Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
Community
dacorvo HF Staff commited on
Commit
6f1a1e4
·
verified ·
1 Parent(s): ccff189

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. .gitattributes +12 -0
  2. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/granite/hf-internal-testing/tiny-random-GraniteForCausalLM/28037758016041af399a.json +51 -0
  3. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/granite/hf-internal-testing/tiny-random-GraniteForCausalLM/32ec2689f2e2b8d58073.json +51 -0
  4. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/granite/hf-internal-testing/tiny-random-GraniteForCausalLM/3fdcfb75cf15e31ca9b5.json +51 -0
  5. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/llama/llamafactory/tiny-random-Llama-3/610dad84114ec035cc32.json +55 -0
  6. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/llama/llamafactory/tiny-random-Llama-3/b645ea39b56c5af10179.json +55 -0
  7. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/llama/llamafactory/tiny-random-Llama-3/b84d42926b5badee9c7a.json +55 -0
  8. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/phi3/yujiepan/phi-4-tiny-random/2e3b2c893e5dd12fa28c.json +52 -0
  9. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/phi3/yujiepan/phi-4-tiny-random/5f969dd5ec1310f92701.json +52 -0
  10. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/phi3/yujiepan/phi-4-tiny-random/84c8e75e4defdd9fa414.json +52 -0
  11. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/qwen2/Qwen/Qwen2.5-0.5B/2dcbb3f89df1561e38ca.json +49 -0
  12. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/qwen2/yujiepan/qwen2.5-128k-tiny-random/1fd1a2e4204366d503f6.json +53 -0
  13. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/qwen2/yujiepan/qwen2.5-128k-tiny-random/2a5b61ac8a8b5940982a.json +53 -0
  14. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/qwen2/yujiepan/qwen2.5-128k-tiny-random/b1b72b82166aaabf0a64.json +53 -0
  15. neuronxcc-2.17.194.0+d312836f/MODULE_1649fc77b87fff02e370+613edded/compile_flags.json +1 -0
  16. neuronxcc-2.17.194.0+d312836f/MODULE_1649fc77b87fff02e370+613edded/model.done +0 -0
  17. neuronxcc-2.17.194.0+d312836f/MODULE_1649fc77b87fff02e370+613edded/model.hlo_module.pb +3 -0
  18. neuronxcc-2.17.194.0+d312836f/MODULE_1649fc77b87fff02e370+613edded/model.neff +3 -0
  19. neuronxcc-2.17.194.0+d312836f/MODULE_22cf23062ec53b3fd95d+613edded/compile_flags.json +1 -0
  20. neuronxcc-2.17.194.0+d312836f/MODULE_22cf23062ec53b3fd95d+613edded/model.done +0 -0
  21. neuronxcc-2.17.194.0+d312836f/MODULE_22cf23062ec53b3fd95d+613edded/model.hlo_module.pb +3 -0
  22. neuronxcc-2.17.194.0+d312836f/MODULE_22cf23062ec53b3fd95d+613edded/model.neff +3 -0
  23. neuronxcc-2.17.194.0+d312836f/MODULE_24ff9ac2787ce9a1d276+613edded/model.neff +1 -1
  24. neuronxcc-2.17.194.0+d312836f/MODULE_2ef52130792b59d66c66+613edded/compile_flags.json +1 -0
  25. neuronxcc-2.17.194.0+d312836f/MODULE_2ef52130792b59d66c66+613edded/model.done +0 -0
  26. neuronxcc-2.17.194.0+d312836f/MODULE_2ef52130792b59d66c66+613edded/model.hlo_module.pb +3 -0
  27. neuronxcc-2.17.194.0+d312836f/MODULE_2ef52130792b59d66c66+613edded/model.neff +3 -0
  28. neuronxcc-2.17.194.0+d312836f/MODULE_320f2622d4d0c9fdd0f1+613edded/model.neff +1 -1
  29. neuronxcc-2.17.194.0+d312836f/MODULE_3aa61c04bd8d9ca8722a+613edded/model.neff +1 -1
  30. neuronxcc-2.17.194.0+d312836f/MODULE_3da832fdaa3d62981800+613edded/compile_flags.json +1 -0
  31. neuronxcc-2.17.194.0+d312836f/MODULE_3da832fdaa3d62981800+613edded/model.done +0 -0
  32. neuronxcc-2.17.194.0+d312836f/MODULE_3da832fdaa3d62981800+613edded/model.hlo_module.pb +3 -0
  33. neuronxcc-2.17.194.0+d312836f/MODULE_3da832fdaa3d62981800+613edded/model.neff +3 -0
  34. neuronxcc-2.17.194.0+d312836f/MODULE_40b73cdf6e14999378ef+613edded/compile_flags.json +1 -0
  35. neuronxcc-2.17.194.0+d312836f/MODULE_40b73cdf6e14999378ef+613edded/model.done +0 -0
  36. neuronxcc-2.17.194.0+d312836f/MODULE_40b73cdf6e14999378ef+613edded/model.hlo_module.pb +3 -0
  37. neuronxcc-2.17.194.0+d312836f/MODULE_40b73cdf6e14999378ef+613edded/model.neff +3 -0
