jimfhahn/bibframe-olmo-1b-onnx

BIBFRAME-OLMo-1B for Transformers.js (WebGPU)

Fine-tuned OLMo-1B model for correcting BIBFRAME RDF/XML records, exported for browser-based inference with Transformers.js and WebGPU.

Usage with Transformers.js

import { pipeline } from '@huggingface/transformers';

const generator = await pipeline(
  'text-generation',
  'jimfhahn/bibframe-olmo-1b-onnx',
  { device: 'webgpu', dtype: 'fp16' }
);

const result = await generator(prompt, {
  max_new_tokens: 512,
  temperature: 0.3,
});

Model Details

• Base model: amd/AMD-OLMo-1B (native transformers format)
• Fine-tuning: LoRA on BIBFRAME correction task
• Export: ONNX FP16 optimized for WebGPU
• Size: ~2GB (FP16)

Prompt Format (ChatML)

The model was trained with ChatML format:

<|im_start|>system
You are a BIBFRAME expert assistant...<|im_end|>
<|im_start|>user
Fix the following invalid BIBFRAME RDF/XML:
...
<|im_end|>
<|im_start|>assistant

Related

• jimfhahn/bibframe-olmo-1b-v2 - LoRA adapter