Skip to content

Commit

Permalink
fix docs typo
Browse files Browse the repository at this point in the history
  • Loading branch information
nemonameless committed Oct 11, 2023
1 parent 8946e24 commit 5c6d798
Showing 1 changed file with 6 additions and 6 deletions.
12 changes: 6 additions & 6 deletions paddlemix/examples/eva02/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -117,7 +117,7 @@ export $PATH=$PATH:$INSTALL_DIR

1. 如果采用分布式策略,分布式并行关系有:`nnodes * nproc_per_node == tensor_parallel_degree * sharding_parallel_degree * dp_parallel_degree`,其中`dp_parallel_degree`参数根据其他几个值计算出来,因此需要保证`nnodes * nproc_per_node >= tensor_parallel_degree * sharding_parallel_degree`
2. `model_name` 可单独使用创建模型,如果更换teacher,则需自己改写`paddlemix/EVA/EVA02/eva02_Ti_for_pretrain`中config.json and model_config.json的teacher_config这个字段的内容,比如将默认的 `paddlemix/EVA/EVA01-CLIP-g-14` 改为 "paddlemix/EVA/EVA02-CLIP-bigE-14"。而student_config是dict,student模型本身是train from scratch的;
3. 如果model_name=None,也可采用teacher_name 和 student_name来创建模型,但它们必须都各自具有config.json和model_state.pdparams,一般eval或加载全量权重debug时采用model_name=None的形式
3. 如果 model_name=None,也可采用 teacher_name 和 student_name 来创建模型,但它们必须都各自具有config.json和model_state.pdparams,一般eval或加载全量权重debug时采用 model_name=None 的形式
4. `TEA_PRETRAIN_CKPT`通常情况下设置为None,模型训练前已加载来自`teacher_name`中的对应teacher预训练权重。但是**如果设置 MP_DEGREE > 1**时,则必须再次设置`TEA_PRETRAIN_CKPT`的路径去加载,一般设置绝对路径,也可从对应的下载链接单独下载相应的`model_state.pdparams`并放置;


Expand Down Expand Up @@ -145,8 +145,8 @@ TRAINING_MODEL_RESUME="None"
TRAINER_INSTANCES='127.0.0.1'
MASTER='127.0.0.1:8080'

TRAINERS_NUM=1 # machine num
TRAINING_GPUS_PER_NODE=8
TRAINERS_NUM=1 # nnodes, machine num
TRAINING_GPUS_PER_NODE=8 # nproc_per_node
DP_DEGREE=8 # dp_parallel_degree
MP_DEGREE=1 # tensor_parallel_degree
SHARDING_DEGREE=1 # sharding_parallel_degree
Expand Down Expand Up @@ -225,7 +225,7 @@ model_name="paddlemix/EVA/EVA02/eva02_Ti_for_pretrain" # should modify teacher_c
teacher_name="paddlemix/EVA/EVA02-CLIP-bigE-14"
student_name="paddlemix/EVA/EVA02/eva02_Ti_pt_in21k_p14"
TEA_PRETRAIN_CKPT=None # /root/.paddlenlp/models/paddlemix/EVA/EVA01-CLIP-bigE-14/model_state.pdparams # must add if MP_DEGREE > 1
TEA_PRETRAIN_CKPT=None # /root/.paddlenlp/models/paddlemix/EVA/EVA02-CLIP-bigE-14/model_state.pdparams # must add if MP_DEGREE > 1
STU_PRETRAIN_CKPT=None
```

Expand Down Expand Up @@ -275,8 +275,8 @@ TRAINING_MODEL_RESUME="None"
TRAINER_INSTANCES='127.0.0.1'
MASTER='127.0.0.1:8080'

TRAINERS_NUM=1 # machine num
TRAINING_GPUS_PER_NODE=8
TRAINERS_NUM=1 # nnodes, machine num
TRAINING_GPUS_PER_NODE=8 # nproc_per_node
DP_DEGREE=8 # dp_parallel_degree
MP_DEGREE=1 # tensor_parallel_degree
SHARDING_DEGREE=1 # sharding_parallel_degree
Expand Down

0 comments on commit 5c6d798

Please sign in to comment.