  38. neuronxcc-2.17.194.0+d312836f/MODULE_48437cc74469a8ccaec8+613edded/model.neff +1 -1
  39. neuronxcc-2.17.194.0+d312836f/MODULE_48bfe9ceb9631fdca2d4+613edded/model.neff +1 -1
  40. neuronxcc-2.17.194.0+d312836f/MODULE_4dea33d55b39527f7fc4+613edded/compile_flags.json +1 -0
  41. neuronxcc-2.17.194.0+d312836f/MODULE_4dea33d55b39527f7fc4+613edded/model.done +0 -0
  42. neuronxcc-2.17.194.0+d312836f/MODULE_4dea33d55b39527f7fc4+613edded/model.hlo_module.pb +3 -0
  43. neuronxcc-2.17.194.0+d312836f/MODULE_4dea33d55b39527f7fc4+613edded/model.neff +3 -0
  44. neuronxcc-2.17.194.0+d312836f/MODULE_54947fa9c379486e4f5b+613edded/compile_flags.json +1 -0
  45. neuronxcc-2.17.194.0+d312836f/MODULE_54947fa9c379486e4f5b+613edded/model.done +0 -0
  46. neuronxcc-2.17.194.0+d312836f/MODULE_54947fa9c379486e4f5b+613edded/model.hlo_module.pb +3 -0
  47. neuronxcc-2.17.194.0+d312836f/MODULE_54947fa9c379486e4f5b+613edded/model.neff +3 -0
  48. neuronxcc-2.17.194.0+d312836f/MODULE_6819e64fb96e87ffece0+613edded/model.neff +1 -1
  49. neuronxcc-2.17.194.0+d312836f/MODULE_6f479fa0ca80b88a5b46+613edded/compile_flags.json +1 -0
  50. neuronxcc-2.17.194.0+d312836f/MODULE_6f479fa0ca80b88a5b46+613edded/model.done +0 -0
.gitattributes CHANGED
@@ -2237,3 +2237,15 @@ neuronxcc-2.16.372.0+4a9b2326/MODULE_d6a9bb00aa7189af3ffb/unet/model.neuron filt
2237
  neuronxcc-2.16.372.0+4a9b2326/MODULE_d6a9bb00aa7189af3ffb/vae_decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
2238
  neuronxcc-2.16.372.0+4a9b2326/MODULE_d6a9bb00aa7189af3ffb/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
2239
  neuronxcc-2.16.372.0+4a9b2326/MODULE_fd6170cedb4fe53c8433+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
2237
  neuronxcc-2.16.372.0+4a9b2326/MODULE_d6a9bb00aa7189af3ffb/vae_decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
2238
  neuronxcc-2.16.372.0+4a9b2326/MODULE_d6a9bb00aa7189af3ffb/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
2239
  neuronxcc-2.16.372.0+4a9b2326/MODULE_fd6170cedb4fe53c8433+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2240
+ neuronxcc-2.17.194.0+d312836f/MODULE_1649fc77b87fff02e370+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2241
+ neuronxcc-2.17.194.0+d312836f/MODULE_22cf23062ec53b3fd95d+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2242
+ neuronxcc-2.17.194.0+d312836f/MODULE_2ef52130792b59d66c66+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2243
+ neuronxcc-2.17.194.0+d312836f/MODULE_3da832fdaa3d62981800+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2244
+ neuronxcc-2.17.194.0+d312836f/MODULE_40b73cdf6e14999378ef+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2245
+ neuronxcc-2.17.194.0+d312836f/MODULE_4dea33d55b39527f7fc4+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2246
+ neuronxcc-2.17.194.0+d312836f/MODULE_54947fa9c379486e4f5b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2247
+ neuronxcc-2.17.194.0+d312836f/MODULE_6f479fa0ca80b88a5b46+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2248
+ neuronxcc-2.17.194.0+d312836f/MODULE_83cb40c0c38bacf5b8fd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2249
+ neuronxcc-2.17.194.0+d312836f/MODULE_913f4e1e2b4632438fe9+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2250
+ neuronxcc-2.17.194.0+d312836f/MODULE_e9bcfc17d832317203bd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
2251
+ neuronxcc-2.17.194.0+d312836f/MODULE_fce469267b2ad1b5d80e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/granite/hf-internal-testing/tiny-random-GraniteForCausalLM/28037758016041af399a.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-GraniteForCausalLM",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "GraniteForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "attention_multiplier": 1.0,
11
+ "embedding_multiplier": 1.0,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 32,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 64,
16
+ "logits_scaling": 1.0,
17
+ "max_position_embeddings": 2048,
18
+ "mlp_bias": false,
19
+ "model_type": "granite",
20
+ "neuron": {
21
+ "all_reduce_dtype": null,
22
