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 sha256:1d6f8ed23d9b947a0f415630dc9848156677834f5353666fce8510dba62cea2d
|
| 3 |
+
size 1638675174
|
onnx/model_fp16.onnx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0e54d13b44aa0457cc2a50815f5b47e39f20bce3a6aab3d954280042b38003ec
|
| 3 |
+
size 819612409
|
onnx/model_q4.onnx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bcfd45cdfdc459425a306afb360228e9bbdf9b7ce8954f8f8097964662cf89f8
|
| 3 |
+
size 417376345
|
onnx/model_q4f16.onnx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:027d0926afdd99a4d214dcf13026032ba56e9926b77564d21b94d30de752253f
|
| 3 |
+
size 299634248
|