-
# Summary
Add support for at least inference on greco:
https://habana.ai/inference/greco/
# Problem statement
As today, there does nt seem to be any connection/wrappers between the synapseAI sdk…
-
I use batch size = 32 or 64, gaudi is occupied all 32gb/32gb. Is it the mechanism of gaudi and Habana ?
So that 32gb does not mean any thing?
-
- I set world size = 4, and habana auto selects 4 last ones.
- But I can not use the gaudi index from 0 to 3 in another process
![image](https://user-images.githubusercontent.com/81377071/163561386-…
-
## 🚀 Feature
Recently Intel's Habana Labs announced [AWS nodes for their Gaudi Accelerator](https://venturebeat.com/2021/10/26/amazon-launches-aws-instances-powered-by-habanas-ai-accelerator-chip/)…
-
**What happened**:
When we run a MPI Job like this, notice the Launcher and Worker have different "Resources/Requests:"
```
apiVersion: kubeflow.org/v1alpha2
kind: MPIJob
metadata:
name: f…
-
1. Can you add "(HPU)" after Habana's Gaudi processor?
🤗 Optimum Habana is the interface between the 🤗 Transformers library and [Habana's Gaudi processor](https://docs.habana.ai/en/latest/index.html…
-
With the following cmd roberta large failed at 8x
python ../gaudi_spawn.py --world_size 8 --use_mpi run_qa.py --model_name_or_path roberta-large --gaudi_config_name ../gaudi_config.json --dataset…
-
Got forbidden error while downloading pretrained BERT and vocab in this step[ Download the pretrained model for the BERT Base and Huggingface DistilBERT model variant](https://github.com/HabanaAI/Mode…
-
With gaudi_config.json
{
"execution_mode": "lazy",
"use_habana_mixed_precision": true,
"world_size": 8,
"use_fused_adam": true,
"use_fused_clip_norm": true
}
The bf16 /…
-
I run a model has 50M parameters and try with different batch size, one is 96 and other is 128. Server uses bs = 128 is crashed and work fine with bs = 96. My 1080 has 12GB memory could handle batch s…