snarks | 0|multiple_choice_grade|71.82|± | 3.35|
|bigbench_sports_understanding | 0|multiple_choice_grade|72.92|± | 1.42|
|bigbench_temporal_sequences | 0|multiple_choice_grade|44.20|± | 1.57|
|bigbench_tracking_shuffled_objects_five_objects | 0|multiple_choice_grade|22.80|± | 1.19|
|bigbench_tracking_shuffled_objects_seven_objects| 0|multiple_choice_grade|18.23|± | 0.92|
|bigbench_tracking_shuffled_objects_three_objects| 0|multiple_choice_grade|56.00|± | 2.87|
Average: 45.57%
Average score: 57.63%
## 💻 Usage
```python
!pip install -qU transformers accelerate
from transformers import AutoTokenizer
import transformers
import torch
model = "cstr/Spaetzle-v8-7b"
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"])
```
## 🧩 Configuration
The model uses ChatML and should work well with this (as it is merged from models which (mostly) saw ChatML templates in training).
```yaml
models:
- model: mayflowergmbh/Wiedervereinigung-7b-dpo-laser
# no parameters necessary for base model
- model: flemmingmiguel/NeuDist-Ro-7B
parameters:
density: 0.60
weight: 0.30
- model: johannhartmann/Brezn3
parameters:
density: 0.65
weight: 0.40
- model: ResplendentAI/Flora_DPO_7B
parameters:
density: 0.6
weight: 0.3
merge_method: dare_ties
base_model: mayflowergmbh/Wiedervereinigung-7b-dpo-laser
parameters:
int8_mask: true
dtype: bfloat16
random_seed: 0
tokenizer_source: base
``` |