Instructions to use sori-forest/aether-office with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use sori-forest/aether-office with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf sori-forest/aether-office:Q3_K_M # Run inference directly in the terminal: llama cli -hf sori-forest/aether-office:Q3_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf sori-forest/aether-office:Q3_K_M # Run inference directly in the terminal: llama cli -hf sori-forest/aether-office:Q3_K_M
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf sori-forest/aether-office:Q3_K_M # Run inference directly in the terminal: ./build/bin/llama cli -hf sori-forest/aether-office:Q3_K_M
Use Docker
docker model run hf.co/sori-forest/aether-office:Q3_K_M
- Llama new
- LM Studio
- Jan
- vLLM
How to use sori-forest/aether-office with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "sori-forest/aether-office" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "sori-forest/aether-office", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/sori-forest/aether-office:Q3_K_M
- Ollama
How to use sori-forest/aether-office with Ollama:
ollama run hf.co/sori-forest/aether-office:Q3_K_M
- Unsloth Desktop
- Pi
How to use sori-forest/aether-office with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf sori-forest/aether-office:Q3_K_M
Configure the model in Pi
# Install Pi: npm install -g @earendil-works/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:8080/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "sori-forest/aether-office:Q3_K_M" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use sori-forest/aether-office with Docker Model Runner:
docker model run hf.co/sori-forest/aether-office:Q3_K_M
- Lemonade
How to use sori-forest/aether-office with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull sori-forest/aether-office:Q3_K_M
Run and chat with the model
lemonade run user.aether-office-Q3_K_M
List all available models
lemonade list
- Hermes Agent
How to use sori-forest/aether-office with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf sori-forest/aether-office:Q3_K_M
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:8080/v1 hermes config set model.default sori-forest/aether-office:Q3_K_M
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use sori-forest/aether-office with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf sori-forest/aether-office:Q3_K_M
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:8080/v1 \ --custom-model-id "sori-forest/aether-office:Q3_K_M" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
Aether Office v1.0
An offline office-document assistant for Korean, English, and Japanese. No internet, no cloud โ your documents never leave your computer.
English ยท ํ๊ตญ์ด ยท ๆฅๆฌ่ช
| File | Size | For |
|---|---|---|
aether-office-v1.0-Q4_K_M.gguf |
2.78 GB | Recommended. Sized with 8 GB RAM devices in mind (not yet tested on such devices) |
aether-office-v1.0-Q3_K_M.gguf |
2.32 GB | Smaller file. Not evaluated โ all results below are for Q4_K_M |
English
What it does
- Summarize documents (keeps requested line counts and length)
- Write and rewrite business emails and memos in Korean, English, or Japanese business style
- Extract fields from invoices, contracts, and orders into JSON only
- Answer only from the document you give, and say plainly when the answer is not in it
- Translate business text between Korean, English, and Japanese
- Simple office formulas (Excel, regex, short scripts)
File size is 2.78 GB, chosen with 8 GB RAM devices in mind. It has not yet been tested on an actual 8 GB device.
Quick start (LM Studio)
- Download
aether-office-v1.0-Q4_K_M.ggufand import it into LM Studio (My Models โ Import). - Load the model and turn Thinking off in the model settings. Aether Office was trained to answer directly.
- Paste the contents of the
systemfile into the System Prompt box. Recommended: temperature 0.3, top_p 0.9.
Using the local API on Windows? Use http://127.0.0.1:1234 instead of localhost. On Windows, localhost can add about 2 seconds to every request.
Evaluation
Fixed 30-question set (ko/en/ja ร summary, email, rewrite, JSON extraction, grounded answers, "not in the document" answers, translation). Automatic checks for format, required facts, language and refusals. Thinking off, same settings for both models, each model run twice.
| Previous internal build | Aether Office v1.0 | |
|---|---|---|
| 30-question set, run 1 (all checks passed) | 29 / 30 | 29 / 30 |
| 30-question set, run 2 | 28 / 30 | 29 / 30 |
| Real-use test: Japanese 2-line summary of a holiday notice โ keeps the "orders after Dec 25, 15:00 ship from Jan 5" date line | 3 / 8 answers | 5 / 5 answers |
- Aether Office v1.0 results were measured with the
systemprompt included in this repository. - For reference, the base Qwen3.5-4B scored 17/30 on the same set (measured with the previous internal build, not re-measured here).
