Add Zephyr to mistral variants
Browse files
inference-cache-config/mistral-variants.json
CHANGED
|
@@ -1,5 +1,43 @@
|
|
| 1 |
{
|
| 2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
{
|
| 4 |
"batch_size": 1,
|
| 5 |
"sequence_length": 4096,
|
|
@@ -13,7 +51,7 @@
|
|
| 13 |
"auto_cast_type": "bf16"
|
| 14 |
}
|
| 15 |
],
|
| 16 |
-
|
| 17 |
{
|
| 18 |
"batch_size": 1,
|
| 19 |
"sequence_length": 4096,
|
|
@@ -27,7 +65,7 @@
|
|
| 27 |
"auto_cast_type": "bf16"
|
| 28 |
}
|
| 29 |
],
|
| 30 |
-
|
| 31 |
{
|
| 32 |
"batch_size": 1,
|
| 33 |
"sequence_length": 4096,
|
|
@@ -41,7 +79,7 @@
|
|
| 41 |
"auto_cast_type": "bf16"
|
| 42 |
}
|
| 43 |
],
|
| 44 |
-
|
| 45 |
{
|
| 46 |
"batch_size": 1,
|
| 47 |
"sequence_length": 4096,
|
|
@@ -55,7 +93,7 @@
|
|
| 55 |
"auto_cast_type": "bf16"
|
| 56 |
}
|
| 57 |
],
|
| 58 |
-
|
| 59 |
{
|
| 60 |
"batch_size": 1,
|
| 61 |
"sequence_length": 4096,
|
|
|
|
| 1 |
{
|
| 2 |
+
"HuggingFaceH4/zephyr-7b-beta": [
|
| 3 |
+
{
|
| 4 |
+
"batch_size": 1,
|
| 5 |
+
"sequence_length": 4096,
|
| 6 |
+
"num_cores": 2,
|
| 7 |
+
"auto_cast_type": "bf16"
|
| 8 |
+
},
|
| 9 |
+
{
|
| 10 |
+
"batch_size": 1,
|
| 11 |
+
"sequence_length": 4096,
|
| 12 |
+
"num_cores": 8,
|
| 13 |
+
"auto_cast_type": "bf16"
|
| 14 |
+
},
|
| 15 |
+
{
|
| 16 |
+
"batch_size": 4,
|
| 17 |
+
"sequence_length": 4096,
|
| 18 |
+
"num_cores": 2,
|
| 19 |
+
"auto_cast_type": "bf16"
|
| 20 |
+
},
|
| 21 |
+
{
|
| 22 |
+
"batch_size": 4,
|
| 23 |
+
"sequence_length": 4096,
|
| 24 |
+
"num_cores": 8,
|
| 25 |
+
"auto_cast_type": "bf16"
|
| 26 |
+
},
|
| 27 |
+
{
|
| 28 |
+
"batch_size": 8,
|
| 29 |
+
"sequence_length": 4096,
|
| 30 |
+
"num_cores": 8,
|
| 31 |
+
"auto_cast_type": "bf16"
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"batch_size": 16,
|
| 35 |
+
"sequence_length": 4096,
|
| 36 |
+
"num_cores": 8,
|
| 37 |
+
"auto_cast_type": "bf16"
|
| 38 |
+
}
|
| 39 |
+
],
|
| 40 |
+
"NousResearch/Genstruct-7B": [
|
| 41 |
{
|
| 42 |
"batch_size": 1,
|
| 43 |
"sequence_length": 4096,
|
|
|
|
| 51 |
"auto_cast_type": "bf16"
|
| 52 |
}
|
| 53 |
],
|
| 54 |
+
"BioMistral/BioMistral-7B": [
|
| 55 |
{
|
| 56 |
"batch_size": 1,
|
| 57 |
"sequence_length": 4096,
|
|
|
|
| 65 |
"auto_cast_type": "bf16"
|
| 66 |
}
|
| 67 |
],
|
| 68 |
+
"NousResearch/Hermes-2-Pro-Mistral-7B": [
|
| 69 |
{
|
| 70 |
"batch_size": 1,
|
| 71 |
"sequence_length": 4096,
|
|
|
|
| 79 |
"auto_cast_type": "bf16"
|
| 80 |
}
|
| 81 |
],
|
| 82 |
+
"NousResearch/Nous-Hermes-2-Mistral-7B-DPO": [
|
| 83 |
{
|
| 84 |
"batch_size": 1,
|
| 85 |
"sequence_length": 4096,
|
|
|
|
| 93 |
"auto_cast_type": "bf16"
|
| 94 |
}
|
| 95 |
],
|
| 96 |
+
"ibm/merlinite-7b": [
|
| 97 |
{
|
| 98 |
"batch_size": 1,
|
| 99 |
"sequence_length": 4096,
|