Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
Community
optimum-internal-testing-user commited on
Commit
2ae1719
·
verified ·
1 Parent(s): 28745ea

Synchronizing local compiler cache.

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/4cb32492bb3736b0f508.json +48 -0
  2. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/b3a5049f78aa165f1949.json +48 -0
  3. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/ec2660c61019ca183223.json +48 -0
  4. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/gpt2/gpt2/bcf3b4c10abca97b8c19.json +48 -0
  5. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/1f43dd9159545f0247d5.json +45 -0
  6. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/588998ffbea59807bd3d.json +45 -0
  7. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/83ecbabedf314419f3ca.json +45 -0
  8. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/12398fc9bb3b90b69c6d.json +45 -0
  9. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/3159c7cc5b0c26793327.json +45 -0
  10. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/e068af6507bc381e1370.json +45 -0
  11. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/4ac2b0ca2f3012645cd2.json +40 -0
  12. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/66507451739e53a47483.json +40 -0
  13. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/ce2619dd085de8aadeff.json +40 -0
  14. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mixtral/dacorvo/Mixtral-tiny/3509514a093c68deeaca.json +41 -0
  15. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mixtral/dacorvo/Mixtral-tiny/7a7bef14f18dbabbd689.json +41 -0
  16. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mixtral/dacorvo/Mixtral-tiny/7f1476291e21a1e7a585.json +41 -0
  17. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/549b38a19302cd6ff930.json +39 -0
  18. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/6b898ea4e83643172eba.json +39 -0
  19. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/d1b1af5fd67eb0577ee7.json +39 -0
  20. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/phi3/yujiepan/phi-4-tiny-random/607ae698be806304fb50.json +42 -0
  21. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/phi3/yujiepan/phi-4-tiny-random/da1641bf50b2440d8d42.json +42 -0
  22. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/phi3/yujiepan/phi-4-tiny-random/ff5ccda1b163a504cfcc.json +42 -0
  23. neuronxcc-2.17.194.0+d312836f/MODULE_032f74178031f5ed3c74+613edded/model.neff +1 -1
  24. neuronxcc-2.17.194.0+d312836f/MODULE_0a88901e8c98f54e4c10+613edded/model.neff +1 -1
  25. neuronxcc-2.17.194.0+d312836f/MODULE_0cfdfd8c26d66b282d5a+613edded/model.neff +0 -0
  26. neuronxcc-2.17.194.0+d312836f/MODULE_1090c91454701eac92a9+613edded/model.neff +1 -1
  27. neuronxcc-2.17.194.0+d312836f/MODULE_113f3268e3fd4d66fe81+8a3305d3/model.neff +0 -0
  28. neuronxcc-2.17.194.0+d312836f/MODULE_151e013069d6b102df91+613edded/model.neff +1 -1
  29. neuronxcc-2.17.194.0+d312836f/MODULE_24ff9ac2787ce9a1d276+613edded/model.neff +1 -1
  30. neuronxcc-2.17.194.0+d312836f/MODULE_255bd8f176e49c261e93+613edded/model.neff +1 -1
  31. neuronxcc-2.17.194.0+d312836f/MODULE_272d7dabaeb48e6d7210+613edded/model.neff +1 -1
  32. neuronxcc-2.17.194.0+d312836f/MODULE_2b6914194b931d7496fc+613edded/model.neff +1 -1
  33. neuronxcc-2.17.194.0+d312836f/MODULE_2c290567bb51ca0069bf+613edded/model.neff +0 -0
  34. neuronxcc-2.17.194.0+d312836f/MODULE_320f2622d4d0c9fdd0f1+613edded/model.neff +1 -1
  35. neuronxcc-2.17.194.0+d312836f/MODULE_3aa61c04bd8d9ca8722a+613edded/model.neff +1 -1
  36. neuronxcc-2.17.194.0+d312836f/MODULE_3d7f2d2bef4f6fdd2c74+613edded/model.neff +1 -1
  37. neuronxcc-2.17.194.0+d312836f/MODULE_3e5f6b34247d2b457ec5+613edded/model.neff +1 -1
  38. neuronxcc-2.17.194.0+d312836f/MODULE_40f92bf9469aae653e93+613edded/model.neff +1 -1
  39. neuronxcc-2.17.194.0+d312836f/MODULE_48437cc74469a8ccaec8+613edded/model.neff +1 -1
  40. neuronxcc-2.17.194.0+d312836f/MODULE_48bfe9ceb9631fdca2d4+613edded/model.neff +1 -1
  41. neuronxcc-2.17.194.0+d312836f/MODULE_5a9cc28c7336fd9ea3ab+613edded/model.neff +1 -1
  42. neuronxcc-2.17.194.0+d312836f/MODULE_5de5ea910315ba9c0def+613edded/model.neff +1 -1
  43. neuronxcc-2.17.194.0+d312836f/MODULE_6819e64fb96e87ffece0+613edded/model.neff +1 -1
  44. neuronxcc-2.17.194.0+d312836f/MODULE_7062a76356ca462bcc78+613edded/model.neff +1 -1
  45. neuronxcc-2.17.194.0+d312836f/MODULE_878400bd48be9c48dd23+613edded/model.neff +0 -0
  46. neuronxcc-2.17.194.0+d312836f/MODULE_8bb2b91314df12f54a63+613edded/model.neff +1 -1
  47. neuronxcc-2.17.194.0+d312836f/MODULE_960d05b842008ddeae63+8a3305d3/model.neff +0 -0
  48. neuronxcc-2.17.194.0+d312836f/MODULE_99824fde40f6515f5b3d+613edded/model.neff +1 -1
  49. neuronxcc-2.17.194.0+d312836f/MODULE_9be95eab2454f5ade94b+613edded/model.neff +1 -1
  50. neuronxcc-2.17.194.0+d312836f/MODULE_a297084c31f9c1ff1d5d+613edded/model.neff +1 -1
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/4cb32492bb3736b0f508.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-BloomForCausalLM",
4
+ "_task": "text-generation",
5
+ "apply_residual_connection_post_layernorm": false,
6
+ "architectures": [
7
+ "BloomForCausalLM"
8
+ ],
9
+ "attention_dropout": 0.1,
10
+ "dtype": "float32",
11
+ "gradient_checkpointing": false,
