Spaces:
Running
Running
Ctrl+K
- 00-scored-evidence-summary
- claim-1-poet-x-achieves-a-3x-gpu-memory-reduction-and-8x-r
- claim-1-poet-x-enables-pretraining-of-billion-parameter-ll
- claim-2-maintains-generalization-and-stability-benefits-of
- claim-2-poet-x-enables-pretraining-of-an-llm-with-up-to-13
- claim-3-on-llama-3b-pretraining-on-c4-with-l-max-256-poet
- claim-4-for-a-llama-13b-model-at-sequence-length-2048-poet
- claim-5-a-customized-cuda-kernel-for-poet-x-s-permutation
- claim-6-quantized-poet-xq-with-block-size-512-achieves-the
- conclusion
- reproduction-protocol-and-provenance
- 1.16 kB