[mthreads] deepspeed llama2 #354
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…gOpen#346) * [kunlunxin] fix tacotron2 running error and add 1x1 & 2x8 config * [kunlunxin] modify tacotron2 test_config * [kunlunxin] update tacotron2 readme * [kunlunxin] modify tacotron2 torch.load()
* update iluvatar/swin_transformer-pytorch * update * update * update * fix batch size mistake in readme * correct val_loss to final acc1 * add finnal_acc1 and mem in readme * correct readme mem --------- Co-authored-by: 魏杰 <[email protected]> Co-authored-by: 杨智超 <[email protected]> Co-authored-by: clveryang <[email protected]>
Co-authored-by: zhouyu <[email protected]>
shang-mt
force-pushed
the
llama2-base
branch
3 times, most recently
from
4bade8b to
9c7f506
Compare
Co-authored-by: sen.li <[email protected]>
* Update README.md * Update README.md
* iluvatar bertlarge MLM inference case * update ixrt readme --------- Co-authored-by: 杨智超 <[email protected]>
yuzhou03
reviewed
Dec 11, 2023
yuzhou03
approved these changes
Dec 11, 2023
shang-mt
force-pushed
the
llama2-base
branch
2 times, most recently
from
b9abb42 to
9a02a7d
Compare
yuzhou03
force-pushed
the
llama2-base
branch
from
9a02a7d to
ad1500b
Compare
* support bert_hf fp32/amp/bf16 training for mthreads * update readme * prevent overrun * 1x1/2x8 not support
* support resnet50 training on mthreads * fix typo * support rn50 amp training on mthreads * add test config (should revert this commit) * update config & readme * add get_system_info fn * update * 1x1/2x8 not support --------- Co-authored-by: Zhou Yu <[email protected]>
upvenly
reviewed
Dec 12, 2023
| @@ -54,18 +54,19 @@ def get_argument_parser(): | |||
|
|
|||
| def train(model_engine, dataloader): | |||
| model_engine.train() | |||
| device = torch.device('musa:'+str(args.local_rank)) | |||
| @@ -6,5 +6,6 @@ def get_llama_model(model_config_dir, flashattn): | |||
| config = LlamaConfig.from_pretrained(model_config_dir) | |||
| config._flash_attn_2_enabled = flashattn | |||
| model = LlamaForCausalLM(config) | |||
| model.gradient_checkpointing_enable() | |||
There was a problem hiding this comment.
这个地方需要修改成以配置文件控制开关，NV默认关，厂商自行开关。
* fixllama * add t/tflops
shang-mt
force-pushed
the
llama2-base
branch
from
ad1500b to
5fae218
Compare
shh2000
reviewed
Dec 14, 2023
|
|
||
| - ##### 优化策略 | ||
|
|
||
| - 无 |
There was a problem hiding this comment.
在这里标注一下使用的优化策略，例如
-flash attention（1/2/sdp-attn）
-checkpointing
shh2000
reviewed
Dec 15, 2023
upvenly
approved these changes
Dec 15, 2023
shh2000
reviewed
Dec 15, 2023
| datafilename = "openwebtext_llama2_100M.npy" | ||
| epochs = 1 | ||
| theoryflops = 98000000000000.0 | ||
| flashattn = True |
There was a problem hiding this comment.
flashattn=True # using sdp attention
shh2000
reviewed
Dec 20, 2023
shang-mt
force-pushed
the
llama2-base
branch
from
a1c98e3 to
8b18240
Compare
shh2000
approved these changes
Dec 20, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.