12
+ "hidden_dropout": 0.1,
13
+ "hidden_size": 32,
14
+ "id2label": {
15
+ "0": "LABEL_0",
16
+ "1": "LABEL_1",
17
+ "2": "LABEL_2"
18
+ },
19
+ "initializer_range": 0.02,
20
+ "is_decoder": true,
21
+ "label2id": {
22
+ "LABEL_0": 0,
23
+ "LABEL_1": 1,
24
+ "LABEL_2": 2
25
+ },
26
+ "layer_norm_epsilon": 1e-05,
27
+ "model_type": "bloom",
28
+ "n_head": 4,
29
+ "n_layer": 5,
30
+ "n_positions": 512,
31
+ "neuron": {
32
+ "auto_cast_type": "fp32",
33
+ "batch_size": 1,
34
+ "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM",
35
+ "checkpoint_revision": "92b07e9b7b4f986fa7c54e2ac3b9201b4ba5212e",
36
+ "compiler_type": "neuronx-cc",
37
+ "compiler_version": "2.17.194.0+d312836f",
38
+ "num_cores": 2,
39
+ "sequence_length": 100,
40
+ "task": "text-generation"
41
+ },
42
+ "pretraining_tp": 1,
43
+ "seq_length": 7,
44
+ "slow_but_exact": true,
45
+ "type_vocab_size": 16,
46
+ "use_cache": true,
47
+ "vocab_size": 1024
48
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/b3a5049f78aa165f1949.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-BloomForCausalLM",
4
+ "_task": "text-generation",
5
+ "apply_residual_connection_post_layernorm": false,
6
+ "architectures": [
7
+ "BloomForCausalLM"
8
+ ],
9
+ "attention_dropout": 0.1,
10
+ "dtype": "float32",
11
+ "gradient_checkpointing": false,
12
+ "hidden_dropout": 0.1,
13
+ "hidden_size": 32,
14
+ "id2label": {
15
+ "0": "LABEL_0",
16
+ "1": "LABEL_1",
17
+ "2": "LABEL_2"
18
+ },
19
+ "initializer_range": 0.02,
20
+ "is_decoder": true,
21
+ "label2id": {
22
+ "LABEL_0": 0,
23
+ "LABEL_1": 1,
24
+ "LABEL_2": 2
25
+ },
26
+ "layer_norm_epsilon": 1e-05,
27
+ "model_type": "bloom",
28
+ "n_head": 4,
29
+ "n_layer": 5,
30
+ "n_positions": 512,
31
+ "neuron": {
32
+ "auto_cast_type": "fp16",
33
+ "batch_size": 2,
34
+ "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM",
35
+ "checkpoint_revision": "92b07e9b7b4f986fa7c54e2ac3b9201b4ba5212e",
36
+ "compiler_type": "neuronx-cc",
37
+ "compiler_version": "2.17.194.0+d312836f",
38
+ "num_cores": 2,
39
+ "sequence_length": 100,
40
+ "task": "text-generation"
41
+ },
42
+ "pretraining_tp": 1,
43
+ "seq_length": 7,
44
+ "slow_but_exact": true,
45
+ "type_vocab_size": 16,
46
+ "use_cache": true,
47
+ "vocab_size": 1024
48
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/ec2660c61019ca183223.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-BloomForCausalLM",
4
+ "_task": "text-generation",
5
+ "apply_residual_connection_post_layernorm": false,
6
+ "architectures": [
7
+ "BloomForCausalLM"
8
+ ],
9
+ "attention_dropout": 0.1,
10
+ "dtype": "float32",
11
+ "gradient_checkpointing": false,
12
+ "hidden_dropout": 0.1,
13
+ "hidden_size": 32,
14
+ "id2label": {
15
+ "0": "LABEL_0",
16
+ "1": "LABEL_1",
17
+ "2": "LABEL_2"
18
+ },
19
+ "initializer_range": 0.02,
20
+ "is_decoder": true,
21
+ "label2id": {
22
+ "LABEL_0": 0,
23
+ "LABEL_1": 1,
24
+ "LABEL_2": 2
25
+ },
26
+ "layer_norm_epsilon": 1e-05,
27
+ "model_type": "bloom",
28
+ "n_head": 4,
29
+ "n_layer": 5,
30
+ "n_positions": 512,
31
+ "neuron": {
32
+ "auto_cast_type": "fp16",
33
+ "batch_size": 1,
34
+ "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM",
35
+ "checkpoint_revision": "92b07e9b7b4f986fa7c54e2ac3b9201b4ba5212e",
36
+ "compiler_type": "neuronx-cc",
37
+ "compiler_version": "2.17.194.0+d312836f",
38
+ "num_cores": 2,
39
+ "sequence_length": 100,
40
+ "task": "text-generation"
41
+ },
42
+ "pretraining_tp": 1,
43
+ "seq_length": 7,
44
+ "slow_but_exact": true,
45
+ "type_vocab_size": 16,
46
+ "use_cache": true,
47
+ "vocab_size": 1024
48
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/gpt2/gpt2/bcf3b4c10abca97b8c19.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "gpt2",
4
+ "_task": "text-generation",
5
+ "activation_function": "gelu_new",
6
+ "architectures": [
7
+ "GPT2LMHeadModel"
8
+ ],
9
+ "attn_pdrop": 0.1,
10
+ "embd_pdrop": 0.1,
11
+ "initializer_range": 0.02,
12
+ "layer_norm_epsilon": 1e-05,
13
+ "model_type": "gpt2",
14
+ "n_ctx": 1024,
15
+ "n_embd": 768,
16
+ "n_head": 12,
17
+ "n_inner": null,
18
+ "n_layer": 12,
19
+ "n_positions": 1024,
20
+ "neuron": {
21
+ "auto_cast_type": "fp16",
22
+ "batch_size": 4,
23
+ "checkpoint_id": "gpt2",
24
+ "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e",
25
+ "compiler_type": "neuronx-cc",
26
+ "compiler_version": "2.17.194.0+d312836f",
27
+ "num_cores": 2,
28
+ "sequence_length": 1024,
29
+ "task": "text-generation"
30
+ },
31
+ "reorder_and_upcast_attn": false,
32
+ "resid_pdrop": 0.1,
33
+ "scale_attn_by_inverse_layer_idx": false,
34
+ "scale_attn_weights": true,
35
+ "summary_activation": null,
36
+ "summary_first_dropout": 0.1,
37
+ "summary_proj_to_labels": true,
38
+ "summary_type": "cls_index",
39
+ "summary_use_proj": true,
40
+ "task_specific_params": {
41
+ "text-generation": {
42
+ "do_sample": true,
43
+ "max_length": 50
44
+ }
45
+ },
46
+ "use_cache": true,
47
+ "vocab_size": 50257
48
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/1f43dd9159545f0247d5.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-gpt2",
4
+ "_task": "text-generation",
5
+ "activation_function": "gelu_new",
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "attn_pdrop": 0.1,
8
+ "embd_pdrop": 0.1,
9
+ "gradient_checkpointing": false,
10
