Commit
•
60ee357
1
Parent(s):
6ef5cee
Remove hmp from config and README (#1)
Browse files- Remove hmp from config and README (97dafc920ab239d70134771e9ea8dc3b4cd0e49d)
- Update README.md (9a03693df5b9117733f36e331a78dd78c8be8f07)
Co-authored-by: Jan Wieczorek <[email protected]>
- README.md +3 -6
- gaudi_config.json +1 -46
README.md
CHANGED
|
@@ -13,16 +13,13 @@ This model only contains the `GaudiConfig` file for running **Stable Diffusion v
|
|
| 13 |
**This model contains no model weights, only a GaudiConfig.**
|
| 14 |
|
| 15 |
This enables to specify:
|
| 16 |
-
- `
|
| 17 |
-
- `hmp_opt_level`: optimization level for HMP, see [here](https://docs.habana.ai/en/latest/PyTorch/PyTorch_Mixed_Precision/PT_Mixed_Precision.html#configuration-options) for a detailed explanation
|
| 18 |
-
- `hmp_bf16_ops`: list of operators that should run in bf16
|
| 19 |
-
- `hmp_fp32_ops`: list of operators that should run in fp32
|
| 20 |
-
- `hmp_is_verbose`: verbosity
|
| 21 |
|
| 22 |
## Usage
|
| 23 |
|
| 24 |
The `GaudiStableDiffusionPipeline` (`GaudiDDIMScheduler`) is instantiated the same way as the `StableDiffusionPipeline` (`DDIMScheduler`) in the 🤗 Diffusers library.
|
| 25 |
-
The only difference is that there are a few new training arguments specific to HPUs
|
|
|
|
| 26 |
|
| 27 |
Here is an example with one prompt:
|
| 28 |
```python
|
|
|
|
| 13 |
**This model contains no model weights, only a GaudiConfig.**
|
| 14 |
|
| 15 |
This enables to specify:
|
| 16 |
+
- `use_torch_autocast`: whether to use Torch Autocast for managing mixed precision
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
## Usage
|
| 19 |
|
| 20 |
The `GaudiStableDiffusionPipeline` (`GaudiDDIMScheduler`) is instantiated the same way as the `StableDiffusionPipeline` (`DDIMScheduler`) in the 🤗 Diffusers library.
|
| 21 |
+
The only difference is that there are a few new training arguments specific to HPUs.\
|
| 22 |
+
It is strongly recommended to train this model doing bf16 mixed-precision training for optimal performance and accuracy.
|
| 23 |
|
| 24 |
Here is an example with one prompt:
|
| 25 |
```python
|
gaudi_config.json
CHANGED
|
@@ -1,50 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"use_habana_mixed_precision": true,
|
| 3 |
-
"hmp_is_verbose": false,
|
| 4 |
"use_fused_adam": true,
|
| 5 |
"use_fused_clip_norm": true,
|
| 6 |
-
"
|
| 7 |
-
"addmm",
|
| 8 |
-
"batch_norm",
|
| 9 |
-
"bmm",
|
| 10 |
-
"conv1d",
|
| 11 |
-
"conv2d",
|
| 12 |
-
"conv3d",
|
| 13 |
-
"conv_transpose1d",
|
| 14 |
-
"conv_transpose2d",
|
| 15 |
-
"conv_transpose3d",
|
| 16 |
-
"dot",
|
| 17 |
-
"dropout",
|
| 18 |
-
"dropout1d",
|
| 19 |
-
"dropout2d",
|
| 20 |
-
"dropout3d",
|
| 21 |
-
"group_norm",
|
| 22 |
-
"instance_norm",
|
| 23 |
-
"layer_norm",
|
| 24 |
-
"leaky_relu",
|
| 25 |
-
"linear",
|
| 26 |
-
"matmul",
|
| 27 |
-
"mean",
|
| 28 |
-
"mm",
|
| 29 |
-
"mv",
|
| 30 |
-
"relu",
|
| 31 |
-
"t"
|
| 32 |
-
],
|
| 33 |
-
"hmp_fp32_ops": [
|
| 34 |
-
"binary_cross_entropy",
|
| 35 |
-
"binary_cross_entropy_with_logits",
|
| 36 |
-
"cross_entropy",
|
| 37 |
-
"div",
|
| 38 |
-
"divide",
|
| 39 |
-
"embedding",
|
| 40 |
-
"embedding_bag",
|
| 41 |
-
"log",
|
| 42 |
-
"log2",
|
| 43 |
-
"log_softmax",
|
| 44 |
-
"nll_loss",
|
| 45 |
-
"smooth_l1_loss",
|
| 46 |
-
"softmax",
|
| 47 |
-
"topk",
|
| 48 |
-
"truediv"
|
| 49 |
-
]
|
| 50 |
}
|
|
|
|
| 1 |
{
|
|
|
|
|
|
|
| 2 |
"use_fused_adam": true,
|
| 3 |
"use_fused_clip_norm": true,
|
| 4 |
+
"use_torch_autocast": true
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
}
|