- Results move by 1โ2 questions from run to run, so the difference on the 30-question set is within normal variation. The clearer change is the Japanese summary date line.
- The automatic checks do not fully measure writing quality.
Known limitations
- Run-to-run variation: the same question can pass in one run and fail in the next (1โ2 of 30 questions per run).
- Human review was limited: apart from the automatic checks, the answers a person read closely were the Japanese summary items only.
- Summaries can occasionally drop or alter a number or a comparison (e.g. a total amount, "vs. last month"). Check figures that matter.
- AM/PM can flip: a time can change between morning and afternoon (e.g. "3 p.m." became "3 a.m." when rewriting a memo as a notice). Always check dates and times.
- Word swaps: a word can be replaced by a similar but different word, changing the meaning (e.g. Korean ์ ๊ณ "goods received" โ ์ ๊ธ "payment received"; "ship date" โ "delivery date"). The base model (Qwen3.5-4B) showed the same behavior, and this training did not fully fix it. Review drafts before sending.
- Emails may add small details you did not ask for (placeholder company names).
- Japanese honorifics: high-stakes keigo can be off. Have a person check before sending.
- Code: only short office formulas and scripts. Not a general coding assistant.
- Text only. Images are not supported in this GGUF. Only Korean, English, and Japanese were trained.
How it was made
- Base: Qwen3.5-4B (Apache-2.0). LoRA (r=16), 2 epochs, 1,400 examples (English 473, Japanese 472, Korean 455).
- The examples are the previous internal build's training set minus 40 summaries that dropped dates or amounts (summary data clean-up).
- Internal build name: v2.1.1.
- Training examples were drafted locally by Gemma 4 E4B (Apache-2.0), filtered by automatic checks (language, number formatting, invented numbers, JSON validity, unsafe code) and reviewed by people with AI assistance.
- Evaluation questions were kept out of the training data.
ํ๊ตญ์ด
๋ฌด์์ ํ๋์
- ๋ฌธ์ ์์ฝ (์ค ์ยท๊ธธ์ด ์ง์๋ฅผ ์งํต๋๋ค)
- ์ ๋ฌด ์ด๋ฉ์ผยท๊ณต๋ฌธ ์์ฑ๊ณผ ๋ค๋ฌ๊ธฐ (ํ๊ตญ์ดยท์์ดยท์ผ๋ณธ์ด ๋น์ฆ๋์ค ๋งํฌ)
- ์ฒญ๊ตฌ์ยท๊ณ์ฝ์ยท์ฃผ๋ฌธ์์์ ์ ๋ณด๋ฅผ ๋ฝ์ JSON๋ง ์ถ๋ ฅ
- ์ ๋ ฅํ ๋ฌธ์ ์์์๋ง ๋ตํ๊ธฐ, ๋ฌธ์์ ์์ผ๋ฉด ์๋ค๊ณ ๋งํ๊ธฐ
- ํยท์ยท์ผ ์ ๋ฌด ๋ฒ์ญ
- ๊ฐ๋จํ ์ฌ๋ฌด ์๋ํ (์์ ์์, ์ ๊ท์, ์งง์ ์คํฌ๋ฆฝํธ)
์ธํฐ๋ท ์์ด ๋ด ์ปดํจํฐ์์๋ง ๋์๊ฐ๋ฏ๋ก, ํ์ฌ ๋ฌธ์๋ฅผ ๋ฐ์ผ๋ก ๋ด๋ณด๋ด์ง ์์๋ ๋ฉ๋๋ค. ๊ถ์ฅ ํ์ผ(Q4_K_M)์ 2.78GB๋ก, ๋ฉ๋ชจ๋ฆฌ 8GB ๊ธฐ๊ธฐ๋ฅผ ์ผ๋์ ๋ ํฌ๊ธฐ์ ๋๋ค (์ค์ 8GB ๊ธฐ๊ธฐ์์๋ ์์ง ์ํํ์ง ์์์ต๋๋ค). ๋ ์์ Q3_K_M(2.32GB)์ ํ๊ฐํ์ง ์์์ต๋๋ค.
