language:
- en
license: cc-by-nc-4.0
model-index:
- name: Iambe-20b-DARE-v2
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: 62.8
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=athirdpath/Iambe-20b-DARE-v2
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: 84.53
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=athirdpath/Iambe-20b-DARE-v2
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: 60.45
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=athirdpath/Iambe-20b-DARE-v2
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: 53.85
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=athirdpath/Iambe-20b-DARE-v2
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: 77.03
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=athirdpath/Iambe-20b-DARE-v2
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: 33.28
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=athirdpath/Iambe-20b-DARE-v2
name: Open LLM Leaderboard
Strange quirk: This model seems to need a context size of EXACTLY 4096 ONLY. I'm assuming this is a dares_ties effect?
Iambe-20b-DARE-v2
Alpaca prompt formatting
Description
Named after a charming daughter of Echo and Pan in Greek myth, Iambe-20b-DARE-v2 is an improved DARE merge building on my recent experiments.
Iambe is intended to have the best realistically possible understanding of anatomy and of a scene's state for a 20b merge, while remaining personable and authentic in "voice".
Update Methodology
Noromaid and the general "no-robots" vibe didn't come through like I'd hoped in v1. My hypothesis is that the "soul" MythoMax and Noromaid have is probably distributed widely over many low-value deltas, due to the "ephemeral" nature of such a thing.
My old base model was likely giving DARE conniption fits, so I replaced that with a truly vanilla 20b base model.
CleverGirl was updated to the DARE version, as Sir Hillary said, simply because it was there.
Without a large base of dare_ties models to compare to, I'm basically feeling my way through this intuitively, so here's to good results!
Recipe
merge_method: dare_ties
base_model: athirdpath/BigLlama-20b-v1.1
model: Noromaid-20b-v0.1.1
weight: 0.38 / density: 0.60
model: athirdpath/athirdpath/Eileithyia-20b
weight: 0.22 / density: 0.40
model: athirdpath/CleverGirl-20b-Blended-v1.1-DARE
weight: 0.40 / density: 0.33
int8_mask: true
dtype: bfloat16
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 61.99 |
| AI2 Reasoning Challenge (25-Shot) | 62.80 |
| HellaSwag (10-Shot) | 84.53 |
| MMLU (5-Shot) | 60.45 |
| TruthfulQA (0-shot) | 53.85 |
| Winogrande (5-shot) | 77.03 |
| GSM8k (5-shot) | 33.28 |