+ "allow_flash_attention": true,
23
+ "attention_layout": "HSB",
24
+ "attn_output_transposed": false,
25
+ "auto_cast_type": "fp16",
26
+ "backend": "hlo",
27
+ "batch_size": 2,
28
+ "checkpoint_id": "hf-internal-testing/tiny-random-GraniteForCausalLM",
29
+ "checkpoint_revision": "c3074ebc0ac2fe545305f5e5f6cce2cc9b2aa0c5",
30
+ "collectives_layout": "HSB",
31
+ "continuous_batching": true,
32
+ "fuse_qkv": true,
33
+ "group_query_attention": null,
34
+ "log_softmax_scores": false,
35
+ "neuronxcc_version": "2.17.194.0+d312836f",
36
+ "optimum_neuron_version": "0.2.0.dev3",
37
+ "output_all_logits": false,
38
+ "sequence_length": 100,
39
+ "tp_degree": 2
40
+ },
41
+ "num_attention_heads": 4,
42
+ "num_hidden_layers": 2,
43
+ "num_key_value_heads": 4,
44
+ "residual_multiplier": 1.0,
45
+ "rms_norm_eps": 1e-06,
46
+ "rope_scaling": null,
47
+ "rope_theta": 10000.0,
48
+ "tie_word_embeddings": false,
49
+ "use_cache": true,
50
+ "vocab_size": 49152
51
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/granite/hf-internal-testing/tiny-random-GraniteForCausalLM/32ec2689f2e2b8d58073.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-GraniteForCausalLM",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "GraniteForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "attention_multiplier": 1.0,
11
+ "embedding_multiplier": 1.0,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 32,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 64,
16
+ "logits_scaling": 1.0,
17
+ "max_position_embeddings": 2048,
18
+ "mlp_bias": false,
19
+ "model_type": "granite",
20
+ "neuron": {
21
+ "all_reduce_dtype": null,
22
+ "allow_flash_attention": true,
23
+ "attention_layout": "HSB",
24
+ "attn_output_transposed": false,
25
+ "auto_cast_type": "fp16",
26
+ "backend": "hlo",
27
+ "batch_size": 1,
28
+ "checkpoint_id": "hf-internal-testing/tiny-random-GraniteForCausalLM",
29
+ "checkpoint_revision": "c3074ebc0ac2fe545305f5e5f6cce2cc9b2aa0c5",
30
+ "collectives_layout": "HSB",
31
+ "continuous_batching": false,
32
+ "fuse_qkv": true,
33
+ "group_query_attention": null,
34
+ "log_softmax_scores": false,
35
+ "neuronxcc_version": "2.17.194.0+d312836f",
36
+ "optimum_neuron_version": "0.2.0.dev3",
37
+ "output_all_logits": false,
38
+ "sequence_length": 100,
39
+ "tp_degree": 2
40
+ },
41
+ "num_attention_heads": 4,
42
+ "num_hidden_layers": 2,
43
+ "num_key_value_heads": 4,
44
+ "residual_multiplier": 1.0,
45
+ "rms_norm_eps": 1e-06,
46
+ "rope_scaling": null,
47
+ "rope_theta": 10000.0,
48
+ "tie_word_embeddings": false,
49
+ "use_cache": true,
50
+ "vocab_size": 49152
51
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/granite/hf-internal-testing/tiny-random-GraniteForCausalLM/3fdcfb75cf15e31ca9b5.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-GraniteForCausalLM",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "GraniteForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "attention_multiplier": 1.0,
11
+ "embedding_multiplier": 1.0,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 32,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 64,
16
+ "logits_scaling": 1.0,
17
+ "max_position_embeddings": 2048,
18
+ "mlp_bias": false,
19
+ "model_type": "granite",
20
+ "neuron": {
21
+ "all_reduce_dtype": null,
22
+ "allow_flash_attention": true,
23
+ "attention_layout": "HSB",
24
+ "attn_output_transposed": false,
25
+ "auto_cast_type": "fp32",
26
+ "backend": "hlo",
27
+ "batch_size": 1,
28
+ "checkpoint_id": "hf-internal-testing/tiny-random-GraniteForCausalLM",
29
+ "checkpoint_revision": "c3074ebc0ac2fe545305f5e5f6cce2cc9b2aa0c5",
30
+ "collectives_layout": "HSB",
31
+ "continuous_batching": false,
32
+ "fuse_qkv": true,
33
+ "group_query_attention": null,
34
+ "log_softmax_scores": false,
35
+ "neuronxcc_version": "2.17.194.0+d312836f",
36
+ "optimum_neuron_version": "0.2.0.dev3",
37
+ "output_all_logits": false,
38
+ "sequence_length": 100,
39
+ "tp_degree": 2
40
+ },
41
+ "num_attention_heads": 4,
42
+ "num_hidden_layers": 2,
43
+ "num_key_value_heads": 4,
44
+ "residual_multiplier": 1.0,
45
+ "rms_norm_eps": 1e-06,
46