์์ํ๊ธฐ (LM Studio)
aether-office-v1.0-Q4_K_M.gguf๋ฅผ ๋ฐ์ LM Studio์ ๋ถ๋ฌ์ต๋๋ค (My Models โ Import).- ๋ชจ๋ธ์ ๋ถ๋ฌ์จ ๋ค ์ค์ ์์ Thinking(์๊ฐ)์ ๋๋๋ค. ๋ฐ๋ก ๋ตํ๋๋ก ํ์ต๋์์ต๋๋ค.
systemํ์ผ ๋ด์ฉ์ System Prompt ์นธ์ ๋ถ์ฌ ๋ฃ์ต๋๋ค. ๊ถ์ฅ: temperature 0.3, top_p 0.9.
์๋์ฐ์์ ๋ก์ปฌ API๋ฅผ ์ด๋ค๋ฉด localhost ๋์ http://127.0.0.1:1234 ๋ฅผ ์ฐ์ธ์. ์๋์ฐ์์๋ localhost ๊ฐ ์์ฒญ๋ง๋ค ์ฝ 2์ด๋ฅผ ๋ ์ธ ์ ์์ต๋๋ค.
ํ๊ฐ
๊ณ ์ ๋ 30๋ฌธํญ(ํยท์ยท์ผ ร ์์ฝ, ์ด๋ฉ์ผ, ์ฌ์์ฑ, JSON ์ถ์ถ, ๊ทผ๊ฑฐ ๋ต๋ณ, "๋ฌธ์์ ์์" ๋ต๋ณ, ๋ฒ์ญ)์ ํ์ยทํต์ฌ ์ฌ์คยท์ธ์ดยท๊ฑฐ์ ์ฌ๋ถ๋ก ์๋ ์ฑ์ ํ์ต๋๋ค. Thinking ๋, ๋ ๋ชจ๋ธ ๊ฐ์ ์กฐ๊ฑด, ๊ฐ๊ฐ 2ํ.
| ์ด์ ๋ด๋ถ ๋ฒ์ | Aether Office v1.0 | |
|---|---|---|
| 30๋ฌธํญ 1ํ์ฐจ (์ ํญ๋ชฉ ํต๊ณผ) | 29 / 30 | 29 / 30 |
| 30๋ฌธํญ 2ํ์ฐจ | 28 / 30 | 29 / 30 |
| ์ค์ฌ์ฉ ์ํ: ์ผ๋ณธ์ด ์ฐ๋ง์ฐ์ ํด์ ์๋ด 2์ค ์์ฝ์์ "12์ 25์ผ 15์ ์ดํ ์ฃผ๋ฌธ์ 1์ 5์ผ ์ดํ ๋ฐ์ก" ๋ ์ง ์ค ์ ์ง | 8๋ฒ ์ค 3๋ฒ | 5๋ฒ ์ค 5๋ฒ |
- Aether Office v1.0 ๊ฒฐ๊ณผ๋ ์ด ์ ์ฅ์์ ํจ๊ป ์๋
systemํ๋กฌํํธ๋ก ์ฐ ๊ฒ์ ๋๋ค. - ์ฐธ๊ณ ๋ก ํ์ต ์ Qwen3.5-4B๋ ๊ฐ์ ๋ฌธํญ์์ 17/30์ด์์ต๋๋ค (์ด์ ๋ด๋ถ ๋ฒ์ ๋ ์ธก์ , ์ด๋ฒ์ ๋ค์ ์ฌ์ง ์์).
