Offline models on this machine

Inventory of local translation assets for zen / lang-pipe. Path root: /Volumes/qbitOS/03.models/07-lang-ugrad/ + Ollama under /Volumes/qbitOS/03.models/01-ollama.

On disk · top skill pack

Engine	Skill	Path / how	State	Use
qwen3:8b	92%	Ollama · `qwen3:8b`	…	zen `offline` / `ollama` · default
NLLB-200 distilled 600M	94%	`07-lang-ugrad/nllb-200-distilled-600M/`	…	best dedicated offline NMT · ar/fa/el/hi
m2m100 1.2B	90%	`07-lang-ugrad/m2m100-1.2b/`	…	edge parity offline · bulk fanout
IndicTrans2	88%	`07-lang-ugrad/indictrans2/`	…	Indic specialist (hi + family)
mistral:latest	78%	Ollama · `mistral`	…	EU alt offline
m2m100 / llama (edge CF)	90% / 72%	Cloudflare Workers AI	online	zen `edge` · no local weight needed

Scanning disk…

Ollama (live probe)

Checking 127.0.0.1:11434…

Use offline on zen

# default offline = Ollama qwen3:8b
open 'https://lang.ugrad.ai/zen.html?provider=offline&to=fr,es,de,uk,ar,ja,zh'
open 'http://127.0.0.1:8790/zen.html?provider=offline&test=short'

# CLI
cd scripts/language/lang-ugrad-local
./cli/lang-pipe ai --provider ollama --action translate --text "all clear" --target uk

# keep Ollama up (models live under /Volumes/qbitOS/03.models/01-ollama)
ollama serve
ollama list   # expect qwen3:8b · mistral · …

Weight roots

/Volumes/qbitOS/03.models/07-lang-ugrad/
  nllb-200-distilled-600M/   # P0 NMT
  m2m100-1.2b/               # P0 edge-parity NMT
  indictrans2/               # P2 Indic
  llama-3.1-8b-instruct-fp8/ # optional empty · use qwen3

/Volumes/qbitOS/03.models/01-ollama/   # Ollama blobs (symlink from ~/.ollama/models)

Zen · offline short Zen · convoy offline Full engine ranking Charts mt-models.json OFFLINE_PACK.md