+ "rope_scaling": null,
47
+ "rope_theta": 10000.0,
48
+ "tie_word_embeddings": false,
49
+ "use_cache": true,
50
+ "vocab_size": 49152
51
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/llama/llamafactory/tiny-random-Llama-3/610dad84114ec035cc32.json ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "all_reduce_dtype": null,
20
+ "allow_flash_attention": true,
21
+ "attention_layout": "BSH",
22
+ "attn_output_transposed": false,
23
+ "auto_cast_type": "fp16",
24
+ "backend": "hlo",
25
+ "batch_size": 2,
26
+ "checkpoint_id": "llamafactory/tiny-random-Llama-3",
27
+ "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8",
28
+ "collectives_layout": "HSB",
29
+ "continuous_batching": true,
30
+ "fuse_qkv": true,
31
+ "group_query_attention": null,
32
+ "log_softmax_scores": false,
33
+ "neuronxcc_version": "2.17.194.0+d312836f",
34
+ "optimum_neuron_version": "0.2.0.dev3",
35
+ "output_all_logits": false,
36
+ "sequence_length": 100,
37
+ "tp_degree": 2
38
+ },
39
+ "num_attention_heads": 4,
40
+ "num_hidden_layers": 2,
41
+ "num_key_value_heads": 4,
42
+ "pretraining_tp": 1,
43
+ "rms_norm_eps": 1e-05,
44
+ "rope_scaling": {
45
+ "factor": 8.0,
46
+ "high_freq_factor": 4.0,
47
+ "low_freq_factor": 1.0,
48
+ "original_max_position_embeddings": 8192,
49
+ "rope_type": "llama3"
50
+ },
51
+ "rope_theta": 500000.0,
52
+ "tie_word_embeddings": false,
53
+ "use_cache": true,
54
+ "vocab_size": 128256
55
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/llama/llamafactory/tiny-random-Llama-3/b645ea39b56c5af10179.json ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "all_reduce_dtype": null,
20
+ "allow_flash_attention": true,
21
+ "attention_layout": "BSH",
22
+ "attn_output_transposed": false,
23
+ "auto_cast_type": "fp16",
24
+ "backend": "hlo",
25
+ "batch_size": 1,
26
+ "checkpoint_id": "llamafactory/tiny-random-Llama-3",
27
+ "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8",
28
+ "collectives_layout": "HSB",
29
+ "continuous_batching": false,
30
+ "fuse_qkv": true,
31
+ "group_query_attention": null,
32
+ "log_softmax_scores": false,
33
+ "neuronxcc_version": "2.17.194.0+d312836f",
34
+ "optimum_neuron_version": "0.2.0.dev3",
35
+ "output_all_logits": false,
36
+ "sequence_length": 100,
37
+ "tp_degree": 2
38
+ },
39
+ "num_attention_heads": 4,
40
+ "num_hidden_layers": 2,
41
+ "num_key_value_heads": 4,
42
+ "pretraining_tp": 1,
43
+ "rms_norm_eps": 1e-05,
44
+ "rope_scaling": {
45
+ "factor": 8.0,
46
+ "high_freq_factor": 4.0,
47
+ "low_freq_factor": 1.0,
48
+ "original_max_position_embeddings": 8192,
49
+ "rope_type": "llama3"
50
+ },
51
+ "rope_theta": 500000.0,
52
+ "tie_word_embeddings": false,
53
+ "use_cache": true,
54
+ "vocab_size": 128256
55
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/llama/llamafactory/tiny-random-Llama-3/b84d42926b5badee9c7a.json ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "all_reduce_dtype": null,
20
+ "allow_flash_attention": true,
21
+ "attention_layout": "BSH",
22
+ "attn_output_transposed": false,
23
+ "auto_cast_type": "fp32",
24
+ "backend": "hlo",
25
+ "batch_size": 1,
26
+ "checkpoint_id": "llamafactory/tiny-random-Llama-3",
27
+ "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8",
28
+ "collectives_layout": "HSB",
29
+ "continuous_batching": false,
30
+ "fuse_qkv": true,
31
+ "group_query_attention": null,
32
+ "log_softmax_scores": false,
33
+ "neuronxcc_version": "2.17.194.0+d312836f",
34
+ "optimum_neuron_version": "0.2.0.dev3",
35
+ "output_all_logits": false,
36
+ "sequence_length": 100,
37
+ "tp_degree": 2
38
+ },
39
+ "num_attention_heads": 4,
40
+ "num_hidden_layers": 2,
41
+ "num_key_value_heads": 4,
42
+ "pretraining_tp": 1,
43
+ "rms_norm_eps": 1e-05,
44
+ "rope_scaling": {
45
+ "factor": 8.0,
46
+ "high_freq_factor": 4.0,
47
+ "low_freq_factor": 1.0,
48
+ "original_max_position_embeddings": 8192,
49
+ "rope_type": "llama3"
50
+ },
51
+ "rope_theta": 500000.0,
52
+ "tie_word_embeddings": false,
53
+ "use_cache": true,
54
+ "vocab_size": 128256
55
