Edit model card

tech-keyword-extractor

This model is a fine-tuned version of facebook/bart-large on a private dataset. It achieves the following results on the evaluation set:

  • Loss: 0.8795

Model description

This model extracts tech terms, tools, company names from texts so they can easily be aggregated. It is trained to extract tech terms, tools, languages, platforms but may be used on other texts.

Intended uses & limitations

Use to extract keywords from texts.

Example text: "If a task raises an exception, or a worker process dies, Celery will by default lose the job. So if you happen to reboot or redeploy, any running jobs with be lost to the sands of time."

Output: "Celery, Exception Handling, Worker Process"

Example text: "Spin 2.0 – open-source tool for building and running WebAssembly applications -"

Output: "Spin 2.0, WebAssembly, Open Source"

Example text: "Do you think that low-code and no-code is a threat for developers in the long term?"

Output: "Low Code, No Code, Developers"

Example text: "I'm reaching out for some guidance on choosing the right no-code or low-code platform for my web app development projects. As a proficient back-end developer with a strong grasp of AWS, I have always struggled with front-end development"

Output: "No Code, Low Code, Web App Development, AWS"

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 5e-05
  • train_batch_size: 4
  • eval_batch_size: 4
  • seed: 42
  • gradient_accumulation_steps: 16
  • total_train_batch_size: 64
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • num_epochs: 3

Training results

Training Loss Epoch Step Validation Loss
1.5095 0.44 50 1.1766
1.1875 0.89 100 0.9652
1.0428 1.33 150 1.0587
0.9392 1.78 200 0.8968
0.786 2.22 250 1.0131
0.8503 2.67 300 0.8795

Framework versions

  • Transformers 4.35.2
  • Pytorch 2.1.0+cu118
  • Datasets 2.15.0
  • Tokenizers 0.15.0
Downloads last month
3,829
Safetensors
Model size
406M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for ilsilfverskiold/tech-keywords-extractor

Finetuned
(137)
this model