- ํ์ฐจ๋ง๋ค 1~2๋ฌธํญ์ฉ ๊ฒฐ๊ณผ๊ฐ ํ๋ค๋ฆฌ๋ฏ๋ก, 30๋ฌธํญ ์ ์ ์ฐจ์ด๋ ํ๋ค๋ฆผ ๋ฒ์ ์์ ๋๋ค. ๋ ๋ถ๋ช ํ ๋ณํ๋ ์ผ๋ณธ์ด ์์ฝ์ ๋ ์ง ์ค์ ๋๋ค.
- ์๋ ์ฑ์ ์ ๋ฌธ์ฅ์ ํ์ง๊น์ง ์์ ํ ์ฌ์ง๋ ๋ชปํฉ๋๋ค.
ํ๊ณ
- ํ์ฐจ๋ง๋ค ํ๋ค๋ฆผ: ๊ฐ์ ๋ฌธํญ์ด ํ ๋ฒ์ ํต๊ณผํ๊ณ ๋ค์์๋ ์คํจํ ์ ์์ต๋๋ค (ํ์ฐจ๋ง๋ค 30๋ฌธํญ ์ค 1~2๊ฐ).
- ์ฌ๋์ ํ์ธ ๋ฒ์: ์๋ ์ฑ์ ์ธ์ ์ฌ๋์ด ๋ต์ ์ง์ ์ฝ์ด ํ์ธํ ๊ฒ์ ์ผ๋ณธ์ด ์์ฝ ๋ฌธํญ๋ฟ์ ๋๋ค.
- ์์ฝ์์ ์ซ์๋ ๋น๊ต ๊ธฐ์ค(ํฉ๊ณ ๊ธ์ก, "์ ์ ๋๋น" ๋ฑ)์ ๋น ๋จ๋ฆฌ๊ฑฐ๋ ๋ฐ๊ฟ ์ ์์ต๋๋ค. ์ค์ํ ์ซ์๋ ํ์ธํ์ธ์.
- ์ค์ /์คํ ๋ฐ๋: ์๊ฐ์ ์ค์ /์คํ๊ฐ ๋ฐ๋ ์ ์์ต๋๋ค (์: ๋ฉ๋ชจ๋ฅผ ๊ณต์ง๋ฌธ์ผ๋ก ๋ฐ๊ฟ ๋ "์คํ 3์" โ "์ค์ 3์"). ๋ ์งยท์๊ฐ์ ๋ฐ๋์ ํ์ธํ์ธ์.
- ๋จ์ด ๋ฐ๋: ๋จ์ด๋ฅผ ๋น์ทํ ๋ค๋ฅธ ๋จ์ด๋ก ๋ฐ๊ฟ ๋ป์ด ๋ฌ๋ผ์ง ์ ์์ต๋๋ค (์: ์ ๊ณ โ์ ๊ธ, ๋ฐ์ก์ผโ๋์ฐฉ์ผ). ๊ธฐ๋ณธ ๋ชจ๋ธ(Qwen3.5-4B)์์๋ ๊ฐ์ ํ์์ด ์์๊ณ , ์ด๋ฒ ํ์ต์ผ๋ก ์์ ํ ๊ณ ์ณ์ง์ง ์์์ต๋๋ค. ๋ณด๋ด๊ธฐ ์ ์ ํ์ธํ์ธ์.
- ์ด๋ฉ์ผ์ ์์ฒญํ์ง ์์ ์ธ๋ถ ๋ด์ฉ(๊ฐ์์ ํ์ฌ ์ด๋ฆ)์ ๋ฃ์ ๋๊ฐ ์์ต๋๋ค.
- ์ผ๋ณธ์ด ๊ณ ๊ธ ๊ฒฝ์ด๋ ์ด์ํ ์ ์์ผ๋ฏ๋ก, ๋ณด๋ด๊ธฐ ์ ์ ์ฌ๋์ด ํ์ธํ์ธ์.
