Add precision error note
Browse files
README.md
CHANGED
|
@@ -15,4 +15,10 @@ base_model:
|
|
| 15 |
> This repository contains models that have been converted to the GGUF format with various quantizations from an IBM Granite base model.
|
| 16 |
>
|
| 17 |
> Please reference the base model's full model card here:
|
| 18 |
-
> https://huggingface.co/ibm-granite/granite-4.0-1b
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
> This repository contains models that have been converted to the GGUF format with various quantizations from an IBM Granite base model.
|
| 16 |
>
|
| 17 |
> Please reference the base model's full model card here:
|
| 18 |
+
> https://huggingface.co/ibm-granite/granite-4.0-1b
|
| 19 |
+
|
| 20 |
+
### Known Issues
|
| 21 |
+
|
| 22 |
+
This model often uses the full numerical range of a 32-bit float (`f32`), so variants with smaller numerical ranges may run into precision errors at inference. The `F16` variant is known to fail on many hardware combinations.
|
| 23 |
+
|
| 24 |
+
**The recommended full-precision variant is `bf16`**.
|