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/phi3/yujiepan/phi-4-tiny-random/2e3b2c893e5dd12fa28c.json ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "all_reduce_dtype": null,
20
+ "allow_flash_attention": false,
21
+ "attention_layout": "HSB",
22
+ "attn_output_transposed": false,
23
+ "auto_cast_type": "fp16",
24
+ "backend": "hlo",
25
+ "batch_size": 1,
26
+ "checkpoint_id": "yujiepan/phi-4-tiny-random",
27
+ "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a",
28
+ "collectives_layout": "HSB",
29
+ "continuous_batching": false,
30
+ "fuse_qkv": true,
31
+ "group_query_attention": "replicated-heads",
32
+ "log_softmax_scores": false,
33
+ "neuronxcc_version": "2.17.194.0+d312836f",
34
+ "optimum_neuron_version": "0.2.0.dev3",
35
+ "output_all_logits": false,
36
+ "sequence_length": 100,
37
+ "tp_degree": 2
38
+ },
39
+ "num_attention_heads": 2,
40
+ "num_hidden_layers": 2,
41
+ "num_key_value_heads": 1,
42
+ "original_max_position_embeddings": 16384,
43
+ "partial_rotary_factor": 1.0,
44
+ "resid_pdrop": 0.0,
45
+ "rms_norm_eps": 1e-05,
46
+ "rope_scaling": null,
47
+ "rope_theta": 250000,
48
+ "sliding_window": null,
49
+ "tie_word_embeddings": false,
50
+ "use_cache": true,
51
+ "vocab_size": 100352
52
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/phi3/yujiepan/phi-4-tiny-random/5f969dd5ec1310f92701.json ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "all_reduce_dtype": null,
20
+ "allow_flash_attention": false,
21
+ "attention_layout": "HSB",
22
+ "attn_output_transposed": false,
23
+ "auto_cast_type": "fp32",
24
+ "backend": "hlo",
25
+ "batch_size": 1,
26
+ "checkpoint_id": "yujiepan/phi-4-tiny-random",
27
+ "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a",
28
+ "collectives_layout": "HSB",
29
+ "continuous_batching": false,
30
+ "fuse_qkv": true,
31
+ "group_query_attention": "replicated-heads",
32
+ "log_softmax_scores": false,
33
+ "neuronxcc_version": "2.17.194.0+d312836f",
34
+ "optimum_neuron_version": "0.2.0.dev3",
35
+ "output_all_logits": false,
36
+ "sequence_length": 100,
37
+ "tp_degree": 2
38
+ },
39
+ "num_attention_heads": 2,
40
+ "num_hidden_layers": 2,
41
+ "num_key_value_heads": 1,
42
+ "original_max_position_embeddings": 16384,
43
+ "partial_rotary_factor": 1.0,
44
+ "resid_pdrop": 0.0,
45
+ "rms_norm_eps": 1e-05,
46
+ "rope_scaling": null,
47
+ "rope_theta": 250000,
48
+ "sliding_window": null,
49
+ "tie_word_embeddings": false,
50
+ "use_cache": true,
51
+ "vocab_size": 100352
52
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/phi3/yujiepan/phi-4-tiny-random/84c8e75e4defdd9fa414.json ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "all_reduce_dtype": null,
20
+ "allow_flash_attention": false,
21
+ "attention_layout": "HSB",
22
+ "attn_output_transposed": false,
23
+ "auto_cast_type": "fp16",
24
+ "backend": "hlo",
25
+ "batch_size": 2,
26
+ "checkpoint_id": "yujiepan/phi-4-tiny-random",
27
+ "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a",
28
+ "collectives_layout": "HSB",
29
+ "continuous_batching": true,
30
+ "fuse_qkv": true,
31
+ "group_query_attention": "replicated-heads",
32
+ "log_softmax_scores": false,
33
+ "neuronxcc_version": "2.17.194.0+d312836f",
34
+ "optimum_neuron_version": "0.2.0.dev3",
35
+ "output_all_logits": false,
36
+ "sequence_length": 100,
37
+ "tp_degree": 2
38
+ },
39
+ "num_attention_heads": 2,
40
+ "num_hidden_layers": 2,
41
+ "num_key_value_heads": 1,
42
+ "original_max_position_embeddings": 16384,
43
+ "partial_rotary_factor": 1.0,
44
+ "resid_pdrop": 0.0,
45
+ "rms_norm_eps": 1e-05,
46
+ "rope_scaling": null,
47
+ "rope_theta": 250000,
48
+ "sliding_window": null,
49
+ "tie_word_embeddings": false,
50
+ "use_cache": true,
51
+ "vocab_size": 100352
52
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/qwen2/Qwen/Qwen2.5-0.5B/2dcbb3f89df1561e38ca.json ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "Qwen/Qwen2.5-0.5B",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Qwen2ForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "hidden_act": "silu",
10
+ "hidden_size": 896,
11
+ "initializer_range": 0.02,
12
+ "intermediate_size": 4864,
13
+ "max_position_embeddings": 32768,
14
+ "max_window_layers": 24,
