Instructions to use UnimeType/Transliteration-2.0-9B-Safetensors with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use UnimeType/Transliteration-2.0-9B-Safetensors with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="UnimeType/Transliteration-2.0-9B-Safetensors") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("UnimeType/Transliteration-2.0-9B-Safetensors") model = AutoModelForCausalLM.from_pretrained("UnimeType/Transliteration-2.0-9B-Safetensors", device_map="auto") messages = [ {"role": "user", "content": "Who are you?"}, ] inputs = tokenizer.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=256) print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use UnimeType/Transliteration-2.0-9B-Safetensors with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "UnimeType/Transliteration-2.0-9B-Safetensors" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "UnimeType/Transliteration-2.0-9B-Safetensors", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/UnimeType/Transliteration-2.0-9B-Safetensors
- SGLang
How to use UnimeType/Transliteration-2.0-9B-Safetensors 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 "UnimeType/Transliteration-2.0-9B-Safetensors" \ --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": "UnimeType/Transliteration-2.0-9B-Safetensors", "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 "UnimeType/Transliteration-2.0-9B-Safetensors" \ --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": "UnimeType/Transliteration-2.0-9B-Safetensors", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use UnimeType/Transliteration-2.0-9B-Safetensors with Docker Model Runner:
docker model run hf.co/UnimeType/Transliteration-2.0-9B-Safetensors
UnimeType Transliteration 2.0 9B โ v2.6 (12-Language Offline IME Engine ยท Standard BF16 Safetensors)
Current Release:
v2.6(dataset_v5.1ยท 12 Languages / 13 Modes + Blind-Typing Typo Recovery) ยท Protocol:unimetype-convert-short-v3(0system-prompt tokens, 100% drop-in compatible with allTransliteration-2.0clients)
UnimeType gives you one keyboard for all your languages โ no switching keyboards across apps or mid-sentence. Keep a clean ASCII layout for single-key shortcuts in Figma or Blender, LaTeX commands in Overleaf, WASD controls in games, Terminal/IDE workflows, and switching between Slack, Mail, WhatsApp, WeChat, or LINE, then convert romanized text in-place only when you want native script.
UnimeType Transliteration 2.0 9B (v2.6) is a deterministic on-device input-method transliteration engine fine-tuned from unquantized BF16 Qwen/Qwen3.5-9B. It converts romanized or ASCII-typed input across 12 languages (13 output modes) โ Chinese Pinyin, Japanese Romaji (natural & kana), Hindi Hinglish, Arabic Arabizi, Russian, Greek, Vietnamese, Spanish, French, Portuguese, German, and Italian โ while keeping inline English terms, @handles, URLs, code spans, file paths, numbers, and line breaks untouched.
- ๐ ๏ธ Local Inference Toolkit, Ollama Modelfiles & Short-v3 Spec (GitHub) โ (One-command Ollama, LM Studio, Apple Silicon
mlx-lm&llama.cpplaunchers +Smoke-60local eval) - ๐ Launch Interactive Playground Space โ (Inspect multi-clause presets or run live in-browser WebAssembly inference)
- โจ๏ธ Download the UnimeType Keyboard App โ (One keyboard across macOS, Windows, Linux, iOS, Android & ChromeOS โ zero keyboard switching across apps or mid-sentence)
- ๐ Evaluation Methodology & Local LM Studio Setup Guide โ
Official v2.6 Model Family (0.8B ยท 4B ยท 9B)
| Tier & Footprint | Apple Silicon (MLX 6-bit) |
PyTorch / vLLM (BF16 Safetensors) |
llama.cpp / LM Studio / Ollama (GGUF) |
|---|---|---|---|
0.8B ยท Ultra-light mobile & browser WASM (~0.5โ0.7 GB) |
MLX 6-bit | BF16 Safetensors | GGUF Q4_K_M & Q8_0 |
4B ยท Balanced everyday desktop IME (~2.6โ3.5 GB) |
MLX 6-bit | BF16 Safetensors | GGUF Q4_K_M & Q8_0 |
9B ยท Flagship offline accuracy (~5.6โ7.4 GB, beats cloud on Core-5) |
MLX 6-bit | BF16 Safetensors (this repo) | GGUF Q4_K_M & Q8_0 |
What's New in v2.6
- 12 Languages / 13 Output Modes with Zero System Prompt: Expands from 4 languages (
zh,ja,hi,ar) to 12 languages (chinese,japanesenatural&kana,hindi,arabic,russian,greek,vietnamese,spanish,french,portuguese,german,italian) using a single-line compact JSON user message and an empty system prompt (0prompt-prefix overhead). - New
9BFlagship Offline Tier Surpasses Cloud on Core Benchmarks:Transliteration-2.0-9Breaches92.09%strict accuracy on the708-casev4-aligned suite (+1.27 ppahead ofDeepSeek V4.1 Flashwith a ~2,800-token production system prompt) and90.94%onCore-5(+0.64 ppahead of cloud), while matching cloud accuracy onDogfood-30(76.7%) andOOD-Unseen-20probe repair (90.0%,18/20). - Blind-Typing Adjacent-Key Typo Repair & Glossary Precision: Recovers realistic fast-typing transpositions (e.g.,
shujuukโๆฐๆฎๅบ,yanchanghiuโๆผๅฑไผ,yanuhuoโ้ช่ดง), honors dynamiclexiconcontact/brand entries without false-positive homophone hijacking, and cleanly emitsUNIMETYPE_NO_ALTERNATIVEwhen closed-setreconvertalternatives are exhausted. - Unconditional Non-Thinking Chat Template: Ships with
<|im_start|>assistant\n<think>\n\n</think>\n\nbaked intochat_template.jinjaacross MLX, Safetensors, and GGUF so LM Studio,mlx-lm, andllama.cppemit converted text on the very first token with zero reasoning-token latency.
