Remove variants from main mistral config
Browse files
inference-cache-config/mistral.json
CHANGED
|
@@ -36,57 +36,5 @@
|
|
| 36 |
"num_cores": 8,
|
| 37 |
"auto_cast_type": "bf16"
|
| 38 |
}
|
| 39 |
-
],
|
| 40 |
-
"HuggingFaceH4/zephyr-7b-beta": [
|
| 41 |
-
{
|
| 42 |
-
"batch_size": 1,
|
| 43 |
-
"sequence_length": 4096,
|
| 44 |
-
"num_cores": 2,
|
| 45 |
-
"auto_cast_type": "bf16"
|
| 46 |
-
},
|
| 47 |
-
{
|
| 48 |
-
"batch_size": 1,
|
| 49 |
-
"sequence_length": 4096,
|
| 50 |
-
"num_cores": 8,
|
| 51 |
-
"auto_cast_type": "bf16"
|
| 52 |
-
},
|
| 53 |
-
{
|
| 54 |
-
"batch_size": 4,
|
| 55 |
-
"sequence_length": 4096,
|
| 56 |
-
"num_cores": 2,
|
| 57 |
-
"auto_cast_type": "bf16"
|
| 58 |
-
},
|
| 59 |
-
{
|
| 60 |
-
"batch_size": 4,
|
| 61 |
-
"sequence_length": 4096,
|
| 62 |
-
"num_cores": 8,
|
| 63 |
-
"auto_cast_type": "bf16"
|
| 64 |
-
},
|
| 65 |
-
{
|
| 66 |
-
"batch_size": 8,
|
| 67 |
-
"sequence_length": 4096,
|
| 68 |
-
"num_cores": 8,
|
| 69 |
-
"auto_cast_type": "bf16"
|
| 70 |
-
},
|
| 71 |
-
{
|
| 72 |
-
"batch_size": 16,
|
| 73 |
-
"sequence_length": 4096,
|
| 74 |
-
"num_cores": 8,
|
| 75 |
-
"auto_cast_type": "bf16"
|
| 76 |
-
}
|
| 77 |
-
],
|
| 78 |
-
"teknium/OpenHermes-2.5-Mistral-7B": [
|
| 79 |
-
{
|
| 80 |
-
"batch_size": 1,
|
| 81 |
-
"sequence_length": 4096,
|
| 82 |
-
"num_cores": 2,
|
| 83 |
-
"auto_cast_type": "bf16"
|
| 84 |
-
},
|
| 85 |
-
{
|
| 86 |
-
"batch_size": 4,
|
| 87 |
-
"sequence_length": 4096,
|
| 88 |
-
"num_cores": 2,
|
| 89 |
-
"auto_cast_type": "bf16"
|
| 90 |
-
}
|
| 91 |
]
|
| 92 |
}
|
|
|
|
| 36 |
"num_cores": 8,
|
| 37 |
"auto_cast_type": "bf16"
|
| 38 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
]
|
| 40 |
}
|