metadata

tags:
  - merge
  - mergekit
  - cstr/Spaetzle-v80-7b
  - cstr/Spaetzle-v79-7b
  - cstr/Spaetzle-v81-7b
  - cstr/Spaetzle-v71-7b
base_model:
  - cstr/Spaetzle-v80-7b
  - cstr/Spaetzle-v79-7b
  - cstr/Spaetzle-v81-7b
  - cstr/Spaetzle-v71-7b
license: cc-by-nc-4.0
language:
  - de
  - en

Spaetzle-v85-7b

Spaetzle-v85-7b is a merge of the following models using LazyMergekit:

Evaluation

EQ-Bench (v2_de): 65.32, Parseable: 171.0

Model	AGIEval	GPT4All	TruthfulQA	Bigbench	Average
Spaetzle-v85-7b	44.35	75.99	67.23	46.55	58.53

From Intel/low_bit_open_llm_leaderboard:

Metric	Value
ARC-c	62.63
ARC-e	85.56
Boolq	87.77
HellaSwag	66.66
Lambada	70.35
MMLU	61.61
Openbookqa	37.2
Piqa	82.48
Truthfulqa	50.43
Winogrande	78.3
Average	68.3

AGIEval

Task	Version	Metric	Value		Stderr
agieval_aqua_rat	0	acc	23.23	±	2.65
		acc_norm	22.44	±	2.62
agieval_logiqa_en	0	acc	37.33	±	1.90
		acc_norm	37.94	±	1.90
agieval_lsat_ar	0	acc	25.22	±	2.87
		acc_norm	23.04	±	2.78
agieval_lsat_lr	0	acc	49.41	±	2.22
		acc_norm	50.78	±	2.22
agieval_lsat_rc	0	acc	64.68	±	2.92
		acc_norm	63.20	±	2.95
agieval_sat_en	0	acc	77.67	±	2.91
		acc_norm	78.16	±	2.89
agieval_sat_en_without_passage	0	acc	46.12	±	3.48
		acc_norm	45.15	±	3.48
agieval_sat_math	0	acc	35.45	±	3.23
		acc_norm	34.09	±	3.20

Average: 44.35%

GPT4All

Task	Version	Metric	Value		Stderr
arc_challenge	0	acc	63.82	±	1.40
		acc_norm	64.76	±	1.40
arc_easy	0	acc	85.90	±	0.71
		acc_norm	82.32	±	0.78
boolq	1	acc	87.61	±	0.58
hellaswag	0	acc	67.39	±	0.47
		acc_norm	85.36	±	0.35
openbookqa	0	acc	38.80	±	2.18
		acc_norm	48.80	±	2.24
piqa	0	acc	83.03	±	0.88
		acc_norm	84.17	±	0.85
winogrande	0	acc	78.93	±	1.15

Average: 75.99%

TruthfulQA

Task	Version	Metric	Value		Stderr
truthfulqa_mc	1	mc1	50.80	±	1.75
		mc2	67.23	±	1.49

Average: 67.23%

Bigbench

Task	Version	Metric	Value		Stderr
bigbench_causal_judgement	0	multiple_choice_grade	54.74	±	3.62
bigbench_date_understanding	0	multiple_choice_grade	68.29	±	2.43
bigbench_disambiguation_qa	0	multiple_choice_grade	39.53	±	3.05
bigbench_geometric_shapes	0	multiple_choice_grade	22.28	±	2.20
		exact_str_match	12.26	±	1.73
bigbench_logical_deduction_five_objects	0	multiple_choice_grade	32.80	±	2.10
bigbench_logical_deduction_seven_objects	0	multiple_choice_grade	23.00	±	1.59
bigbench_logical_deduction_three_objects	0	multiple_choice_grade	59.00	±	2.84
bigbench_movie_recommendation	0	multiple_choice_grade	45.60	±	2.23
bigbench_navigate	0	multiple_choice_grade	51.10	±	1.58
bigbench_reasoning_about_colored_objects	0	multiple_choice_grade	70.10	±	1.02
bigbench_ruin_names	0	multiple_choice_grade	52.68	±	2.36
bigbench_salient_translation_error_detection	0	multiple_choice_grade	33.57	±	1.50
bigbench_snarks	0	multiple_choice_grade	71.27	±	3.37
bigbench_sports_understanding	0	multiple_choice_grade	74.54	±	1.39
bigbench_temporal_sequences	0	multiple_choice_grade	40.00	±	1.55
bigbench_tracking_shuffled_objects_five_objects	0	multiple_choice_grade	21.52	±	1.16
bigbench_tracking_shuffled_objects_seven_objects	0	multiple_choice_grade	18.86	±	0.94
bigbench_tracking_shuffled_objects_three_objects	0	multiple_choice_grade	59.00	±	2.84

Average: 46.55%

Average score: 58.53%

🧩 Configuration

models:
  - model: cstr/Spaetzle-v84-7b
    # no parameters necessary for base model
  - model: cstr/Spaetzle-v80-7b
    parameters:
      density: 0.65
      weight: 0.2
  - model: cstr/Spaetzle-v79-7b
    parameters:
      density: 0.65
      weight: 0.2
  - model: cstr/Spaetzle-v81-7b
    parameters:
      density: 0.65
      weight: 0.2
  - model: cstr/Spaetzle-v71-7b
    parameters:
      density: 0.65
      weight: 0.2
merge_method: dare_ties
base_model: cstr/Spaetzle-v84-7b
parameters:
  int8_mask: true
dtype: bfloat16
random_seed: 0
tokenizer_source: base

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "cstr/Spaetzle-v85-7b"
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"])