metadata
model-index:
- name: EEVE-Math-10.8B
results:
- task:
type: text-generation
dataset:
name: gsm8k-ko
type: gsm8k
metrics:
- name: pass@1
type: pass@1
value: 0.539
verified: false
base_model: yanolja/EEVE-Korean-10.8B-v1.0
license: cc-by-sa-4.0
language:
- ko
tags:
- math
datasets:
- kuotient/orca-math-word-problems-193k-korean
EEVE-Math-10.8B
EEVE-Math
νλ‘μ νΈλ
- Orca-Math-200k λ²μ (Orca-Math: Unlocking the potential of SLMs in Grade School Math)
- gsm8k λ²μ, lm_eval νμ©
- Mergekitμ μ΄μ©ν dare-ties μ¬μ© (DARE)
μ λν λ΄μ©μ ν¬κ΄νκ³ μμ΅λλ€.
μ΄ λͺ¨λΈμ orca-math-word-problems-193k-korean λ°μ΄ν°μ μ μ΄μ©νμ¬ νμ΅λμμ΅λλ€. μλ΅ μ€ μΌλΆλ LaTeX νμμ μ΄μ©νμ¬ κ²°κ³Όλ₯Ό λ°ννμ§λ§, μμ±λ νμμ΄ μλ μ μμ΅λλ€. νμ¬ M1 stageκΉμ§ μ§νλμμ΅λλ€.
Model | gsm8k-ko(pass@1) |
---|---|
Base | 0.4049 |
SFT(M1) | 0.508 |
SFT(M1) -> SFT | 0.539 |
SFT(M1) -> KTO(M2) | - |
SFT -> KTO(M2) -> KTO(final) | - |
Specifications
- SFT(M1) -> SFT λ¨κ³
Base Model
yanolja/EEVE-Korean-10.8B-v1.0
Dataset
orca-math-word-problems-193k-korean
Evaluation
gsm8k-ko, kobest
git clone https://github.com/kuotient/lm-evaluation-harness.git
cd lm-evaluation-harness
pip install -e .
lm_eval --model hf \
--model_args pretrained=yanolja/EEVE-Korean-Instruct-2.8B-v1.0 \
--tasks gsm8k-ko \
--device cuda:0 \
--batch_size auto:4
Model | gsm8k(pass@1) | boolq(acc) | copa(acc) | hellaswag(acc) | Overall |
---|---|---|---|---|---|
yanolja/EEVE-Korean-10.8B-v1.0 | 0.4049 | - | - | - | - |
yanolja/EEVE-Korean-Instruct-10.8B-v1.0 | 0.4511 | 0.8668 | 0.7450 | 0.4940 | 0.6392 |
EEVE-Math-10.8B | 0.5390 | 0.8027 | 0.7260 | 0.4760 | 0.6359 |
EEVE-Instruct-Math-10.8B | 0.4845 | 0.8519 | 0.7410 | 0.4980 | 0.6439 |