dcb183f4bd
* draft add new model like * some cleaning of the config * nits * add nested configs * nits * update * update * added layer norms + triton kernels * consider only LPLayerNorm for now. * update * all keys match. * Update * fixing nits here and there * working forward pass. * removed einops dependency * nits * format * add alibi * byebye head mask * refactor attention * nits. * format * fix nits. * nuke ande updates * nuke tokenizer test * don't reshape query with kv heads * added a bit of documentation. * remove unneeded things * nuke more stuff * nit * logits match - same generations * rm unneeded methods * 1 remaining failing CI test * nit * fix nits * fix docs * fix docs * rm tokenizer * fixup * fixup * fixup and fix tests * fixed configuration object. * use correct activation * few minor fixes * clarify docs a bit * logits match à 1e-12 * skip and unskip a test * added some slow tests. * fix readme * add more details * Update docs/source/en/model_doc/mpt.md Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Apply suggestions from code review Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * fix configuration issues * more fixes in config * added more models * Apply suggestions from code review Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * remove unneeded position ids * fix some comments * Apply suggestions from code review Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * revert suggestion * mpt alibi + added batched generation * Update src/transformers/models/mpt/__init__.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * remove init config * Update src/transformers/models/mpt/configuration_mpt.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * fix nit * add another slow test * Apply suggestions from code review Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> * fits in one line * some refactor because make fixup doesn't pass * add ft notebook * update md * correct doc path --------- Co-authored-by: younesbelkada <younesbelkada@gmail.com> Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com> Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
2.8 KiB
MPT
Overview
The MPT model was proposed by the MosaicML team and released with multiple sizes and finetuned variants. The MPT models is a series of open source and commercially usable LLMs pre-trained on 1T tokens.
MPT models are GPT-style decoder-only transformers with several improvements: performance-optimized layer implementations, architecture changes that provide greater training stability, and the elimination of context length limits by replacing positional embeddings with ALiBi.
- MPT base: MPT base pre-trained models on next token prediction
- MPT instruct: MPT base models fine-tuned on instruction based tasks
- MPT storywriter: MPT base models fine-tuned for 2500 steps on 65k-token excerpts of fiction books contained in the books3 corpus, this enables the model to handle very long sequences
The original code is available at the llm-foundry repository.
Read more about it in the release blogpost
Tips:
-
Learn more about some techniques behind training of the model in this section of llm-foundry repository
-
If you want to use the advanced version of the model (triton kernels, direct flash attention integration), you can still use the original model implementation by adding
trust_remote_code=Truewhen callingfrom_pretrained. -
Fine-tuning Notebook on how to fine-tune MPT-7B on a free Google Colab instance to turn the model into a Chatbot.
MptConfig
autodoc MptConfig - all
MptModel
autodoc MptModel - forward
MptForCausalLM
autodoc MptForCausalLM - forward
MptForSequenceClassification
autodoc MptForSequenceClassification - forward
MptForTokenClassification
autodoc MptForTokenClassification - forward
MptForQuestionAnswering
autodoc MptForQuestionAnswering - forward