- ์ฝ๋๋ ์งง์ ์ฌ๋ฌด์ฉ ์์ยท์คํฌ๋ฆฝํธ ์์ค์ด๋ฉฐ, ์ผ๋ฐ ์ฝ๋ฉ์ฉ ๋ชจ๋ธ์ด ์๋๋๋ค.
- ๊ธ์ ์ ์ฉ์ ๋๋ค. ์ด GGUF๋ ์ด๋ฏธ์ง๋ฅผ ์ง์ํ์ง ์์ต๋๋ค. ํ๊ตญ์ดยท์์ดยท์ผ๋ณธ์ด๋ง ํ์ตํ์ต๋๋ค.
๋ง๋ ๋ฐฉ๋ฒ
- ๋ฐํ: Qwen3.5-4B (Apache-2.0). LoRA (r=16), 2ํ ํ์ต, ์์ 1,400๊ฐ (์์ด 473, ์ผ๋ณธ์ด 472, ํ๊ตญ์ด 455).
- ์์๋ ์ด์ ๋ด๋ถ ๋ฒ์ ์ ํ์ต ๋ฐ์ดํฐ์์ ๋ ์งยท๊ธ์ก์ ๋น ๋จ๋ฆฐ ์์ฝ 40๊ฐ๋ฅผ ๋บ ๊ฒ์ ๋๋ค (์์ฝ ๋ฐ์ดํฐ ํ์ง ์ ๋ฆฌ).
- ๋ด๋ถ ๊ฐ๋ฐ ์ด๋ฆ v2.1.1.
- ํ์ต ์์๋ Gemma 4 E4B(Apache-2.0)๋ก ๋ก์ปฌ์์ ์ด์์ ๋ง๋ค๊ณ , ์๋ ๊ฒ์ฌ(์ธ์ด, ์ซ์ ํ๊ธฐ, ์ง์ด๋ธ ์ซ์, JSON ํ์, ์ํํ ์ฝ๋)์ AI ๋์์ ๋ฐ์ ์ฌ๋ ๊ฒํ ๋ฅผ ๊ฑฐ์ณค์ต๋๋ค.
- ํ๊ฐ ๋ฌธํญ์ ํ์ต ๋ฐ์ดํฐ์ ๋ฃ์ง ์์์ต๋๋ค.
ๆฅๆฌ่ช
ใงใใใใจ
- ๆๆธใฎ่ฆ็ด(่กๆฐใป้ทใใฎๆๅฎใๅฎใใพใ)
- ใใธใในใกใผใซใป็คพๅ ๆๆธใฎไฝๆใจๆธใ็ดใ(ๆฅใป่ฑใป้ใฎใใธใในๆไฝ)
- ่ซๆฑๆธใปๅฅ็ดๆธใปๆณจๆๆธใใ้ ็ฎใๆใๅบใใJSONใฎใฟใงๅบๅ
- ๅ ฅๅใใๆๆธใฎๅ ๅฎนใ ใใงๅ็ญใใ่จ่ผใใชใใใฐใ่จ่ผใฏใใใพใใใใจ็ญใใ
- ๆฅใป่ฑใป้ใฎใใธใใน็ฟป่จณ
- ็ฐกๅใชไบๅ่ชๅๅ(Excel้ขๆฐใๆญฃ่ฆ่กจ็พใ็ญใในใฏใชใใ)
ใคใณใฟใผใใใไธ่ฆใ็คพๅ ๆๆธใๅค้จใซ้ใใใซไฝฟใใพใใ ๆจๅฅจใใกใคใซ(Q4_K_M)ใฏ 2.78GB ใงใใกใขใช 8GB ใฎๆฉๅจใๆณๅฎใใๅคงใใใงใ(ๅฎ้ใฎ 8GB ๆฉๅจใงใฏใพใ ่ฉฆใใฆใใพใใ)ใใใๅฐใใ Q3_K_M(2.32GB)ใฏ่ฉไพกใใฆใใพใใใ
ไฝฟใๆน(LM Studio)
aether-office-v1.0-Q4_K_M.ggufใใใฆใณใญใผใใใLM Studio ใซ่ชญใฟ่พผใฟใพใ(My Models โ Import)ใ- ใขใใซ่จญๅฎใง Thinking(ๆ่)ใใชใใซใใพใใใใใซ็ญใใใใๅญฆ็ฟใใใฆใใพใใ