+ "hidden_act": "gelu",
11
+ "hidden_dropout_prob": 0.1,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 37,
14
+ "layer_norm_epsilon": 1e-05,
15
+ "model_type": "gpt2",
16
+ "n_ctx": 512,
17
+ "n_embd": 32,
18
+ "n_head": 4,
19
+ "n_inner": null,
20
+ "n_layer": 5,
21
+ "n_positions": 512,
22
+ "neuron": {
23
+ "auto_cast_type": "fp16",
24
+ "batch_size": 2,
25
+ "checkpoint_id": "hf-internal-testing/tiny-random-gpt2",
26
+ "checkpoint_revision": "71034c5d8bde858ff824298bdedc65515b97d2b9",
27
+ "compiler_type": "neuronx-cc",
28
+ "compiler_version": "2.17.194.0+d312836f",
29
+ "num_cores": 2,
30
+ "sequence_length": 100,
31
+ "task": "text-generation"
32
+ },
33
+ "reorder_and_upcast_attn": false,
34
+ "resid_pdrop": 0.1,
35
+ "scale_attn_by_inverse_layer_idx": false,
36
+ "scale_attn_weights": true,
37
+ "summary_activation": null,
38
+ "summary_first_dropout": 0.1,
39
+ "summary_proj_to_labels": true,
40
+ "summary_type": "cls_index",
41
+ "summary_use_proj": true,
42
+ "type_vocab_size": 16,
43
+ "use_cache": true,
44
+ "vocab_size": 1000
45
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/588998ffbea59807bd3d.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-gpt2",
4
+ "_task": "text-generation",
5
+ "activation_function": "gelu_new",
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "attn_pdrop": 0.1,
8
+ "embd_pdrop": 0.1,
9
+ "gradient_checkpointing": false,
10
+ "hidden_act": "gelu",
11
+ "hidden_dropout_prob": 0.1,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 37,
14
+ "layer_norm_epsilon": 1e-05,
15
+ "model_type": "gpt2",
16
+ "n_ctx": 512,
17
+ "n_embd": 32,
18
+ "n_head": 4,
19
+ "n_inner": null,
20
+ "n_layer": 5,
21
+ "n_positions": 512,
22
+ "neuron": {
23
+ "auto_cast_type": "fp16",
24
+ "batch_size": 1,
25
+ "checkpoint_id": "hf-internal-testing/tiny-random-gpt2",
26
+ "checkpoint_revision": "71034c5d8bde858ff824298bdedc65515b97d2b9",
27
+ "compiler_type": "neuronx-cc",
28
+ "compiler_version": "2.17.194.0+d312836f",
29
+ "num_cores": 2,
30
+ "sequence_length": 100,
31
+ "task": "text-generation"
32
+ },
33
+ "reorder_and_upcast_attn": false,
34
+ "resid_pdrop": 0.1,
35
+ "scale_attn_by_inverse_layer_idx": false,
36
+ "scale_attn_weights": true,
37
+ "summary_activation": null,
38
+ "summary_first_dropout": 0.1,
39
+ "summary_proj_to_labels": true,
40
+ "summary_type": "cls_index",
41
+ "summary_use_proj": true,
42
+ "type_vocab_size": 16,
43
+ "use_cache": true,
44
+ "vocab_size": 1000
45
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/83ecbabedf314419f3ca.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-gpt2",
4
+ "_task": "text-generation",
5
+ "activation_function": "gelu_new",
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "attn_pdrop": 0.1,
8
+ "embd_pdrop": 0.1,
9
+ "gradient_checkpointing": false,
10
+ "hidden_act": "gelu",
11
+ "hidden_dropout_prob": 0.1,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 37,
14
+ "layer_norm_epsilon": 1e-05,
15
+ "model_type": "gpt2",
16
+ "n_ctx": 512,
17
+ "n_embd": 32,
18
+ "n_head": 4,
19
+ "n_inner": null,
20
+ "n_layer": 5,
21
+ "n_positions": 512,
22
+ "neuron": {
23
+ "auto_cast_type": "fp32",
24
+ "batch_size": 1,
25
+ "checkpoint_id": "hf-internal-testing/tiny-random-gpt2",
26
+ "checkpoint_revision": "71034c5d8bde858ff824298bdedc65515b97d2b9",
27
+ "compiler_type": "neuronx-cc",
28
+ "compiler_version": "2.17.194.0+d312836f",
29
+ "num_cores": 2,
30
+ "sequence_length": 100,
31
+ "task": "text-generation"
32
+ },
33
+ "reorder_and_upcast_attn": false,
34
+ "resid_pdrop": 0.1,
35
+ "scale_attn_by_inverse_layer_idx": false,
36
+ "scale_attn_weights": true,
37
+ "summary_activation": null,
38
+ "summary_first_dropout": 0.1,
39
+ "summary_proj_to_labels": true,
40
+ "summary_type": "cls_index",
41
+ "summary_use_proj": true,
42
+ "type_vocab_size": 16,
43
+ "use_cache": true,
44
+ "vocab_size": 1000
45
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/12398fc9bb3b90b69c6d.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "llamafactory/tiny-random-Llama-3",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "LlamaForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "head_dim": 4,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 16,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 64,
15
+ "max_position_embeddings": 131072,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "neuron": {
19
+ "auto_cast_type": "fp32",
20
+ "batch_size": 1,
21
+ "checkpoint_id": "llamafactory/tiny-random-Llama-3",
22
+ "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8",
23
+ "compiler_type": "neuronx-cc",
24
+ "compiler_version": "2.17.194.0+d312836f",
25
+ "num_cores": 2,
26
+ "sequence_length": 100,
27
+ "task": "text-generation"
28
+ },
29
+ "num_attention_heads": 4,
30
+ "num_hidden_layers": 2,
31
+ "num_key_value_heads": 4,
32
+ "pretraining_tp": 1,
33
+ "rms_norm_eps": 1e-05,
34
+ "rope_scaling": {
35
+ "factor": 8.0,
36
+ "high_freq_factor": 4.0,
37
+ "low_freq_factor": 1.0,
38
+ "original_max_position_embeddings": 8192,
39
+ "rope_type": "llama3"
40
+ },
41
+ "rope_theta": 500000.0,
42
+ "tie_word_embeddings": false,
43
+ "use_cache": true,
44
+ "vocab_size": 128256
45
