Commit ·
b3368c8
1
Parent(s): 82e2ba3
Add config.json for QLoRA variants
Browse files- llama-3.2-1B/QLoRA/config.json +40 -0
- llama-3.2-3B/QLoRA/config.json +40 -0
llama-3.2-1B/QLoRA/config.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"$schema": "https://huggingface.co/software-mansion/react-native-executorch-spec/resolve/main/config.schema.json",
|
| 3 |
+
"model": "llama_3_2",
|
| 4 |
+
"family": "llama",
|
| 5 |
+
"capabilities": [
|
| 6 |
+
"text-generation"
|
| 7 |
+
],
|
| 8 |
+
"backend": "xnnpack",
|
| 9 |
+
"license": "llama-3.2-community",
|
| 10 |
+
"variants": [
|
| 11 |
+
{
|
| 12 |
+
"file": "llama3_2_qat_lora.pte",
|
| 13 |
+
"precision": "qat_lora",
|
| 14 |
+
"quantized": true,
|
| 15 |
+
"default": true,
|
| 16 |
+
"methods": {
|
| 17 |
+
"enable_dynamic_shape": true,
|
| 18 |
+
"get_bos_id": 128000,
|
| 19 |
+
"get_eos_ids": [
|
| 20 |
+
128009,
|
| 21 |
+
128001
|
| 22 |
+
],
|
| 23 |
+
"get_max_context_len": 2048,
|
| 24 |
+
"get_max_seq_len": 2048,
|
| 25 |
+
"get_n_layers": null,
|
| 26 |
+
"get_vocab_size": 128256,
|
| 27 |
+
"use_kv_cache": true,
|
| 28 |
+
"use_sdpa_with_kv_cache": true,
|
| 29 |
+
"forward": {
|
| 30 |
+
"inputs": [],
|
| 31 |
+
"outputs": []
|
| 32 |
+
}
|
| 33 |
+
}
|
| 34 |
+
}
|
| 35 |
+
],
|
| 36 |
+
"size": "1b",
|
| 37 |
+
"_methods_needing_shapes": [
|
| 38 |
+
"forward"
|
| 39 |
+
]
|
| 40 |
+
}
|
llama-3.2-3B/QLoRA/config.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"$schema": "https://huggingface.co/software-mansion/react-native-executorch-spec/resolve/main/config.schema.json",
|
| 3 |
+
"model": "llama_3_2",
|
| 4 |
+
"family": "llama",
|
| 5 |
+
"capabilities": [
|
| 6 |
+
"text-generation"
|
| 7 |
+
],
|
| 8 |
+
"backend": "xnnpack",
|
| 9 |
+
"license": "llama-3.2-community",
|
| 10 |
+
"variants": [
|
| 11 |
+
{
|
| 12 |
+
"file": "llama3_2-3B_qat_lora.pte",
|
| 13 |
+
"precision": "qat_lora",
|
| 14 |
+
"quantized": true,
|
| 15 |
+
"default": true,
|
| 16 |
+
"methods": {
|
| 17 |
+
"enable_dynamic_shape": true,
|
| 18 |
+
"get_bos_id": 128000,
|
| 19 |
+
"get_eos_ids": [
|
| 20 |
+
128009,
|
| 21 |
+
128001
|
| 22 |
+
],
|
| 23 |
+
"get_max_context_len": 2048,
|
| 24 |
+
"get_max_seq_len": 2048,
|
| 25 |
+
"get_n_layers": null,
|
| 26 |
+
"get_vocab_size": 128256,
|
| 27 |
+
"use_kv_cache": true,
|
| 28 |
+
"use_sdpa_with_kv_cache": true,
|
| 29 |
+
"forward": {
|
| 30 |
+
"inputs": [],
|
| 31 |
+
"outputs": []
|
| 32 |
+
}
|
| 33 |
+
}
|
| 34 |
+
}
|
| 35 |
+
],
|
| 36 |
+
"size": "3b",
|
| 37 |
+
"_methods_needing_shapes": [
|
| 38 |
+
"forward"
|
| 39 |
+
]
|
| 40 |
+
}
|