systemใใกใคใซใฎๅ ๅฎนใ System Prompt ๆฌใซ่ฒผใไปใใพใใๆจๅฅจ:temperature 0.3ใtop_p 0.9ใ
Windows ใงใญใผใซใซ API ใไฝฟใๅ ดๅใฏใlocalhost ใงใฏใชใ http://127.0.0.1:1234 ใไฝฟใฃใฆใใ ใใใlocalhost ใงใฏใชใฏใจในใใใจใซ็ด2็งไฝๅใซใใใใใจใใใใพใใ
่ฉไพก
ๅบๅฎใฎ30ๅ(ๆฅใป่ฑใป้ ร ่ฆ็ดใใกใผใซใๆธใ็ดใใJSONๆฝๅบใๆๆธใซๅบใฅใๅ็ญใใ่จ่ผใชใใๅ็ญใ็ฟป่จณ)ใใๅฝขๅผใปๅฟ ่ฆใชไบๅฎใป่จ่ชใปๅ็ญๆๅฆใฎๆ็กใง่ชๅๆก็นใใพใใใThinking ใชใใๅไธๆกไปถใๅใขใใซ2ๅใ
| ไปฅๅใฎๅ ้จ็ | Aether Office v1.0 | |
|---|---|---|
| 30ๅใป1ๅ็ฎ(ๅ จ้ ็ฎๅๆ ผ) | 29 / 30 | 29 / 30 |
| 30ๅใป2ๅ็ฎ | 28 / 30 | 29 / 30 |
| ๅฎ็จใในใ:ๅนดๆซๅนดๅงไผๆฅญใฎใ็ฅใใใ2่กใง่ฆ็ดใใใ12ๆ25ๆฅ15ๆไปฅ้ใฎใๆณจๆใฏ1ๆ5ๆฅไปฅ้ใฎ็บ้ใใฎๆฅไปใๆฎใใใ | 8ๅไธญ3ๅ | 5ๅไธญ5ๅ |
- Aether Office v1.0 ใฎ็ตๆใฏใใใฎใชใใธใใชใซๅซใพใใ
systemใใญใณใใใงๆธฌๅฎใใใใฎใงใใ - ๅ่:ๅญฆ็ฟๅใฎ Qwen3.5-4B ใฏๅใ30ๅใง 17/30 ใงใใ(ไปฅๅใฎๅ ้จ็ใฎๆใซๆธฌๅฎใไปๅใฏๅๆธฌๅฎใใฆใใพใใ)ใ
- ๅใใจใซ1ใ2ๅใฎๅทฎใๅบใใใใ30ๅใฎ็นๆฐๅทฎใฏ้ๅธธใฎใฐใใคใใฎ็ฏๅฒๅ ใงใใใใใฏใฃใใใใๅคๅใฏใๆฅๆฌ่ช่ฆ็ดใฎๆฅไปใฎ่กใงใใ
- ่ชๅๆก็นใงใฏๆ็ซ ใฎ่ณชใพใงใฏๅๅใซๆธฌใใพใใใ
ๅถ้ไบ้
- ๅใใจใฎใฐใใคใ: ๅใๅ้กใใใใๅใงใฏๅๆ ผใๅฅใฎๅใงใฏไธๅๆ ผใซใชใใใจใใใใพใ(30ๅไธญ1ใ2ๅ)ใ
- ไบบใซใใ็ขบ่ชใฎ็ฏๅฒ: ่ชๅๆก็นไปฅๅคใซไบบใๅ็ญใ่ชญใใง็ขบ่ชใใใฎใฏใๆฅๆฌ่ชใฎ่ฆ็ดๅ้กใ ใใงใใ
