Update README.md
#23
by
rombodawg
- opened
README.md
CHANGED
|
@@ -5,6 +5,20 @@ tags:
|
|
| 5 |
- jamba
|
| 6 |
- mamba
|
| 7 |
- moe
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
---
|
| 9 |
|
| 10 |
# Model Card for Jamba
|
|
@@ -58,6 +72,7 @@ print(tokenizer.batch_decode(outputs))
|
|
| 58 |
```
|
| 59 |
|
| 60 |
<details>
|
|
|
|
| 61 |
<summary><strong>Loading the model in half precision</strong></summary>
|
| 62 |
|
| 63 |
The published checkpoint is saved in BF16. In order to load it into RAM in BF16/FP16, you need to specify `torch_dtype`:
|
|
|
|
| 5 |
- jamba
|
| 6 |
- mamba
|
| 7 |
- moe
|
| 8 |
+
model-index:
|
| 9 |
+
- name: Jamba-v0.1
|
| 10 |
+
results:
|
| 11 |
+
- task:
|
| 12 |
+
type: text-generation
|
| 13 |
+
dataset:
|
| 14 |
+
type: openai_humaneval
|
| 15 |
+
name: HumanEval
|
| 16 |
+
metrics:
|
| 17 |
+
- name: pass@1
|
| 18 |
+
type: pass@1
|
| 19 |
+
value: 0.329
|
| 20 |
+
|
| 21 |
+
verified: false
|
| 22 |
---
|
| 23 |
|
| 24 |
# Model Card for Jamba
|
|
|
|
| 72 |
```
|
| 73 |
|
| 74 |
<details>
|
| 75 |
+
|
| 76 |
<summary><strong>Loading the model in half precision</strong></summary>
|
| 77 |
|
| 78 |
The published checkpoint is saved in BF16. In order to load it into RAM in BF16/FP16, you need to specify `torch_dtype`:
|