15
+ "model_type": "qwen2",
16
+ "neuron": {
17
+ "all_reduce_dtype": null,
18
+ "allow_flash_attention": true,
19
+ "attention_layout": "HSB",
20
+ "attn_output_transposed": false,
21
+ "auto_cast_type": "fp16",
22
+ "backend": "hlo",
23
+ "batch_size": 4,
24
+ "checkpoint_id": "Qwen/Qwen2.5-0.5B",
25
+ "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987",
26
+ "collectives_layout": "HSB",
27
+ "continuous_batching": true,
28
+ "fuse_qkv": false,
29
+ "group_query_attention": "shard-over-heads",
30
+ "log_softmax_scores": false,
31
+ "neuronxcc_version": "2.17.194.0+d312836f",
32
+ "optimum_neuron_version": "0.2.0.dev3",
33
+ "output_all_logits": false,
34
+ "sequence_length": 4096,
35
+ "tp_degree": 2
36
+ },
37
+ "num_attention_heads": 14,
38
+ "num_hidden_layers": 24,
39
+ "num_key_value_heads": 2,
40
+ "rms_norm_eps": 1e-06,
41
+ "rope_scaling": null,
42
+ "rope_theta": 1000000.0,
43
+ "sliding_window": 32768,
44
+ "tie_word_embeddings": true,
45
+ "use_cache": true,
46
+ "use_mrope": false,
47
+ "use_sliding_window": false,
48
+ "vocab_size": 151936
49
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/qwen2/yujiepan/qwen2.5-128k-tiny-random/1fd1a2e4204366d503f6.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "yujiepan/qwen2.5-128k-tiny-random",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Qwen2ForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "hidden_act": "silu",
10
+ "hidden_size": 8,
11
+ "initializer_range": 0.02,
12
+ "intermediate_size": 16,
13
+ "max_position_embeddings": 32768,
14
+ "max_window_layers": 1,
15
+ "model_type": "qwen2",
16
+ "neuron": {
17
+ "all_reduce_dtype": null,
18
+ "allow_flash_attention": true,
19
+ "attention_layout": "HSB",
20
+ "attn_output_transposed": false,
21
+ "auto_cast_type": "fp16",
22
+ "backend": "hlo",
23
+ "batch_size": 2,
24
+ "checkpoint_id": "yujiepan/qwen2.5-128k-tiny-random",
25
+ "checkpoint_revision": "c8296d4ca3f87782876d2382fbb6481d1beb8ef0",
26
+ "collectives_layout": "HSB",
27
+ "continuous_batching": true,
28
+ "fuse_qkv": false,
29
+ "group_query_attention": "shard-over-heads",
30
+ "log_softmax_scores": false,
31
+ "neuronxcc_version": "2.17.194.0+d312836f",
32
+ "optimum_neuron_version": "0.2.0.dev3",
33
+ "output_all_logits": false,
34
+ "sequence_length": 100,
35
+ "tp_degree": 2
36
+ },
37
+ "num_attention_heads": 4,
38
+ "num_hidden_layers": 2,
39
+ "num_key_value_heads": 2,
40
+ "rms_norm_eps": 1e-06,
41
+ "rope_scaling": {
42
+ "factor": 4.0,
43
+ "original_max_position_embeddings": 32768,
44
+ "rope_type": "yarn",
45
+ "type": "yarn"
46
+ },
47
+ "rope_theta": 1000000.0,
48
+ "sliding_window": 131072,
49
+ "tie_word_embeddings": false,
50
+ "use_cache": true,
51
+ "use_sliding_window": false,
52
+ "vocab_size": 152064
53
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/qwen2/yujiepan/qwen2.5-128k-tiny-random/2a5b61ac8a8b5940982a.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "yujiepan/qwen2.5-128k-tiny-random",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Qwen2ForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "hidden_act": "silu",
10
+ "hidden_size": 8,
11
+ "initializer_range": 0.02,
12
+ "intermediate_size": 16,
13
+ "max_position_embeddings": 32768,
14
+ "max_window_layers": 1,
15
+ "model_type": "qwen2",
16
+ "neuron": {
17
+ "all_reduce_dtype": null,
18
+ "allow_flash_attention": true,
19
+ "attention_layout": "HSB",
20
+ "attn_output_transposed": false,
21
+ "auto_cast_type": "fp16",
22
+ "backend": "hlo",
23
+ "batch_size": 1,
24
+ "checkpoint_id": "yujiepan/qwen2.5-128k-tiny-random",
25
+ "checkpoint_revision": "c8296d4ca3f87782876d2382fbb6481d1beb8ef0",
26
+ "collectives_layout": "HSB",
27
+ "continuous_batching": false,
28
+ "fuse_qkv": false,
29
+ "group_query_attention": "shard-over-heads",
30
+ "log_softmax_scores": false,
31
+ "neuronxcc_version": "2.17.194.0+d312836f",
32
+ "optimum_neuron_version": "0.2.0.dev3",
33
+ "output_all_logits": false,
34
+ "sequence_length": 100,
35
+ "tp_degree": 2
36
+ },
37
+ "num_attention_heads": 4,
38
+ "num_hidden_layers": 2,
39
+ "num_key_value_heads": 2,
