Upload ONNX weights
#1
by
Xenova
HF staff
- opened
- config.json +6 -0
- onnx/model.onnx +3 -0
- onnx/model_fp16.onnx +3 -0
- onnx/model_q4.onnx +3 -0
- onnx/model_q4f16.onnx +3 -0
config.json
CHANGED
|
@@ -27,6 +27,12 @@
|
|
| 27 |
"tie_word_embeddings": true,
|
| 28 |
"torch_dtype": "bfloat16",
|
| 29 |
"transformers_version": "4.42.3",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
"use_cache": true,
|
| 31 |
"vocab_size": 49152
|
| 32 |
}
|
|
|
|
| 27 |
"tie_word_embeddings": true,
|
| 28 |
"torch_dtype": "bfloat16",
|
| 29 |
"transformers_version": "4.42.3",
|
| 30 |
+
"transformers.js_config": {
|
| 31 |
+
"kv_cache_dtype": {
|
| 32 |
+
"q4f16": "float16",
|
| 33 |
+
"fp16": "float16"
|
| 34 |
+
}
|
| 35 |
+
},
|
| 36 |
"use_cache": true,
|
| 37 |
"vocab_size": 49152
|
| 38 |
}
|
onnx/model.onnx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:761b9a4d535baafef46c86546942dc6e14f869bdefb7c85f07ec390f2167faeb
|
| 3 |
+
size 653920749
|
onnx/model_fp16.onnx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7c2af0dc2059e925c9408e8a9aac1d901c4d8db9ce3a1c2ade10966b35d2c40c
|
| 3 |
+
size 327218892
|
onnx/model_q4.onnx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3b32d1d5a31364b71e843a61493df65691c267e79890225788def79e2c16148c
|
| 3 |
+
size 199708970
|
onnx/model_q4f16.onnx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:33cb1d22cd714e37c998deba351c9233f7f2dac5bb4a155bbe0811a6e6eb3553
|
| 3 |
+
size 133950950
|