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/3159c7cc5b0c26793327.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "llamafactory/tiny-random-Llama-3",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "LlamaForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "head_dim": 4,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 16,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 64,
15
+ "max_position_embeddings": 131072,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "neuron": {
19
+ "auto_cast_type": "fp16",
20
+ "batch_size": 2,
21
+ "checkpoint_id": "llamafactory/tiny-random-Llama-3",
22
+ "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8",
23
+ "compiler_type": "neuronx-cc",
24
+ "compiler_version": "2.17.194.0+d312836f",
25
+ "num_cores": 2,
26
+ "sequence_length": 100,
27
+ "task": "text-generation"
28
+ },
29
+ "num_attention_heads": 4,
30
+ "num_hidden_layers": 2,
31
+ "num_key_value_heads": 4,
32
+ "pretraining_tp": 1,
33
+ "rms_norm_eps": 1e-05,
34
+ "rope_scaling": {
35
+ "factor": 8.0,
36
+ "high_freq_factor": 4.0,
37
+ "low_freq_factor": 1.0,
38
+ "original_max_position_embeddings": 8192,
39
+ "rope_type": "llama3"
40
+ },
41
+ "rope_theta": 500000.0,
42
+ "tie_word_embeddings": false,
43
+ "use_cache": true,
44
+ "vocab_size": 128256
45
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/e068af6507bc381e1370.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "llamafactory/tiny-random-Llama-3",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "LlamaForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "head_dim": 4,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 16,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 64,
15
+ "max_position_embeddings": 131072,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "neuron": {
19
+ "auto_cast_type": "fp16",
20
+ "batch_size": 1,
21
+ "checkpoint_id": "llamafactory/tiny-random-Llama-3",
22
+ "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8",
23
+ "compiler_type": "neuronx-cc",
24
+ "compiler_version": "2.17.194.0+d312836f",
25
+ "num_cores": 2,
26
+ "sequence_length": 100,
27
+ "task": "text-generation"
28
+ },
29
+ "num_attention_heads": 4,
30
+ "num_hidden_layers": 2,
31
+ "num_key_value_heads": 4,
32
+ "pretraining_tp": 1,
33
+ "rms_norm_eps": 1e-05,
34
+ "rope_scaling": {
35
+ "factor": 8.0,
36
+ "high_freq_factor": 4.0,
37
+ "low_freq_factor": 1.0,
38
+ "original_max_position_embeddings": 8192,
39
+ "rope_type": "llama3"
40
+ },
41
+ "rope_theta": 500000.0,
42
+ "tie_word_embeddings": false,
43
+ "use_cache": true,
44
+ "vocab_size": 128256
45
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/4ac2b0ca2f3012645cd2.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "dacorvo/tiny-random-MistralForCausalLM",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "MistralForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "attention_probs_dropout_prob": 0.1,
10
+ "head_dim": 8,
11
+ "hidden_act": "gelu",
12
+ "hidden_dropout_prob": 0.1,
13
+ "hidden_size": 32,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 37,
16
+ "is_decoder": true,
17
+ "max_position_embeddings": 512,
18
+ "model_type": "mistral",
19
+ "neuron": {
20
+ "auto_cast_type": "fp16",
21
+ "batch_size": 1,
22
+ "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM",
23
+ "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a",
24
+ "compiler_type": "neuronx-cc",
25
+ "compiler_version": "2.17.194.0+d312836f",
26
+ "num_cores": 2,
27
+ "sequence_length": 100,
28
+ "task": "text-generation"
29
+ },
30
+ "num_attention_heads": 4,
31
+ "num_hidden_layers": 2,
32
+ "num_key_value_heads": 2,
33
+ "rms_norm_eps": 1e-06,
34
+ "rope_theta": 10000.0,
35
+ "sliding_window": 4096,
36
+ "tie_word_embeddings": false,
37
+ "type_vocab_size": 16,
38
+ "use_cache": true,
39
+ "vocab_size": 32000
40
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/66507451739e53a47483.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "dacorvo/tiny-random-MistralForCausalLM",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "MistralForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "attention_probs_dropout_prob": 0.1,
10
+ "head_dim": 8,
11
+ "hidden_act": "gelu",
12
+ "hidden_dropout_prob": 0.1,
13
+ "hidden_size": 32,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 37,
16
+ "is_decoder": true,
17
+ "max_position_embeddings": 512,
18
+ "model_type": "mistral",
19
+ "neuron": {
20
+ "auto_cast_type": "fp32",
21
+ "batch_size": 1,
22
+ "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM",
23
+ "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a",
24
+ "compiler_type": "neuronx-cc",
25
+ "compiler_version": "2.17.194.0+d312836f",
26
+ "num_cores": 2,
27
+ "sequence_length": 100,
28
+ "task": "text-generation"
29
+ },
30
+ "num_attention_heads": 4,
31
+ "num_hidden_layers": 2,
32
+ "num_key_value_heads": 2,
33
+ "rms_norm_eps": 1e-06,
34
+ "rope_theta": 10000.0,
35
+ "sliding_window": 4096,
36
+ "tie_word_embeddings": false,
37
+ "type_vocab_size": 16,
38
+ "use_cache": true,
39
+ "vocab_size": 32000
40
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/ce2619dd085de8aadeff.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "dacorvo/tiny-random-MistralForCausalLM",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "MistralForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "attention_probs_dropout_prob": 0.1,
