license: llama2
model-index:
- name: LMCocktail-10.7B-v1
results:
- task:
type: text-generation
name: Text Generation
dataset:
name: AI2 Reasoning Challenge (25-Shot)
type: ai2_arc
config: ARC-Challenge
split: test
args:
num_few_shot: 25
metrics:
- type: acc_norm
value: 70.65
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yhyu13/LMCocktail-10.7B-v1
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: HellaSwag (10-Shot)
type: hellaswag
split: validation
args:
num_few_shot: 10
metrics:
- type: acc_norm
value: 88.13
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yhyu13/LMCocktail-10.7B-v1
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: MMLU (5-Shot)
type: cais/mmlu
config: all
split: test
args:
num_few_shot: 5
metrics:
- type: acc
value: 66.21
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yhyu13/LMCocktail-10.7B-v1
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: TruthfulQA (0-shot)
type: truthful_qa
config: multiple_choice
split: validation
args:
num_few_shot: 0
metrics:
- type: mc2
value: 71.03
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yhyu13/LMCocktail-10.7B-v1
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: Winogrande (5-shot)
type: winogrande
config: winogrande_xl
split: validation
args:
num_few_shot: 5
metrics:
- type: acc
value: 83.35
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yhyu13/LMCocktail-10.7B-v1
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: GSM8k (5-shot)
type: gsm8k
config: main
split: test
args:
num_few_shot: 5
metrics:
- type: acc
value: 64.97
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yhyu13/LMCocktail-10.7B-v1
name: Open LLM Leaderboard
LM-cocktail 10.7B v1
This is a 50%-50% model of the SOLAR model and meow.
https://huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0
https://huggingface.co/rishiraj/meow
who rank #1 and #2 among models <13B in the https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard by 2023/12/20.
Alpaca Eval
I am thrilled to announce that ChatGPT has ranked LMCocktail 10.7B as the second best model next to GPT4 on AlpcaEval in my local community run. You can also check the leaderboard at ./alpaca_eval/chatgpt_fn_--SOLAR-10-7B-LMCocktail/
win_rate standard_error n_total avg_length
gpt4 73.79 1.54 805 1365
SOLAR-10.7B-LMCocktail(new)73.45 1.56 804 1203
claude 70.37 1.60 805 1082
chatgpt 66.09 1.66 805 811
wizardlm-13b 65.16 1.67 805 985
vicuna-13b 64.10 1.69 805 1037
guanaco-65b 62.36 1.71 805 1249
oasst-rlhf-llama-33b 62.05 1.71 805 1079
alpaca-farm-ppo-human 60.25 1.72 805 803
falcon-40b-instruct 56.52 1.74 805 662
text_davinci_003 50.00 0.00 805 307
alpaca-7b 45.22 1.74 805 396
text_davinci_001 28.07 1.56 805 296
Code
The LM-cocktail is novel technique for merging multiple models https://arxiv.org/abs/2311.13534
Code is backed up by this repo https://github.com/FlagOpen/FlagEmbedding.git
Merging scripts available under the ./scripts folder
Result
The SOLAR model is the first model <30B that can answer this question from my test:
What will AI be like in the year 1010 A.D?
without hullicinating into 1010 A.D is a future time (like other llama2 models)
Models greater than that, like Yi-34B could answer this paradoxic question correctly as well, since it is huge enough.
SOLAR 10.7B output
LMCocktail 10.7B output1
LMCocktail 10.7B output2
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 74.06 |
| AI2 Reasoning Challenge (25-Shot) | 70.65 |
| HellaSwag (10-Shot) | 88.13 |
| MMLU (5-Shot) | 66.21 |
| TruthfulQA (0-shot) | 71.03 |
| Winogrande (5-shot) | 83.35 |
| GSM8k (5-shot) | 64.97 |