Hugging Face's logo

microsoft
/
Phi-3-vision-128k-instruct-onnx-cuda

Text Generation
Transformers
ONNX
phi3_v
ONNX
DML
ONNXRuntime
phi3
custom_code
Model card Files Community
3
kvaishnavi commited on
Commit
1f94971
•
1 Parent(s): 3dd613b

Fix EOS token id

Browse files
Files changed (1) hide show
  1. cuda-int4-rtn-block-32/genai_config.json +73 -78
cuda-int4-rtn-block-32/genai_config.json CHANGED
@@ -1,79 +1,74 @@
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
- "cuda": {
11
- "enable_cuda_graph": "0"
12
- }
13
- }
14
- ]
15
- },
16
- "filename": "phi-3-v-128k-instruct-text.onnx",
17
- "head_size": 96,
18
- "hidden_size": 3072,
19
- "inputs": {
20
- "inputs_embeds": "inputs_embeds",
21
- "attention_mask": "attention_mask",
22
- "past_key_names": "past_key_values.%d.key",
23
- "past_value_names": "past_key_values.%d.value"
24
- },
25
- "outputs": {
26
- "logits": "logits",
27
- "present_key_names": "present.%d.key",
28
- "present_value_names": "present.%d.value"
29
- },
30
- "num_attention_heads": 32,
31
- "num_hidden_layers": 32,
32
- "num_key_value_heads": 32
33
- },
34
- "embedding": {
35
- "filename": "phi-3-v-128k-instruct-text-embedding.onnx",
36
- "inputs": {
37
- "input_ids": "input_ids"
38
- },
39
- "outputs": {
40
- "inputs_embeds": "inputs_embeds"
41
- }
42
- },
43
- "vision": {
44
- "filename": "phi-3-v-128k-instruct-vision.onnx",
45
- "inputs": {
46
- "pixel_values": "pixel_values",
47
- "image_sizes": "image_sizes"
48
- },
49
- "outputs": {
50
- "visual_features": "visual_features"
51
- }
52
- },
53
- "eos_token_id": [
54
- 2,
55
- 32000,
56
- 32001,
57
- 32007
58
- ],
59
- "pad_token_id": 32000,
60
- "type": "phi3v",
61
- "vocab_size": 32064
62
- },
63
- "search": {
64
- "diversity_penalty": 0.0,
65
- "do_sample": false,
66
- "early_stopping": true,
67
- "length_penalty": 1.0,
68
- "max_length": 131072,
69
- "min_length": 0,
70
- "no_repeat_ngram_size": 0,
71
- "num_beams": 1,
72
- "num_return_sequences": 1,
73
- "past_present_share_buffer": true,
74
- "repetition_penalty": 1.0,
75
- "temperature": 1.0,
76
- "top_k": 1,
77
- "top_p": 1.0
78
- }
79
  }
 
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
+ "cuda": {
11
+ "enable_cuda_graph": "0"
12
+ }
13
+ }
14
+ ]
15
+ },
16
+ "filename": "phi-3-v-128k-instruct-text.onnx",
17
+ "head_size": 96,
18
+ "hidden_size": 3072,
19
+ "inputs": {
20
+ "inputs_embeds": "inputs_embeds",
21
+ "attention_mask": "attention_mask",
22
+ "past_key_names": "past_key_values.%d.key",
23
+ "past_value_names": "past_key_values.%d.value"
24
+ },
25
+ "outputs": {
26
+ "logits": "logits",
27
+ "present_key_names": "present.%d.key",
28
+ "present_value_names": "present.%d.value"
29
+ },
30
+ "num_attention_heads": 32,
31
+ "num_hidden_layers": 32,
32
+ "num_key_value_heads": 32
33
+ },
34
+ "embedding": {
35
+ "filename": "phi-3-v-128k-instruct-text-embedding.onnx",
36
+ "inputs": {
37
+ "input_ids": "input_ids"
38
+ },
39
+ "outputs": {
40
+ "inputs_embeds": "inputs_embeds"
41
+ }
42
+ },
43
+ "vision": {
44
+ "filename": "phi-3-v-128k-instruct-vision.onnx",
45
+ "inputs": {
46
+ "pixel_values": "pixel_values",
47
+ "image_sizes": "image_sizes"
48
+ },
49
+ "outputs": {
50
+ "visual_features": "visual_features"
51
+ }
52
+ },
53
+ "eos_token_id": 32007,
54
+ "pad_token_id": 32000,
55
+ "type": "phi3v",
56
+ "vocab_size": 32064
57
+ },
58
+ "search": {
59
+ "diversity_penalty": 0.0,
60
+ "do_sample": false,
61
+ "early_stopping": true,
62
+ "length_penalty": 1.0,
63
+ "max_length": 131072,
64
+ "min_length": 0,
65
+ "no_repeat_ngram_size": 0,
66
+ "num_beams": 1,
67
+ "num_return_sequences": 1,
68
+ "past_present_share_buffer": true,
69
+ "repetition_penalty": 1.0,
70
+ "temperature": 1.0,
71
+ "top_k": 1,
72
+ "top_p": 1.0
73
+ }
 
 
 
 
 
74
  }