Ammar Ahmad Awan comments

Results 24 comments of


                                            Ammar Ahmad Awan

Provide TF_CONFIG environment variable for distributed TensorFlow

Hello guys, just wondering if Batch AI is generating the new format of TF_CONFIG now?

[BUG] Different outputs by original model and inference engine

@reymondzzzz - can you please add```do_sample=False``` to your script? Please modify the line where you call HF generate for both the original model and the DS model as follows. ```ds_output...

[BUG] Failed to inference Megatron gpt-3 MoE model with `deepspeed.init_inference`

@Gabriel4256 -- please look at this example: https://github.com/microsoft/Megatron-DeepSpeed/blob/main/examples/generate_text.sh For inference, we don't use the pretrain_gpt.py as an entry-point. Please try the above text generation scenario that uses deepspeed inference. If...

fairseq v2

Looking forward to fairseq-v2! Is there any ETA on the v2 release?

Why the MoE layer forces the input hidden dimenstion the same as the output hidden dimension?

@ykim362 -- I know its a very old issue but do you mind explaining this here?

Fix checkpoint loading when zero optimizer states are not given

Thanks for your contribution @szhengac. I am closing this fairly old/stale PR due to conflicts and perhaps new functionality has already been added. If you still find this relevant, please...

Fix type checking of offload optimizer before checked class was imported

Closing this very old PR as it is no longer relevant (Megatron-LM has been deprecated and Megatron-DeepSpeed has a different codebase altogether). Please reopen if needed.

Add inference comm

Reza, I am closing this old PR as discussed in the software sync. We can reopen if/when we decide to work on it again.

[BUG] deepspeed-chat bloom training error, raise RuntimeError "still have inflight params " after 14 steps training of step3 with offload option turned on

@loadams - please see if this is connected to https://github.com/microsoft/DeepSpeed/issues/3735

Resolve hard dependency on MOE for contiguous_gradients on stage 1

@abhilash1910 -- Thanks for the PR! Have you tested performance and found issues? It will be helpful to add some more details if you can. @RezaYazdaniAminabadi, @jeffra, and @tjruwase -...