Instructions to use Cheva123/JamjuriEDGE with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Cheva123/JamjuriEDGE with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="Cheva123/JamjuriEDGE")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("Cheva123/JamjuriEDGE", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use Cheva123/JamjuriEDGE with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "Cheva123/JamjuriEDGE" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Cheva123/JamjuriEDGE", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/Cheva123/JamjuriEDGE
- SGLang
How to use Cheva123/JamjuriEDGE 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 "Cheva123/JamjuriEDGE" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Cheva123/JamjuriEDGE", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'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 "Cheva123/JamjuriEDGE" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Cheva123/JamjuriEDGE", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use Cheva123/JamjuriEDGE with Docker Model Runner:
docker model run hf.co/Cheva123/JamjuriEDGE
JamjuriEDGE — 4B Edge Model Family
Series hub for JamjuriEDGE: compact 4B models for local, offline and tool-driven workloads.
This is the series landing card. Individual models and datasets are listed below.
✦ English
Series overview
JamjuriEDGE is Chevalabs' 4B edge family: Thai + English, coding, calculation, instruction following, tool calling, IoT automation and offline agent workflows — built on the Qwen3-4B-Instruct-2507 lineage with a Stage-1 core + Stage-2 experts + TIES merges.
- Current release:
Jamjuri-EDGE-Preview-100(Stage-1 parent + E1/E2/E3 adapters + 3 TIES candidates) - Non-thinking, assistant-only SFT, official Qwen3 chat template
- Training data released openly (see datasets below)
Models
| Repo | What's inside | Status |
|---|---|---|
Jamjuri-EDGE-Preview-100 |
Core release: Stage-1 merged parent + 3 TIES candidates (BF16) | ✅ released |
Jamjuri-EDGE-Preview-100-adapters |
Raw LoRA adapters: E1 Control · E2 Math · E3 Code + Stage-1 | ✅ released |
Datasets
| Dataset | Use |
|---|---|
jamjuri-edge-v4-stage2-datasets |
Stage-2 expert data (E1/E2/E3) + Stage-1 curriculum |
jamjuri-edge-v4-stage1 |
Stage-1 training data |
core-edge-v3-clean (private) |
Core v3 clean mixture |
expert1-tool-clean (private) |
E1 tools expert data (v3) |
expert2-thai-clean (private) |
E2 Thai SME expert data (v3) |
expert3-code-clean (private) |
E3 code expert data (v3) |
jamjuri-3stage-th-code (private) |
3-stage Thai + code curriculum |
jamjuri-mix7m-th (private) |
Thai mix 7M |
jamjuri-th-math-sci-8m (private) |
Thai math & science 8M |
coder-5m-swe-openr1-mbpp (private) |
Code 5M (SWE / OpenR1 / MBPP) |
Jamjuri-Edge-Core-v0.1 (private) |
Early core dataset (v0.1) |
core-c2-v0-clean (private) |
Core C2 v0 clean |
Experience the trade-offs: no single checkpoint wins every workload — that's why the series ships multiple candidates and raw adapters.
✦ ภาษาไทย
ภาพรวมซีรีส์
JamjuriEDGE คือ ซีรีส์โมเดล 4B สำหรับงาน edge ของ Chevalabs — ไทย+อังกฤษ, เขียนโค้ด, คิดเลข, ทำตามคำสั่ง, เรียก tool, งาน IoT/automation และ agent แบบออฟไลน์ ต่อยอดจากสาย Qwen3-4B-Instruct-2507 ด้วยสูตร Stage-1 core → Stage-2 experts → TIES merge
- รุ่นล่าสุด:
Jamjuri-EDGE-Preview-100(parent Stage-1 + adapter E1/E2/E3 + TIES 3 ตัว) - โหมด non-thinking เทรนเฉพาะคำตอบ assistant ด้วย chat template ทางการ Qwen3
- ข้อมูลเทรนเปิดทั้งหมด (ดูรายการด้านล่าง)
โมเดลในซีรีส์
| Repo | มีอะไร | สถานะ |
|---|---|---|
Jamjuri-EDGE-Preview-100 |
ตัวหลัก: Stage-1 merged parent + TIES candidates 3 ตัว (BF16) | ✅ ปล่อยแล้ว |
Jamjuri-EDGE-Preview-100-adapters |
adapter ดิบ: E1 Control · E2 Math · E3 Code + Stage-1 | ✅ ปล่อยแล้ว |
ชุดข้อมูล
| Dataset | ใช้ทำอะไร |
|---|---|
jamjuri-edge-v4-stage2-datasets |
ข้อมูล expert Stage-2 (E1/E2/E3) + curriculum Stage-1 |
jamjuri-edge-v4-stage1 |
ข้อมูลเทรน Stage-1 |
core-edge-v3-clean (private) |
ชุด Core v3 (คลีน) |
expert1-tool-clean (private) |
ข้อมูล expert E1 (tools, v3) |
expert2-thai-clean (private) |
ข้อมูล expert E2 (ไทย SME, v3) |
expert3-code-clean (private) |
ข้อมูล expert E3 (โค้ด, v3) |
jamjuri-3stage-th-code (private) |
curriculum 3 stage (ไทย + โค้ด) |
jamjuri-mix7m-th (private) |
ไทย mix 7M |
jamjuri-th-math-sci-8m (private) |
คณิต+วิทย์ไทย 8M |
coder-5m-swe-openr1-mbpp (private) |
โค้ด 5M (SWE / OpenR1 / MBPP) |
Jamjuri-Edge-Core-v0.1 (private) |
ชุด core ยุคแรก (v0.1) |
core-c2-v0-clean (private) |
Core C2 v0 (คลีน) |
จุดขายของซีรีส์นี้คือ ไม่มี checkpoint ไหนชนะทุกงาน — เราจึงปล่อยหลาย candidate และ adapter ดิบให้เลือกใช้หรือเอาไป merge เอง
Series collection: https://huggingface.co/collections/Cheva123/jamjuriedge-4b-series-6ab37890b777c7d22568b9d4