40
+ "rms_norm_eps": 1e-06,
41
+ "rope_scaling": {
42
+ "factor": 4.0,
43
+ "original_max_position_embeddings": 32768,
44
+ "rope_type": "yarn",
45
+ "type": "yarn"
46
+ },
47
+ "rope_theta": 1000000.0,
48
+ "sliding_window": 131072,
49
+ "tie_word_embeddings": false,
50
+ "use_cache": true,
51
+ "use_sliding_window": false,
52
+ "vocab_size": 152064
53
+ }
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev3/inference/qwen2/yujiepan/qwen2.5-128k-tiny-random/b1b72b82166aaabf0a64.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "yujiepan/qwen2.5-128k-tiny-random",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Qwen2ForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "hidden_act": "silu",
10
+ "hidden_size": 8,
11
+ "initializer_range": 0.02,
12
+ "intermediate_size": 16,
13
+ "max_position_embeddings": 32768,
14
+ "max_window_layers": 1,
15
+ "model_type": "qwen2",
16
+ "neuron": {
17
+ "all_reduce_dtype": null,
18
+ "allow_flash_attention": true,
19
+ "attention_layout": "HSB",
20
+ "attn_output_transposed": false,
21
+ "auto_cast_type": "fp32",
22
+ "backend": "hlo",
23
+ "batch_size": 1,
24
+ "checkpoint_id": "yujiepan/qwen2.5-128k-tiny-random",
25
+ "checkpoint_revision": "c8296d4ca3f87782876d2382fbb6481d1beb8ef0",
26
+ "collectives_layout": "HSB",
27
+ "continuous_batching": false,
28
+ "fuse_qkv": false,
29
+ "group_query_attention": "shard-over-heads",
30
+ "log_softmax_scores": false,
31
+ "neuronxcc_version": "2.17.194.0+d312836f",
32
+ "optimum_neuron_version": "0.2.0.dev3",
33
+ "output_all_logits": false,
34
+ "sequence_length": 100,
35
+ "tp_degree": 2
36
+ },
37
+ "num_attention_heads": 4,
38
+ "num_hidden_layers": 2,
39
+ "num_key_value_heads": 2,
40
+ "rms_norm_eps": 1e-06,
41
+ "rope_scaling": {
42
+ "factor": 4.0,
43
+ "original_max_position_embeddings": 32768,
44
+ "rope_type": "yarn",
45
+ "type": "yarn"
46
+ },
47
+ "rope_theta": 1000000.0,
48
+ "sliding_window": 131072,
49
+ "tie_word_embeddings": false,
50
+ "use_cache": true,
51
+ "use_sliding_window": false,
52
+ "vocab_size": 152064
53
+ }
neuronxcc-2.17.194.0+d312836f/MODULE_1649fc77b87fff02e370+613edded/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.17.194.0+d312836f/MODULE_1649fc77b87fff02e370+613edded/model.done ADDED
File without changes
neuronxcc-2.17.194.0+d312836f/MODULE_1649fc77b87fff02e370+613edded/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:206a7ad89370cff8d2b2322c1ac670c89f032f09dacf78c1919f6862a5f988f8
3
+ size 20204
neuronxcc-2.17.194.0+d312836f/MODULE_1649fc77b87fff02e370+613edded/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6d594981e2da8b776c3c5a3eca2164c9f6fa6f0c449ac944452b14c94ddf0495
3
+ size 134144
neuronxcc-2.17.194.0+d312836f/MODULE_22cf23062ec53b3fd95d+613edded/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.17.194.0+d312836f/MODULE_22cf23062ec53b3fd95d+613edded/model.done ADDED
File without changes
neuronxcc-2.17.194.0+d312836f/MODULE_22cf23062ec53b3fd95d+613edded/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bffb7434bff71e7ba712292397b2ad4ee6448cc362b3669d6140ae7eda9fed6f
3
+ size 21402
neuronxcc-2.17.194.0+d312836f/MODULE_22cf23062ec53b3fd95d+613edded/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6112eb7624714812b25dc2f55104031955ba212a83c290b3e87880f61ae48063
3
+ size 144384
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:be1fc03322aa9d157a6b41fd67890f055e0d0878b6504e5f47642cca280b170f
3
  size 134144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:816f6e2b2800b0dd18941740a24948cd8febc13e0f8e9f734cc9f5dc87ad41fa
3
  size 134144
neuronxcc-2.17.194.0+d312836f/MODULE_2ef52130792b59d66c66+613edded/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.17.194.0+d312836f/MODULE_2ef52130792b59d66c66+613edded/model.done ADDED
File without changes
neuronxcc-2.17.194.0+d312836f/MODULE_2ef52130792b59d66c66+613edded/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:14f3f25ac84183e3e0fbaea932e3c43b3d7df2d165a2f3fd8d17bbb0d9b67100
3
+ size 17475
neuronxcc-2.17.194.0+d312836f/MODULE_2ef52130792b59d66c66+613edded/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:95ab464c4cf288bbb4eddc39060737a10a606d9a1780f619bd1249e31e77841c