Quickstart: unimetype-convert-short-v3 Compact JSON Protocol
Usage Note: This checkpoint is a specialized IME conversion engine, not an open-ended chatbot. Send a compact JSON object as the
usermessage with empty System Prompt andtemperature: 0, or pair LM Studio / Ollama directly with the UnimeType Keyboard App for system-wide Caps Lock / Convert key replacement in any app.
Scenario (1 Native Language + English) |
user Message JSON (content) |
Direct assistant Output |
|---|---|---|
Chinese + English (zh) |
{"operation":"convert","input":"qing xian zai staging huanjing pao yi bian suoyou unit test, queren API latency meiyou shenggao zhi hou, women zai anpai production deploy","enabled_languages":["chinese","english"],"primary_language":"chinese"} |
่ฏทๅ
ๅจ staging ็ฏๅข่ทไธ้ๆๆ unit test, ็กฎ่ฎค API latency ๆฒกๆๅ้ซไนๅ, ๆไปฌๅๅฎๆ production deploy |
Japanese + English (ja) |
{"operation":"convert","input":"staging kankyou de atarashii API endpoint no dousa wo kakunin shite kara, mondai ga nakereba ban-gohan no ato ni production deploy shimasu","enabled_languages":["english","japanese"],"primary_language":"japanese","japanese_output_style":"natural"} |
staging ็ฐๅขใงๆฐใใ API endpoint ใฎๅไฝใ็ขบ่ชใใฆใใ, ๅ้กใใชใใใฐๆฉใ้ฃฏใฎๅพใซ production deploy ใใพใ |
Hindi + English (hi) |
{"operation":"convert","input":"maine GitHub par naya PR open kar diya hai, aap ek baar code review kar lijiye, phir hum shaam ko staging server par deploy karenge","enabled_languages":["english","hindi"],"primary_language":"hindi"} |
เคฎเฅเคเคจเฅ GitHub เคชเคฐ เคจเคฏเคพ PR open เคเคฐ เคฆเคฟเคฏเคพ เคนเฅ, เคเคช เคเค เคฌเคพเคฐ code review เคเคฐ เคฒเฅเคเคฟเค, เคซเคฟเคฐ เคนเคฎ เคถเคพเคฎ เคเฅ staging server เคชเคฐ deploy เคเคฐเฅเคเคเฅ |
Arabic + English (ar) |
{"operation":"convert","input":"ana 7a2a3od fel office l7ad el sa3a 8 3ashan a5allas el deployment, law 3andak ay update eb3atli 3ala Slack aw WhatsApp","enabled_languages":["arabic","english"],"primary_language":"arabic"} |
ุฃูุง ููุนุฏ ูู ุงู office ูุญุฏ ุงูุณุงุนุฉ 8 ุนุดุงู ุฃุฎูุต ุงู deployment, ูู ุนูุฏู ุฃู update ุงุจุนุชูู ุนูู Slack ุฃู WhatsApp |
Russian + English (ru) |
{"operation":"convert","input":"pozhaluista prover novyi Docker build na staging pered tem kak my nachnem release","enabled_languages":["english","russian"],"primary_language":"russian"} |
ะฟะพะถะฐะปัะนััะฐ ะฟัะพะฒะตัั ะฝะพะฒัะน Docker build ะฝะฐ staging ะฟะตัะตะด ัะตะผ ะบะฐะบ ะผั ะฝะฐัะฝะตะผ release |
German + English (de) |
{"operation":"convert","input":"koennen wir das API Gateway fuer den neuen Docker Container heute noch ueberpruefen","enabled_languages":["english","german"],"primary_language":"german"} |
kรถnnen wir das API Gateway fรผr den neuen Docker Container heute noch รผberprรผfen |
curl http://127.0.0.1:1234/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "UnimeType/Transliteration-2.0-9B-Safetensors",
"temperature": 0,
"max_tokens": 256,
"messages": [
{
"role": "user",
"content": "{\"operation\":\"convert\",\"input\":\"mingtian 3pm he @alex review auth-service de PR #412, ranhou deploy dao staging\",\"enabled_languages\":[\"chinese\",\"english\"],\"primary_language\":\"chinese\"}"
}
]
}'
Evaluation Benchmarks (v2.6)
1. Held-Out Multi-Suite Benchmark: Offline v2.6 (0.8B / 4B / 9B) vs. Cloud DeepSeek V4.1 Flash
All offline v2.6 models run with 0 system-prompt tokens (system_prompt = "", compact JSON input, temperature = 0), compared head-to-head against UnimeType's production cloud baseline (deepseek-v4.1-flash paired with the full ~2,800-token v30 production system prompt and boundary guards). Zero sentence or OOD probe overlap (0.00%) exists between training data and any evaluation suite.
