Upload 4 files
Browse files- .gitattributes +1 -0
- genai_config.json +49 -0
- llama3.onnx +3 -0
- llama3.onnx.data +3 -0
- t3_embedder.onnx +3 -0
.gitattributes
CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
36 |
+
llama3.onnx.data filter=lfs diff=lfs merge=lfs -text
|
genai_config.json
ADDED
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model": {
|
3 |
+
"bos_token_id": 1,
|
4 |
+
"context_length": 131072,
|
5 |
+
"decoder": {
|
6 |
+
"session_options": {
|
7 |
+
"log_id": "onnxruntime-genai",
|
8 |
+
"provider_options": []
|
9 |
+
},
|
10 |
+
"filename": "model.onnx",
|
11 |
+
"head_size": 64,
|
12 |
+
"hidden_size": 1024,
|
13 |
+
"inputs": {
|
14 |
+
"input_ids": "input_ids",
|
15 |
+
"attention_mask": "attention_mask",
|
16 |
+
"past_key_names": "past_key_values.%d.key",
|
17 |
+
"past_value_names": "past_key_values.%d.value"
|
18 |
+
},
|
19 |
+
"outputs": {
|
20 |
+
"logits": "logits",
|
21 |
+
"present_key_names": "present.%d.key",
|
22 |
+
"present_value_names": "present.%d.value"
|
23 |
+
},
|
24 |
+
"num_attention_heads": 16,
|
25 |
+
"num_hidden_layers": 30,
|
26 |
+
"num_key_value_heads": 16
|
27 |
+
},
|
28 |
+
"eos_token_id": 2,
|
29 |
+
"pad_token_id": 2,
|
30 |
+
"type": "llama",
|
31 |
+
"vocab_size": 8
|
32 |
+
},
|
33 |
+
"search": {
|
34 |
+
"diversity_penalty": 0.0,
|
35 |
+
"do_sample": false,
|
36 |
+
"early_stopping": true,
|
37 |
+
"length_penalty": 1.0,
|
38 |
+
"max_length": 131072,
|
39 |
+
"min_length": 0,
|
40 |
+
"no_repeat_ngram_size": 0,
|
41 |
+
"num_beams": 1,
|
42 |
+
"num_return_sequences": 1,
|
43 |
+
"past_present_share_buffer": true,
|
44 |
+
"repetition_penalty": 1.0,
|
45 |
+
"temperature": 1.0,
|
46 |
+
"top_k": 1,
|
47 |
+
"top_p": 1.0
|
48 |
+
}
|
49 |
+
}
|
llama3.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1ec159c6eb08380e73772c6c303271bf8d9bc8778d0dbee7e7f14a7e391bc420
|
3 |
+
size 169240
|
llama3.onnx.data
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e2767f6de0c43395c12ae61eca9a10c3b7b272c84b0c10e38586f8fa1e2a756e
|
3 |
+
size 2047148032
|
t3_embedder.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e653b840ed2b6bc9eef36dec95021ba426f3192c2957b14acf50418f69dc038c
|
3 |
+
size 79679724
|