{ "

Fine Tune GPT-NeoX

\n

This shows how to fine tune GPT-NeoX with pipeline parallelism.

\n": "

Fine Tune GPT-NEOX

\n

\u8fd9\u8bf4\u660e\u4e86\u5982\u4f55\u5229\u7528\u7ba1\u9053\u5e76\u884c\u5ea6\u5bf9 GPT-NEOX \u8fdb\u884c\u5fae\u8c03\u3002

\n", "

Create fine tuner for biases

\n": "

\u4e3a\u504f\u89c1\u521b\u5efa\u7cbe\u7ec6\u7684\u8c03\u8c10\u5668

\n", "

Create pipeline parallel model

\n": "

\u521b\u5efa\u7ba1\u9053\u5e76\u884c\u6a21\u578b

\n", "

Load GPT-NeoX layers

\n": "

\u52a0\u8f7d GPT-NEOX \u56fe\u5c42

\n", "

Tiny Shakespeare dataset

\n": "

\u5c0f\u838e\u58eb\u6bd4\u4e9a\u6570\u636e\u96c6

\n", "

\n": "

\n", "

Create Fairscale Pipe module

\n": "

\u521b\u5efa\u516c\u5e73\u89c4\u6a21\u7ba1\u9053\u6a21\u5757

\n", "

Create experiment

\n": "

\u521b\u5efa\u5b9e\u9a8c

\n", "

Create the Pipe module

\n": "

\u521b\u5efa\u7ba1\u9053\u6a21\u5757

\n", "

Devices for each GPU

\n": "

\u6bcf\u4e2a GPU \u7684\u8bbe\u5907

\n", "

Get the layer distribution across GPUs

\n": "

\u83b7\u53d6\u8de8 GPU \u7684\u5c42\u5206\u5e03

\n", "

Initialize configs

\n": "

\u521d\u59cb\u5316\u914d\u7f6e

\n", "

Initialize the model. Do this before the loop for cleaner logs.

\n": "

\u521d\u59cb\u5316\u6a21\u578b\u3002\u5728\u5faa\u73af\u4e4b\u524d\u6267\u884c\u6b64\u64cd\u4f5c\u4ee5\u83b7\u5f97\u66f4\u6e05\u6670\u7684\u65e5\u5fd7\u3002

\n", "

Make sure the finetuner is initialized

\n": "

\u786e\u4fdd\u5fae\u8c03\u5668\u5df2\u521d\u59cb\u5316

\n", "

Mark biases as trainable

\n": "

\u5c06\u504f\u89c1\u6807\u8bb0\u4e3a\u53ef\u8bad\u7ec3

\n", "

Start the experiment

\n": "

\u5f00\u59cb\u5b9e\u9a8c

\n", "

Train

\n": "

\u706b\u8f66

\n", "Fine Tune GPT-NeoX": "Fine Tune GPT-NEOX", "Fine tune GPT-NeoX biases with Fairscale pipeline parallel module": "\u4f7f\u7528 Fairscale \u7ba1\u9053\u5e76\u884c\u6a21\u5757\u5fae\u8c03 GPT-NEOX \u504f\u5dee" }