10
+ "head_dim": 8,
11
+ "hidden_act": "gelu",
12
+ "hidden_dropout_prob": 0.1,
13
+ "hidden_size": 32,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 37,
16
+ "is_decoder": true,
17
+ "max_position_embeddings": 512,
18
+ "model_type": "mistral",
19
+ "neuron": {
20
+ "auto_cast_type": "fp16",
21
+ "batch_size": 2,
22
+ "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM",
23
+ "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a",
24
+ "compiler_type": "neuronx-cc",
25
+ "compiler_version": "2.17.194.0+d312836f",
26
+ "num_cores": 2,
27
+ "sequence_length": 100,
28
+ "task": "text-generation"
29
+ },
30
+ "num_attention_heads": 4,
31
+ "num_hidden_layers": 2,
32
+ "num_key_value_heads": 2,
33
+ "rms_norm_eps": 1e-06,
34
+ "rope_theta": 10000.0,
35
+ "sliding_window": 4096,
36
+ "tie_word_embeddings": false,
37
+ "type_vocab_size": 16,
38
+ "use_cache": true,
39
+ "vocab_size": 32000
40
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mixtral/dacorvo/Mixtral-tiny/3509514a093c68deeaca.json ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "dacorvo/Mixtral-tiny",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "MixtralForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "head_dim": 32,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 1024,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 3584,
14
+ "max_position_embeddings": 1024,
15
+ "model_type": "mixtral",
16
+ "neuron": {
17
+ "auto_cast_type": "fp16",
18
+ "batch_size": 1,
19
+ "checkpoint_id": "dacorvo/Mixtral-tiny",
20
+ "checkpoint_revision": "c557ba205ddff6ea911f4719e0d543d6c08356b6",
21
+ "compiler_type": "neuronx-cc",
22
+ "compiler_version": "2.17.194.0+d312836f",
23
+ "num_cores": 2,
24
+ "sequence_length": 100,
25
+ "task": "text-generation"
26
+ },
27
+ "num_attention_heads": 32,
28
+ "num_experts_per_tok": 2,
29
+ "num_hidden_layers": 2,
30
+ "num_key_value_heads": 8,
31
+ "num_local_experts": 8,
32
+ "output_router_logits": false,
33
+ "rms_norm_eps": 1e-05,
34
+ "rope_theta": 10000.0,
35
+ "router_aux_loss_coef": 0.001,
36
+ "router_jitter_noise": 0.0,
37
+ "sliding_window": 4096,
38
+ "tie_word_embeddings": false,
39
+ "use_cache": true,
40
+ "vocab_size": 32000
41
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mixtral/dacorvo/Mixtral-tiny/7a7bef14f18dbabbd689.json ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "dacorvo/Mixtral-tiny",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "MixtralForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "head_dim": 32,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 1024,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 3584,
14
+ "max_position_embeddings": 1024,
15
+ "model_type": "mixtral",
16
+ "neuron": {
17
+ "auto_cast_type": "fp16",
18
+ "batch_size": 2,
19
+ "checkpoint_id": "dacorvo/Mixtral-tiny",
20
+ "checkpoint_revision": "c557ba205ddff6ea911f4719e0d543d6c08356b6",
21
+ "compiler_type": "neuronx-cc",
22
+ "compiler_version": "2.17.194.0+d312836f",
23
+ "num_cores": 2,
24
+ "sequence_length": 100,
25
+ "task": "text-generation"
26
+ },
27
+ "num_attention_heads": 32,
28
+ "num_experts_per_tok": 2,
29
+ "num_hidden_layers": 2,
30
+ "num_key_value_heads": 8,
31
+ "num_local_experts": 8,
32
+ "output_router_logits": false,
33
+ "rms_norm_eps": 1e-05,
34
+ "rope_theta": 10000.0,
35
+ "router_aux_loss_coef": 0.001,
36
+ "router_jitter_noise": 0.0,
37
+ "sliding_window": 4096,
38
+ "tie_word_embeddings": false,
39
+ "use_cache": true,
40
+ "vocab_size": 32000
41
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/mixtral/dacorvo/Mixtral-tiny/7f1476291e21a1e7a585.json ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "dacorvo/Mixtral-tiny",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "MixtralForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "head_dim": 32,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 1024,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 3584,
14
+ "max_position_embeddings": 1024,
15
+ "model_type": "mixtral",
16
+ "neuron": {
17
+ "auto_cast_type": "fp32",
18
+ "batch_size": 1,
19
+ "checkpoint_id": "dacorvo/Mixtral-tiny",
20
+ "checkpoint_revision": "c557ba205ddff6ea911f4719e0d543d6c08356b6",
21
+ "compiler_type": "neuronx-cc",
22
+ "compiler_version": "2.17.194.0+d312836f",
23
+ "num_cores": 2,
24
+ "sequence_length": 100,
25
+ "task": "text-generation"
26
+ },
27
+ "num_attention_heads": 32,
28
+ "num_experts_per_tok": 2,
29
+ "num_hidden_layers": 2,
30
+ "num_key_value_heads": 8,
31
+ "num_local_experts": 8,
32
+ "output_router_logits": false,
33
+ "rms_norm_eps": 1e-05,
34
+ "rope_theta": 10000.0,
35
+ "router_aux_loss_coef": 0.001,
36
+ "router_jitter_noise": 0.0,
37
+ "sliding_window": 4096,
38
+ "tie_word_embeddings": false,
39
+ "use_cache": true,
40
+ "vocab_size": 32000
41
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/549b38a19302cd6ff930.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-OPTForCausalLM",
4
+ "_remove_final_layer_norm": false,
5
+ "_task": "text-generation",
6
+ "activation_function": "relu",
7
+ "architectures": [
8
+ "OPTForCausalLM"
9
+ ],
10
+ "attention_dropout": 0.1,
11
+ "do_layer_norm_before": true,
12
+ "dropout": 0.1,
13
+ "embed_dim": 16,
14
+ "enable_bias": true,
15
+ "ffn_dim": 4,
16
+ "hidden_size": 16,
