chandra-twin-onnx (bilingual)

The on-device "AI twin" answering questions on tungathurthi.com in English and German: a LoRA fine-tune of LFM2.5-230M that speaks in first person as Chandra Tungathurthi (AI consultant, Essen, Germany).

Trained on a bilingual merge (English corpus + German corpus oversampled to ~40%), each training row carrying its own language's system prompt and retrieval context โ€” the runtime selects the language via the system prompt and knowledge base it serves. The model answers from retrieved context (### External Info: blocks) and emits plain prose only (twin-response/v1); served fully client-side via Transformers.js + WebGPU.

Variants

File Precision Size Scorecards
onnx/model_fp16.onnx fp16 ~454 MB Recommended (WebGPU) โ€” English 9/13, German 8/12
onnx/model_q4.onnx int4 block-32 ~211 MB Light option โ€” English 9/13, German 6/12

Exported with LiquidONNX.

Intended use

Powers the chat on Chandra's own website. Answers may be imperfect โ€” the authoritative source is the person: letsconnect@tungathurthi.com.

Downloads last month
342
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for tckb/chandra-twin-onnx

Quantized
(19)
this model