| # deep-todo | |
| Wondering what to do? Not anymore! | |
| Generate arbitrary todo's. | |
| Source: <https://colab.research.google.com/drive/1PlKLrGHaCuvWCKNC4fmQEMElF-iRec9f?usp=sharing> | |
| The todo's come from a random selection of (public) repositories I had on my computer. | |
| ### Sample | |
| A bunch of todo's: | |
| ``` | |
| ---------------------------------------------------------------------------------------------------- | |
| 0: TODO: should we check the other edges?/ | |
| 1: TODO: add more information here. | |
| 2: TODO: We could also add more general functions in this case to avoid/ | |
| 3: TODO: It seems strange to have the same constructor when the base set of/ | |
| 4: TODO: This implementation should be simplified, as it's too complex to handle the/ | |
| 5: TODO: we should be able to relax the intrinsic if not | |
| 6: TODO: Make sure this doesn't go through the next generation of plugins. It would be better if this was | |
| 7: TODO: There is always a small number of errors when we have this type/ | |
| 8: TODO: Add support for 't' values (not 't') for all the constant types/ | |
| 9: TODO: Check that we use loglef_cxx in the loop* | |
| 10: TODO: Support double or double values./ | |
| 11: TODO: Add tests that verify that this function does not work for all targets/ | |
| 12: TODO: we'd expect the result to be identical to the same value in terms of | |
| 13: TODO: We are not using a new type for 'w' as it does not denote 'y' yet, so we could/ | |
| 14: TODO: if we had to find a way to extract the source file directly, we would/ | |
| 15: TODO: this should fold into a flat array that would be/ | |
| 16: TODO: Check if we can make it work with the correct address./ | |
| 17: TODO: support v2i with V2R4+ | |
| 18: TODO: Can a fast-math-flags check be generalized to all types of data? */ | |
| 19: TODO: Add support for other type-specific VOPs. | |
| ``` | |
| Generated by: | |
| ``` | |
| tf.random.set_seed(0) | |
| sample_outputs = model.generate( | |
| input_ids, | |
| do_sample=True, | |
| max_length=40, | |
| top_k=50, | |
| top_p=0.95, | |
| num_return_sequences=20 | |
| ) | |
| print("Output:\\ | |
| " + 100 * '-') | |
| for i, sample_output in enumerate(sample_outputs): | |
| m = tokenizer.decode(sample_output, skip_special_tokens=True) | |
| m = m.split("TODO")[1].strip() | |
| print("{}: TODO{}".format(i, m)) | |
| ``` | |
| ## TODO | |
| - [ ] Fixup the data; it seems to contain multiple todo's per line | |
| - [ ] Preprocess the data in a better way | |
| - [ ] Download github and train it on everything |