17
+ "init_std": 0.02,
18
+ "is_decoder": true,
19
+ "layer_norm_elementwise_affine": true,
20
+ "layerdrop": 0.0,
21
+ "max_position_embeddings": 100,
22
+ "model_type": "opt",
23
+ "neuron": {
24
+ "auto_cast_type": "fp32",
25
+ "batch_size": 1,
26
+ "checkpoint_id": "hf-internal-testing/tiny-random-OPTForCausalLM",
27
+ "checkpoint_revision": "3f3f2ee1e499cb7ad89b877068684fdc9d9513c3",
28
+ "compiler_type": "neuronx-cc",
29
+ "compiler_version": "2.17.194.0+d312836f",
30
+ "num_cores": 2,
31
+ "sequence_length": 100,
32
+ "task": "text-generation"
33
+ },
34
+ "num_attention_heads": 4,
35
+ "num_hidden_layers": 5,
36
+ "use_cache": true,
37
+ "vocab_size": 50265,
38
+ "word_embed_proj_dim": 16
39
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/6b898ea4e83643172eba.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-OPTForCausalLM",
4
+ "_remove_final_layer_norm": false,
5
+ "_task": "text-generation",
6
+ "activation_function": "relu",
7
+ "architectures": [
8
+ "OPTForCausalLM"
9
+ ],
10
+ "attention_dropout": 0.1,
11
+ "do_layer_norm_before": true,
12
+ "dropout": 0.1,
13
+ "embed_dim": 16,
14
+ "enable_bias": true,
15
+ "ffn_dim": 4,
16
+ "hidden_size": 16,
17
+ "init_std": 0.02,
18
+ "is_decoder": true,
19
+ "layer_norm_elementwise_affine": true,
20
+ "layerdrop": 0.0,
21
+ "max_position_embeddings": 100,
22
+ "model_type": "opt",
23
+ "neuron": {
24
+ "auto_cast_type": "fp16",
25
+ "batch_size": 1,
26
+ "checkpoint_id": "hf-internal-testing/tiny-random-OPTForCausalLM",
27
+ "checkpoint_revision": "3f3f2ee1e499cb7ad89b877068684fdc9d9513c3",
28
+ "compiler_type": "neuronx-cc",
29
+ "compiler_version": "2.17.194.0+d312836f",
30
+ "num_cores": 2,
31
+ "sequence_length": 100,
32
+ "task": "text-generation"
33
+ },
34
+ "num_attention_heads": 4,
35
+ "num_hidden_layers": 5,
36
+ "use_cache": true,
37
+ "vocab_size": 50265,
38
+ "word_embed_proj_dim": 16
39
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/d1b1af5fd67eb0577ee7.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-OPTForCausalLM",
4
+ "_remove_final_layer_norm": false,
5
+ "_task": "text-generation",
6
+ "activation_function": "relu",
7
+ "architectures": [
8
+ "OPTForCausalLM"
9
+ ],
10
+ "attention_dropout": 0.1,
11
+ "do_layer_norm_before": true,
12
+ "dropout": 0.1,
13
+ "embed_dim": 16,
14
+ "enable_bias": true,
15
+ "ffn_dim": 4,
16
+ "hidden_size": 16,
17
+ "init_std": 0.02,
18
+ "is_decoder": true,
19
+ "layer_norm_elementwise_affine": true,
20
+ "layerdrop": 0.0,
21
+ "max_position_embeddings": 100,
22
+ "model_type": "opt",
23
+ "neuron": {
24
+ "auto_cast_type": "fp16",
25
+ "batch_size": 2,
26
+ "checkpoint_id": "hf-internal-testing/tiny-random-OPTForCausalLM",
27
+ "checkpoint_revision": "3f3f2ee1e499cb7ad89b877068684fdc9d9513c3",
28
+ "compiler_type": "neuronx-cc",
29
+ "compiler_version": "2.17.194.0+d312836f",
30
+ "num_cores": 2,
31
+ "sequence_length": 100,
32
+ "task": "text-generation"
33
+ },
34
+ "num_attention_heads": 4,
35
+ "num_hidden_layers": 5,
36
+ "use_cache": true,
37
+ "vocab_size": 50265,
38
+ "word_embed_proj_dim": 16
39
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/phi3/yujiepan/phi-4-tiny-random/607ae698be806304fb50.json ADDED
@@ -0,0 +1,42 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "yujiepan/phi-4-tiny-random",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Phi3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "auto_map": {},
11
+ "embd_pdrop": 0.0,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 16,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 32,
16
+ "max_position_embeddings": 16384,
17
+ "model_type": "phi3",
18
+ "neuron": {
19
+ "auto_cast_type": "fp16",
20
+ "batch_size": 2,
21
+ "checkpoint_id": "yujiepan/phi-4-tiny-random",
22
+ "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a",
23
+ "compiler_type": "neuronx-cc",
24
+ "compiler_version": "2.17.194.0+d312836f",
25
+ "num_cores": 2,
26
+ "sequence_length": 100,
27
+ "task": "text-generation"
28
+ },
29
+ "num_attention_heads": 2,
30
+ "num_hidden_layers": 2,
31
+ "num_key_value_heads": 1,
32
+ "original_max_position_embeddings": 16384,
33
+ "partial_rotary_factor": 1.0,
34
+ "resid_pdrop": 0.0,
35
+ "rms_norm_eps": 1e-05,
36
+ "rope_scaling": null,
37
+ "rope_theta": 250000,
38
+ "sliding_window": null,
39
+ "tie_word_embeddings": false,
40
+ "use_cache": true,
41
+ "vocab_size": 100352
42
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/phi3/yujiepan/phi-4-tiny-random/da1641bf50b2440d8d42.json ADDED
@@ -0,0 +1,42 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "yujiepan/phi-4-tiny-random",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Phi3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "auto_map": {},
11
+ "embd_pdrop": 0.0,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 16,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 32,
16
+ "max_position_embeddings": 16384,
17
+ "model_type": "phi3",
18
+ "neuron": {
19
+ "auto_cast_type": "fp32",
20
+ "batch_size": 1,
21
+ "checkpoint_id": "yujiepan/phi-4-tiny-random",
22
+ "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a",
23
+ "compiler_type": "neuronx-cc",
24
+ "compiler_version": "2.17.194.0+d312836f",
25
+ "num_cores": 2,
26
+ "sequence_length": 100,
27
+ "task": "text-generation"
