-
> if graph capture is thread local
Graph capture is [initiated on a Cuda stream](https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__STREAM.html#group__CUDART__STREAM_1g793d7…
-
The detection of whether the object is an entity uses ``$this->_em->getMetadataFactory()->hasMetadataFor(ClassUtils::getClass($value))``. But ``hasMetadataFor`` checks whether the factory has **loaded…
-
I'm having trouble exporting the `Helsinki-NLP/opus-mt-es-en` model for language translation into the optimised OpenVino IR format. Reading through the other issues within this repository highlighted …
-
## Description
As I know, the optimizer decides the num_update according to its _index_update_count saved on each device, which means that If the trainer states on one GPU device and loaded into anot…
-
### 🐛 Describe the bug
NOTE: we are only interested in compiling the decoder, the encoder is also shown in the time traces but should be ignored.
I've been trying for a long time to get `torch.c…
-
Starting point: https://github.com/microsoft/DeepSpeed/issues/966
Test matrix
1. gradient accumulation: one vs many
2. #gpus: one vs many
3. stages: 1 vs 2 vs 3
4. dtype: bf16 vs fp16 vs fp32
…
-
### Reminder
- [X] I have read the README and searched the existing issues.
### System Info
when I just add one line in the `examples/extras/adam_mini/qwen2_full_sft.yaml` got a error below.
```…
-
Hi!
This is a really great tool and it's been fun using it.
I am trying to train the model 'bert-base-multilingual-uncased' using a tokenized dataset in the correct format. But every time I run the…
ghost updated
2 years ago
-
**Describe the feature**
提供多种损失函数的sft训练,比如对比损失
**Paste any useful information**
sft时,除了交叉熵损失,有时需要针对某个特定token计算对比损失、pairloss等等,可否集成这样一个功能呢?
**Additional context**
-
I’ve been using BIGDL-LM to accelerate the chatglm3-6b model. However, I’m curious about the speed. Is the current speed considered normal?
Here are the hardware details:
+ Graphics Card: Intel Corp…