3
+ size 134144
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 sha256:68e874b1f31bc45206f4c2049124f0985cad9a7f6c5b2b71ad453bf290cea9be
3
  size 144384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0d72740519c6f1609852cbefc2ca9c47c9c430eac2d564c11fb3e01d62dacc11
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 sha256:71a975bfc9afe131a009261a81dcde76b6f8e3de48a82027819f81062c48bc35
3
  size 144384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d5c71c790f5cec922792760f98f020c80f8851e65d99cf2165fd2380ae3b9a07
3
  size 144384
neuronxcc-2.17.194.0+d312836f/MODULE_3da832fdaa3d62981800+613edded/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.17.194.0+d312836f/MODULE_3da832fdaa3d62981800+613edded/model.done ADDED
File without changes
neuronxcc-2.17.194.0+d312836f/MODULE_3da832fdaa3d62981800+613edded/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ffcc61769d8fb9bea5e15c42d960c69b9df6f97061385d0bd0a21d0ce6906af6
3
+ size 26028
neuronxcc-2.17.194.0+d312836f/MODULE_3da832fdaa3d62981800+613edded/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b76edbd85e8c2a4c9aa2e4f0a91f8a2d55f739330a4c09432d5068a8eb021e53
3
+ size 154624
neuronxcc-2.17.194.0+d312836f/MODULE_40b73cdf6e14999378ef+613edded/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.17.194.0+d312836f/MODULE_40b73cdf6e14999378ef+613edded/model.done ADDED
File without changes
neuronxcc-2.17.194.0+d312836f/MODULE_40b73cdf6e14999378ef+613edded/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:67d2c56e794836b16b020f5cd066ca11a1d29dafab34df3118d3d816e3ff6e95
3
+ size 23565
neuronxcc-2.17.194.0+d312836f/MODULE_40b73cdf6e14999378ef+613edded/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5871a8f23388ad9e78e2f48f3ee955c4facce1a3f373c6980e68f9cbef93853f
3
+ size 164864
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 sha256:93a20c24604eaa08c2e14387740552dcccd1ea088a21a12050f0885a46a2ddf4
3
  size 154624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:58c71f97d235636814f05e8d6047fbe743ddb7e6d20f90577c8d29be05e7689d
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 sha256:356a341c3e1907677aed4628bfaec17f0d2b48ac090addd2d8b1fa072f9332fb
3
  size 144384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c242498d423f976198e48db300828acb03aeac1809d796be2e6a923bfb44689c
3
  size 144384
neuronxcc-2.17.194.0+d312836f/MODULE_4dea33d55b39527f7fc4+613edded/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.17.194.0+d312836f/MODULE_4dea33d55b39527f7fc4+613edded/model.done ADDED
File without changes
neuronxcc-2.17.194.0+d312836f/MODULE_4dea33d55b39527f7fc4+613edded/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b15bcc1c290cb31694bdf9dcca14c4212f94ac98633cf07e5f78c88252248395
3
+ size 17129
neuronxcc-2.17.194.0+d312836f/MODULE_4dea33d55b39527f7fc4+613edded/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bda494af1d3d8fe727fd6f8b8632c99dae58cb97e60706daad7ed24c100aa459
3
+ size 134144
neuronxcc-2.17.194.0+d312836f/MODULE_54947fa9c379486e4f5b+613edded/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.17.194.0+d312836f/MODULE_54947fa9c379486e4f5b+613edded/model.done ADDED
File without changes
neuronxcc-2.17.194.0+d312836f/MODULE_54947fa9c379486e4f5b+613edded/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f0395aa664ed1ddbb5519230b5f4664bb55551ee04dbb00dee5498927b8c53f4
3
+ size 20595
neuronxcc-2.17.194.0+d312836f/MODULE_54947fa9c379486e4f5b+613edded/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca8e32f83811391c47229b37a069f3e04202bd0b8df874d0638d0c49241e93d5
3
+ size 144384
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:d5324e4e6db89b1e57c3b5a686220b043ff876db7db2ad3c578baaa18b48b385
3
  size 144384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8b759b85a480af8a95b47df41a132f4fa1e2dfc1b6ab7712662c40a6fec780fc
3
  size 144384
neuronxcc-2.17.194.0+d312836f/MODULE_6f479fa0ca80b88a5b46+613edded/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.17.194.0+d312836f/MODULE_6f479fa0ca80b88a5b46+613edded/model.done ADDED
File without changes