28
+ },
29
+ "num_attention_heads": 2,
30
+ "num_hidden_layers": 2,
31
+ "num_key_value_heads": 1,
32
+ "original_max_position_embeddings": 16384,
33
+ "partial_rotary_factor": 1.0,
34
+ "resid_pdrop": 0.0,
35
+ "rms_norm_eps": 1e-05,
36
+ "rope_scaling": null,
37
+ "rope_theta": 250000,
38
+ "sliding_window": null,
39
+ "tie_word_embeddings": false,
40
+ "use_cache": true,
41
+ "vocab_size": 100352
42
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/phi3/yujiepan/phi-4-tiny-random/ff5ccda1b163a504cfcc.json ADDED
@@ -0,0 +1,42 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "yujiepan/phi-4-tiny-random",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Phi3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "auto_map": {},
11
+ "embd_pdrop": 0.0,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 16,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 32,
16
+ "max_position_embeddings": 16384,
17
+ "model_type": "phi3",
18
+ "neuron": {
19
+ "auto_cast_type": "fp16",
20
+ "batch_size": 1,
21
+ "checkpoint_id": "yujiepan/phi-4-tiny-random",
22
+ "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a",
23
+ "compiler_type": "neuronx-cc",
24
+ "compiler_version": "2.17.194.0+d312836f",
25
+ "num_cores": 2,
26
+ "sequence_length": 100,
27
+ "task": "text-generation"
28
+ },
29
+ "num_attention_heads": 2,
30
+ "num_hidden_layers": 2,
31
+ "num_key_value_heads": 1,
32
+ "original_max_position_embeddings": 16384,
33
+ "partial_rotary_factor": 1.0,
34
+ "resid_pdrop": 0.0,
35
+ "rms_norm_eps": 1e-05,
36
+ "rope_scaling": null,
37
+ "rope_theta": 250000,
38
+ "sliding_window": null,
39
+ "tie_word_embeddings": false,
40
+ "use_cache": true,
41
+ "vocab_size": 100352
42
+ }
neuronxcc-2.17.194.0+d312836f/MODULE_032f74178031f5ed3c74+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f41883d5e9092ebe55576d873702b5dffed1e27a8d65f84c8db1b278440ced50
3
  size 400384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e27746423329d5ca3648b0a521481cc8c82c4b3642f47ecab0a048a6baa1bf36
3
  size 400384
neuronxcc-2.17.194.0+d312836f/MODULE_0a88901e8c98f54e4c10+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3d7d2c8d2aca9540eef7a0363cc8fcd4c3dedbefb5369e9e43bc0b6631e620e0
3
  size 472064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:624d26061877fb366d7c41b3f88961284e96702d6313b52a33877d915813c796
3
  size 472064
neuronxcc-2.17.194.0+d312836f/MODULE_0cfdfd8c26d66b282d5a+613edded/model.neff CHANGED
Binary files a/neuronxcc-2.17.194.0+d312836f/MODULE_0cfdfd8c26d66b282d5a+613edded/model.neff and b/neuronxcc-2.17.194.0+d312836f/MODULE_0cfdfd8c26d66b282d5a+613edded/model.neff differ
 
neuronxcc-2.17.194.0+d312836f/MODULE_1090c91454701eac92a9+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c62df78a4a1e38116305fd0a445b154e251a9cc07dc31ecb2db18e2f512f8c0e
3
  size 615424
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f1f5d71caa3c2f2b27838c6aca987ca6cfdbad5860d42c62f5bfd5cafa39322a
3
  size 615424
neuronxcc-2.17.194.0+d312836f/MODULE_113f3268e3fd4d66fe81+8a3305d3/model.neff CHANGED
Binary files a/neuronxcc-2.17.194.0+d312836f/MODULE_113f3268e3fd4d66fe81+8a3305d3/model.neff and b/neuronxcc-2.17.194.0+d312836f/MODULE_113f3268e3fd4d66fe81+8a3305d3/model.neff differ
 
neuronxcc-2.17.194.0+d312836f/MODULE_151e013069d6b102df91+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a51fcb0a14a5be498a18c415670cf6841b2e7ad2c6bdc4b0cd999f222afa5213
3
  size 123904
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:29bb38b2d37442241496dc3e577d781d5701ea84a78bdf807cd9632cb3d4a479
3
  size 123904
neuronxcc-2.17.194.0+d312836f/MODULE_24ff9ac2787ce9a1d276+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:41e3db72ce26e8ba03eb8c3bd6b36a0ca1d6600fa86fd1e831baf1aef87545a8
3
  size 134144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cfeda11cc0f3c212245069cb451b881ea4ab7ec09173824aafd1932c6ddaa02b
3
  size 134144
neuronxcc-2.17.194.0+d312836f/MODULE_255bd8f176e49c261e93+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5ba93de455040e5a12ff2d7ca1787d9033a5af9b2e611f006cb354deb4ad0303
3
  size 123904
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8fb97c20b57d947ca3713b658629285b31e42dd91c75b8ae38dabf1bf1d0e0a5
3
  size 123904
neuronxcc-2.17.194.0+d312836f/MODULE_272d7dabaeb48e6d7210+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1bb40c1d1ceb3818584c2da4a0302801edf34d1ef4fbce6d7121dd420983be30
3
  size 123904
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4a1d4df13c1620cf66c80bf725c56bb6d381c95de7daa138096249a3b24105e5
3
  size 123904
neuronxcc-2.17.194.0+d312836f/MODULE_2b6914194b931d7496fc+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e2b4e21f87f0e86d7b8e3ca43aaecd881c890b116d0d2594381b44dec35dc794
3
  size 390144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b938f5ab46bc43919b7b6e6ee5f662dbe45a1e01b722f35caf06cc309bd9614
3
  size 390144
neuronxcc-2.17.194.0+d312836f/MODULE_2c290567bb51ca0069bf+613edded/model.neff CHANGED
Binary files a/neuronxcc-2.17.194.0+d312836f/MODULE_2c290567bb51ca0069bf+613edded/model.neff and b/neuronxcc-2.17.194.0+d312836f/MODULE_2c290567bb51ca0069bf+613edded/model.neff differ
 
neuronxcc-2.17.194.0+d312836f/MODULE_320f2622d4d0c9fdd0f1+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0a1b282a29fa8b5de9c398a9129365e47c7113cde8df4e977f248d3197caf24b
