Text Generation
Transformers
Safetensors
English
Korean
aether_v2_7way
foundation-model
sovereign-ai
fully-open
open-source
mixture-of-experts
Mixture of Experts
heterogeneous-attention
latin-square
instruct
sft
fine-tuned
korean
vidraft
aether
conversational
custom_code
Instructions to use FINAL-Bench/Aether-7B-5Attn-it with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use FINAL-Bench/Aether-7B-5Attn-it with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="FINAL-Bench/Aether-7B-5Attn-it", trust_remote_code=True) messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("FINAL-Bench/Aether-7B-5Attn-it", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use FINAL-Bench/Aether-7B-5Attn-it with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "FINAL-Bench/Aether-7B-5Attn-it" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "FINAL-Bench/Aether-7B-5Attn-it", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/FINAL-Bench/Aether-7B-5Attn-it
- SGLang
How to use FINAL-Bench/Aether-7B-5Attn-it with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "FINAL-Bench/Aether-7B-5Attn-it" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "FINAL-Bench/Aether-7B-5Attn-it", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "FINAL-Bench/Aether-7B-5Attn-it" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "FINAL-Bench/Aether-7B-5Attn-it", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use FINAL-Bench/Aether-7B-5Attn-it with Docker Model Runner:
docker model run hf.co/FINAL-Bench/Aether-7B-5Attn-it
preserve Korean card as README_ko.md
Browse files- README_ko.md +79 -0
README_ko.md
ADDED
|
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- ko
|
| 5 |
+
- en
|
| 6 |
+
base_model: FINAL-Bench/Aether-7B-5Attn
|
| 7 |
+
pipeline_tag: text-generation
|
| 8 |
+
library_name: transformers
|
| 9 |
+
tags:
|
| 10 |
+
- aether
|
| 11 |
+
- mixture-of-experts
|
| 12 |
+
- korean
|
| 13 |
+
- vidraft
|
| 14 |
+
- sft
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# Aether-7B-5Attn-it
|
| 18 |
+
|
| 19 |
+

|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
**Aether-7B-5Attn base** ๋ฅผ ์ง๋ํ์ต(SFT)ํ ํ์ฒ๋ฆฌ ๋ฒ์ ์
๋๋ค. ๊ฐ๊ด์/๋ฒค์น๋งํฌ ์๋ต ํ์์ ๋ง์ถฐ ํ๋๋์์ต๋๋ค.
|
| 23 |
+
6.59B MoE (ํ์ฑ ์ฝ 3B), 49์ธต, 7ร7 ๋ผํด ๋ฐฉ์ง ์ดํ
์
๋ฐฐ์น. **Apache-2.0**.
|
| 24 |
+
|
| 25 |
+
## ํ์ต๋ฅ (LR) ์ ํ โ loss๊ฐ ์๋๋ผ held-out ์ ๋ต๋ฅ ๋ก
|
| 26 |
+
|
| 27 |
+
full-parameter SFT๋ฅผ ์ธ ํ์ต๋ฅ (2e-6 / 6e-6 / 2e-5)๋ก ๊ฐ๊ฐ ์ํํ ๋ค,
|
| 28 |
+
**ํ์ต loss๊ฐ ์๋๋ผ held-out ๋ฒค์น๋งํฌ ์ ๋ต๋ฅ **๋ก ์ต์ข
๋ณธ์ ์ ํํ์ต๋๋ค.
|
| 29 |
+
(์์ ๋ชจ๋ธ์์ ๋์ LR์ ํ์ต loss๋ ๋ฎ์ถ์ง๋ง ์ค์ ๋ฅ๋ ฅ์ ์์์ํฌ ์ ์์ด, loss๋ก ๊ณ ๋ฅด๋ฉด ์ ๋ฉ๋๋ค.)