| Model / Tier | System Prompt Overhead | 708 Shared (v4-Aligned) |
Core-5 (773 Cases) |
12-Language / 13-Mode (1,573 Cases) |
Dogfood-30 Real Pain Points (Exact / Tolerant) |
Zero-Overlap OOD-Unseen-20 (Exact / Tolerant / Probe Repair) |
|---|---|---|---|---|---|---|
DeepSeek V4.1 Flash (Cloud Baseline) |
~2,800 tok (v30) |
90.82% (643/708) | 90.30% (698/773) | 93.20% (1466/1573) | 53.3% / 76.7% (23/30) | 45% / 75% / 90% (18/20) |
Transliteration-2.0-0.8B (v2.6) |
0 tok (short JSON) | 72.32% (512/708) | 71.15% (550/773) | 69.99% (1101/1573) | 36.7% / 36.7% (11/30) | 25% / 25% / 45% (9/20) |
Transliteration-2.0-4B (v2.6) |
0 tok (short JSON) | 89.12% (631/708) | 87.45% (676/773) | 88.11% (1386/1573) | 50.0% / 56.7% (17/30) | 35% / 35% / 75% (15/20) |
Transliteration-2.0-9B (v2.6) |
0 tok (short JSON) | 92.09% (+1.27 pp vs cloud) |
90.94% (+0.64 pp vs cloud) |
91.54% (1440/1573) | 73.3% / 76.7% (= cloud) |
55% / 60% / 90% (18/20, = cloud) |
Reproducible Local Benchmark (
Smoke-60): Run the open 60-case real-world blind-typing suite (eval/smoke_60.jsonl) on your own hardware via github.com/UnimeType/transliteration โ both4B(P50 376 ms) and9B(P50 722 ms) achieve 100.0% (99/99) protected-span preservation (URL,@handle,#issue, hex colors, CLI flags, file paths, and LaTeX\cite{}) and 100% pass on closed-setreconvertexhaustion (UNIMETYPE_NO_ALTERNATIVE).
2. Per-Language Strict Accuracy Across All 12 Languages / 13 Modes (1,573 Held-Out Cases)
| Language / Output Mode | Cases | DeepSeek V4.1 Flash (Cloud) |
v2.6 0.8B |
v2.6 4B |
v2.6 9B |
|---|---|---|---|---|---|
Chinese Pinyin (chinese) |
261 | 88.1% | 75.9% | 91.2% | 93.5% |
Japanese Kanji-Kana (japanese:natural) |
139 | 95.7% | 60.4% | 85.6% | 87.0% |
Japanese Kana-Only (japanese:kana) |
100 | 90.0% | 66.0% | 75.0% | 86.0% |
Hindi Hinglish (hindi) |
135 | 91.1% | 80.7% | 94.8% | 99.3% |
Arabic Arabizi (arabic) |
138 | 88.4% | 67.4% | 84.1% | 85.5% |
Russian Translit (russian) |
100 | 98.0% | 52.0% | 88.0% | 91.0% |
Greek Greeklish (greek) |
100 | 85.0% | 49.0% | 77.0% | 85.0% |
Vietnamese Telex / ASCII (vietnamese) |
100 | 97.0% | 43.0% | 84.0% | 88.0% |
Spanish ASCII (spanish) |
100 | 96.0% | 71.0% | 85.0% | 93.0% |
French ASCII (french) |
100 | 95.0% | 77.0% | 92.0% | 87.0% |
Portuguese ASCII (portuguese) |
100 | 99.0% | 88.0% | 97.0% | 99.0% |
German ASCII / Umlaut (german) |
100 | 99.0% | 88.0% | 96.0% | 100.0% |
Italian ASCII (italian) |
100 | 99.0% | 83.0% | 91.0% | 94.0% |
Version History & Reproducible Revisions
This repository tracks the latest 2.x release on main while preserving immutable Git tags for every milestone:
v2.6(default /main): Initial public release of the9Bflagship offline tier (dataset_v5.1), supporting 12 languages / 13 output modes with92.09%accuracy on the708-casev4-aligned suite and90.94%onCore-5(773cases).huggingface-cli download UnimeType/Transliteration-2.0-9B-Safetensors --revision v2.6
- Downloads last month
- 38