3
  size 144384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fd2fe87c0cecc2794f7c60fd83d1d0ec84e8c623c456709cc7b39b178be93dc2
3
  size 144384
neuronxcc-2.17.194.0+d312836f/MODULE_3aa61c04bd8d9ca8722a+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0ca9b438c97f9db8c70180cd35353b4f3eec15abe1d8a2b3977704b02dc34926
3
  size 144384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e23f8f5053a85ab214581a20d01be948511efa6c94de639ea9a0faf0e78fb43
3
  size 144384
neuronxcc-2.17.194.0+d312836f/MODULE_3d7f2d2bef4f6fdd2c74+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9bdd5c3e5d25551ab6199eb228376cf904709d3078ccdb90e439fb093c8e7bce
3
  size 134144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b91101d383cee1359c09f57a632e02c300b21e27a7295193cdeea85eb45ee1bb
3
  size 134144
neuronxcc-2.17.194.0+d312836f/MODULE_3e5f6b34247d2b457ec5+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2232a3000290a8daefe6f35989ae17fd67e8abe8f78eceae3f54c0a6a0bef321
3
  size 134144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8b2316f98ea92a283291c9a4274c380422be7c036fce4edeb9a50fe44bece16f
3
  size 134144
neuronxcc-2.17.194.0+d312836f/MODULE_40f92bf9469aae653e93+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ec758f8963cbb918f519b5b9bc8d8de328d3986ef987b35d27dfb37c6b3be486
3
  size 103424
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:79ec5411f21a49ce0ace6124438d98172ea326ed9726cf0829b59de6b9842f38
3
  size 103424
neuronxcc-2.17.194.0+d312836f/MODULE_48437cc74469a8ccaec8+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0a5ddd38c24ef4d267c8f50c4835455c2774d0fce8b5ae9f1e48266153adcabf
3
  size 154624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b1c68d1f3339c7c8578b3548a5cb94c40003af6ca357d051fe24348f19e0abd
3
  size 154624
neuronxcc-2.17.194.0+d312836f/MODULE_48bfe9ceb9631fdca2d4+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:df7012eae0855a49437751cefd3daf51c66d7ac893a3d050fa1d9fe5f5287cb4
3
  size 144384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f1a73c10b974b60e298a2c58c2913629564bf9746b1c62292fc461c41af22ff2
3
  size 144384
neuronxcc-2.17.194.0+d312836f/MODULE_5a9cc28c7336fd9ea3ab+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e3a6d256dc4709764db1aea3313be9fe99cecb8fead79d0002f8cd4900095684
3
  size 482304
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:740741c09a4afc3930fb12221f6fb36235b1de2e5035bce90142e801337869b6
3
  size 482304
neuronxcc-2.17.194.0+d312836f/MODULE_5de5ea910315ba9c0def+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5348979b49246a3e458b5952fcc16c0c6aedf8cf08fc488b40fb31e3a0718110
3
  size 134144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:976bc34408e232169b1ab7193866ead180cd97813d19797f28160bc2015f6782
3
  size 134144
neuronxcc-2.17.194.0+d312836f/MODULE_6819e64fb96e87ffece0+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:64aaa855b0b551030067dd1a2073c92948723acefe8c25896732e26bf558cf45
3
  size 144384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:23ef7eabe51185c0a807a040447dd75ac0d2fb7238e38f9d8245b4f71d713bd8
3
  size 144384
neuronxcc-2.17.194.0+d312836f/MODULE_7062a76356ca462bcc78+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7e082566d6ce3145d645b256f011898c4e72e1de5fbd9de9c672ab1aa799ac7e
3
  size 144384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1c4f650d174476fafcd6f2cd4d3001f5436ce1dfaa5f48cd9a95f05224883b73
3
  size 144384
neuronxcc-2.17.194.0+d312836f/MODULE_878400bd48be9c48dd23+613edded/model.neff CHANGED
Binary files a/neuronxcc-2.17.194.0+d312836f/MODULE_878400bd48be9c48dd23+613edded/model.neff and b/neuronxcc-2.17.194.0+d312836f/MODULE_878400bd48be9c48dd23+613edded/model.neff differ
 
neuronxcc-2.17.194.0+d312836f/MODULE_8bb2b91314df12f54a63+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b5aa5152fe7d42622da2409e928da288d1610b1178bf1c27d3a830968b35fc55
3
  size 400384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cf1903c941e671f0a24c5c263d723e718412ddc66654f2a03aee1f3f8d948156
3
  size 400384
neuronxcc-2.17.194.0+d312836f/MODULE_960d05b842008ddeae63+8a3305d3/model.neff CHANGED
Binary files a/neuronxcc-2.17.194.0+d312836f/MODULE_960d05b842008ddeae63+8a3305d3/model.neff and b/neuronxcc-2.17.194.0+d312836f/MODULE_960d05b842008ddeae63+8a3305d3/model.neff differ
 
neuronxcc-2.17.194.0+d312836f/MODULE_99824fde40f6515f5b3d+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fba423f40d603edaf212999fc1fff50343ce48a156de6b8cc0fd2ccf640f363d
3
  size 134144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f486b3d755c56de833246b62bf3bcdae5f44a49fee0b2df14b35fc9d740b05a
3
  size 134144
neuronxcc-2.17.194.0+d312836f/MODULE_9be95eab2454f5ade94b+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9b8940b8e6fe89c99511745586d1b555b239d2ec45286d6b941848453a7af9c2
3
  size 123904
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:13e6d8d70a234e00d1b734fdf7fe4c9c53dd6430f03c14c0c8b4a42ffd02d27c
3
  size 123904
neuronxcc-2.17.194.0+d312836f/MODULE_a297084c31f9c1ff1d5d+613edded/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:55ce511345939db432b50d42a2cd9b4799f3c9df0ad4cd34916c139cc24335c9
3
  size 134144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5461f58c99a25feea9959fe94dcd65b799cf940d700be5578d33e7da4a5418b1
3
  size 134144