|
| 30 |
+
|
| 31 |
+
| LR | K-AI 4 ํ๊ท | ๋น๊ณ |
|
| 32 |
+
|----|------------|------|
|
| 33 |
+
| 2e-6 | 29.7% | ๊ณผ์์ ํฉ (์ผ๋ถ ๊ณผ๋ชฉ ์ฝํจ) |
|
| 34 |
+
| **6e-6 (์ ํ)** | **34.9%** | ์ต๊ณ ยท๊ท ํ, ์์ ์์ |
|
| 35 |
+
| 2e-5 | 32.3% | ํ ๊ณผ๋ชฉ ๊ธ๋ฝ (ํ์ ์์ ์งํ) |
|
| 36 |
+
|
| 37 |
+
## Held-out ๊ฒฐ๊ณผ (๋ณธ ์ ํ๋ณธ, 6e-6)
|
| 38 |
+
|
| 39 |
+
- **GPQA-Diamond (198)**: 25.3%
|
| 40 |
+
- **K-AI 4 ํ๊ท (195)**: **34.9%**
|
| 41 |
+
- musr_ko 26.5% ยท com2_main_ko 50.0% ยท click 32.0% ยท kommlu_pro 30.4%
|
| 42 |
+
- **base(SFT ์ ) ๋๋น**: base 26.7% โ **34.9% (+8.2pp)** โ ๋์ผ held-outยท๋์ผ ํ๋์ค๋ก ์ธก์
|
| 43 |
+
|
| 44 |
+
*๋ชจ๋ ํ๊ฐ๋ ํ์ต์ ์ฐ์ด์ง ์์ held-out ์
์
๋๋ค. SFT๋ MMLU-auxiliary(๊ฐ๊ด์ ํ์) ์์์ ์ํ๋์๊ณ ,
|
| 45 |
+
ํ๊ฐ ๊ณผ๋ชฉ(GPQAยทK-AI)๊ณผ ๊ฒน์น์ง ์์ต๋๋ค.*
|
| 46 |
+
|
| 47 |
+
## ์ฌ์ ํ์ต ๋ฐ์ดํฐ ๊ตฌ์ฑ (๋ฒ ์ด์ค ์์, ์ ํจ ์ํ๋ง ๋น์ค)
|
| 48 |
+
|
| 49 |
+
| ๋๋ฉ์ธ | ๋น์ค |
|
| 50 |
+
|--------|------|
|
| 51 |
+
| Math (finemath + open-web-math) | 37.8% |
|
| 52 |
+
| Korean (webtext + synth) | 21.6% |
|
| 53 |
+
| English webยทsynthetic (fineweb-edu + cosmopedia) | 21.6% |
|
| 54 |
+
| Code (opc) | 13.5% |
|
| 55 |
+
| phase15 ํ๋ฆฌ๋ธ๋ ๋ | 5.4% |
|
| 56 |
+
|
| 57 |
+
*์ด ๊ตฌ์ฑ์ **๋ฒ ์ด์ค ์ฌ์ ํ์ต** ๋ฐ์ดํฐ์ ๋น์ค์
๋๋ค. ๋ณธ ๋ชจ๋ธ์ ํ์ฒ๋ฆฌ(SFT) ๋ฐ์ดํฐ๋ MMLU-auxiliary(๊ฐ๊ด์)์
๋๋ค.*
|
| 58 |
+
|
| 59 |
+
|
| 60 |
+
## ์ํคํ
์ฒ (๋ฒ ์ด์ค ๊ณ์น)
|
| 61 |
+
|
| 62 |
+
๋ณธ ๋ชจ๋ธ์ Aether-7B-5Attn base ์ ๋์ผํ ์ํคํ
์ฒ์
๋๋ค: 49์ธต์ 7ร7 ๋ผํด ๋ฐฉ์ง์ผ๋ก ๋ฐฐ์นํ ์ด์ข
์ดํ
์
(7๋ผ๋ฒจ / 5๋ฉ์ปค๋์ฆ) + MoE(25์ ๋ฌธ๊ฐ top-7). ๊ตฌ์กฐ ์์ธ๋ [๋ฒ ์ด์ค ๋ชจ๋ธ ์นด๋](https://huggingface.co/FINAL-Bench/Aether-7B-5Attn) ยง3.2 ์ฐธ์กฐ.
|
| 63 |
+
|
| 64 |
+
## ์ฌ์ฉ
|
| 65 |
+
> โ ๏ธ **batch_size=1 ๋ก ์ฌ์ฉํ์ธ์.** ๋ณธ ๋ชจ๋ธ์ ์ผ๋ถ ์ดํ
์
(NSA ๊ณ์ด)์ padding mask ๋ฅผ ์ฌ์ฉํ์ง ์์ผ๋ฏ๋ก, ์ฌ๋ฌ ์ํ์ค๋ฅผ padding ํ์ฌ ๋ฐฐ์น๋ก ๋ฃ์ผ๋ฉด ๊ฒฐ๊ณผ๊ฐ ์กฐ์ฉํ ์ค์ผ๋ ์ ์์ต๋๋ค. ํ ๋ฒ์ ํ๋์ฉ(batch_size=1) ์ถ๋ก ํ์ญ์์ค.
|
| 66 |
+
|
| 67 |
+
|
| 68 |
+
๋ฒ ์ด์ค์ ๋์ผํ๊ฒ `aether_pkg/` ์ ์ปค์คํ
์ํคํ
์ฒ๋ฅผ ์ฌ์ฉํฉ๋๋ค. ๋ก๋ฉ ๋ฐฉ๋ฒ์
|
| 69 |
+
[FINAL-Bench/Aether-7B-5Attn](https://huggingface.co/FINAL-Bench/Aether-7B-5Attn) ์ ๋ฐ๋ฅด์ญ์์ค.
|
| 70 |
+
|
| 71 |
+
|
| 72 |
+
## ์คํ์์ค ์์ ๊ณต๊ฐ LLM ๋น๊ต (6๊ฐ๊ตญ)
|
| 73 |
+
|
| 74 |
+

|
| 75 |
+
|
| 76 |
+
*๋ฒ ์ด์ค Aether ๊ธฐ์ค. ๋น๋๋ํ๋ ์ ์ผํ ํ๊ตญ AI ์คํํธ์
์ด๋ฉฐ 5์ข
(์ต๋ค) ์ดํ
์
ยท๋ผํด๋ฐฉ์ง ๊ตฌ์กฐ๊ฐ ์ฐจ๋ณ์ ์
๋๋ค.*
|
| 77 |
+
|
| 78 |
+
## ๋ฌธ์
|
| 79 |
+
์ฃผ์ํ์ฌ ๋น๋๋ํํธ (VIDRAFT) ยท arxivgpt@gmail.com
|