- ่ฆ็ดใง้้กใชใฉใฎๆฐๅญใๆฏ่ผใฎๅบๆบ(ใๅๆๆฏใใชใฉ)ใๆใใใๅคใใฃใใใใใใจใใใใพใใ้่ฆใชๆฐๅญใฏ็ขบ่ชใใฆใใ ใใใ
- ๅๅใปๅๅพใฎๅ ฅใๆฟใใ: ๆๅปใฎๅๅใปๅๅพใๅ ฅใๆฟใใใใจใใใใพใ(ไพ:ใกใขใใ็ฅใใๆใซๆธใๆใใ้ใซใๅๅพ3ๆใโใๅๅ3ๆใ)ใๆฅไปใปๆๅปใฏๅฟ ใ็ขบ่ชใใฆใใ ใใใ
- ่จ่ใฎ็ฝฎใๆใใ: ไผผใๅฅใฎ่จ่ใซ็ฝฎใๆใใใๆๅณใๅคใใใใจใใใใพใ(ไพ:้ๅฝ่ชใฎใ์ ๊ณ (ๅ ฅๅบซ)ใโใ์ ๊ธ(ๅ ฅ้)ใใใ็บ้ๆฅใโใๅฐ็ๆฅใ)ใใใผในใขใใซ(Qwen3.5-4B)ใงใๅใ็พ่ฑกใใใใไปๅใฎๅญฆ็ฟใงใๅฎๅ จใซใฏ็ดใฃใฆใใพใใใ้ไฟกๅใซ็ขบ่ชใใฆใใ ใใใ
- ใกใผใซใซใไพ้ ผใใฆใใชใๅ ๅฎน(ไปฎใฎ็คพๅ)ใๅ ฅใใใจใใใใพใใ
- ๆนใพใฃใๆฌ่ชใฏไธ่ช็ถใซใชใใใจใใใใพใใ้ไฟกๅใซไบบใ็ขบ่ชใใฆใใ ใใใ
- ใณใผใใฏ็ฐกๅใชไบๅ็จใฎ้ขๆฐใปในใฏใชใใ็จๅบฆใงใไธ่ฌ็ใชใณใผใใฃใณใฐ็จใงใฏใใใพใใใ
- ใใญในใๅฐ็จใงใใใใฎ GGUF ใฏ็ปๅใซๅฏพๅฟใใฆใใพใใใๅญฆ็ฟใใ่จ่ชใฏๆฅๆฌ่ชใป่ฑ่ชใป้ๅฝ่ชใฎใฟใงใใ
ไฝใๆน
- ใใผใน:Qwen3.5-4B(Apache-2.0)ใLoRA(r=16)ใ2ใจใใใฏใๅญฆ็ฟไพ 1,400ไปถ(่ฑ่ช 473ใๆฅๆฌ่ช 472ใ้ๅฝ่ช 455)ใ
- ๅญฆ็ฟไพใฏไปฅๅใฎๅ ้จ็ใฎๅญฆ็ฟใใผใฟใใใๆฅไปใป้้กใๆใใ่ฆ็ด40ไปถใ้คใใใใฎใงใ(่ฆ็ดใใผใฟใฎๅ่ณชๆด็)ใ
- ๅ ้จ้็บๅ v2.1.1ใ
- ๅญฆ็ฟไพใฏ Gemma 4 E4B(Apache-2.0)ใงใญใผใซใซใซไธๆธใใใ่ชๅใใงใใฏ(่จ่ชใๆฐๅญ่กจ่จใไฝใใใๆฐๅญใJSONๅฝขๅผใๅฑ้บใชใณใผใ)ใจใAIใฎ่ฃๅฉใๅใใไบบใฎ็ขบ่ชใ็ตใฆใใพใใ
- ่ฉไพกๅ้กใฏๅญฆ็ฟใใผใฟใซๅซใใฆใใพใใใ
Made by Luciver (sori-forest). Base model: Qwen/Qwen3.5-4B (Apache-2.0).
- Downloads last month
- 106
3-bit
4-bit