| Age | Commit message (Collapse) | Author | Files | Lines | |
|---|---|---|---|---|---|
| 2025-08-24 | try finetuning encoder-decoder model like t5gemmaHEADmaster | vin | 1 | -1/+1 | |
| Unfortunately once it finishes the finetuning process, it doesn't actually output anything. With decoder-only models like small gemma3, it also doesn't output anything meaningful. I'm tempted to just classify each token as an ad or not, finetune an encoder like some long-context BERT, and then a post-processing script to convert back to tokens and their timestamps. May end up being more performant and efficient too... | |||||
| 2025-08-23 | add highlight/visualizer script and update system prompt | vin | 1 | -30/+37 | |
| 2025-08-22 | update main script | vin | 1 | -143/+127 | |
| 2025-07-19 | provide better examples in system prompt and retry on error | vin | 1 | -69/+215 | |
| 2025-07-19 | save output in data/, not where podcasts are | vin | 1 | -4/+14 | |
| 2025-07-19 | change chunking to overlapping time segments | vin | 1 | -55/+52 | |
| 2025-07-19 | output data in jsonl | vin | 1 | -4/+6 | |
| 2025-07-19 | respect ratelimit and change output extension to .gpt | vin | 1 | -40/+38 | |
| 2025-07-19 | remove prototyping llm prints, use tqdm, use arguments as inputs | vin | 1 | -140/+134 | |
| 2025-07-18 | initial payload | vin | 1 | -0/+292 | |
 |
index : podcast-sponsor-remove | |
| Attempt at identify sponsored segments in audio transcripts and removing them. | Vineet K |
| summaryrefslogtreecommitdiff |