Hugging Face's logo

aws-neuron
/
optimum-neuron-cache

Model card Files
xet
 Community
684
dacorvo HF Staff commited on
Commit
189cc4e
·
verified ·
1 Parent(s): 3cd5664

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-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev2/gemma3_text/unsloth/gemma-3-270m-it/7fde6971cc66d8965351.json +81 -0
  3. neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev2/llama/unsloth/Llama-3.2-1B-Instruct/237f3427f0ffd5bd2890.json +63 -0
  4. neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev2/qwen2/Qwen/Qwen2.5-0.5B/682142914fb2798c2038.json +82 -0
  5. neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev2/smollm3/HuggingFaceTB/SmolLM3-3B/8adf13bbd08271a796c8.json +134 -0
  6. neuronxcc-2.21.33363.0+82129205/MODULE_0007bf645cd8f44068db+a02c3a36/model.hlo_module.pb +1 -1
  7. neuronxcc-2.21.33363.0+82129205/MODULE_0007bf645cd8f44068db+a02c3a36/model.neff +1 -1
  8. neuronxcc-2.21.33363.0+82129205/MODULE_0007bf645cd8f44068db+a02c3a36/wrapped_neff.hlo +1 -1
  9. neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff +0 -0
  10. neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff +0 -0
  11. neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff +0 -0
  12. neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff +0 -0
  13. neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff +0 -0
  14. neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff +0 -0
  15. neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff +0 -0
  16. neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff +0 -0
  17. neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff +0 -0
  18. neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff +0 -0
  19. neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff +0 -0
  20. neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff +0 -0
  21. neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff +0 -0
  22. neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff +0 -0
  23. neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff +0 -0
  24. neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff +0 -0
  25. neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff +0 -0
  26. neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff +0 -0
  27. neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff +0 -0
  28. neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff +0 -0
  29. neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff +0 -0
  30. neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff +0 -0
  31. neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff +0 -0
  32. neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff +0 -0
  33. neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff +0 -0
  34. neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff +0 -0
  35. neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff +0 -0
  36. neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff +0 -0
  37. neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff +0 -0
  38. neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff +0 -0
  39. neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff +0 -0
  40. neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff +0 -0
  41. neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff +0 -0
  42. neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff +0 -0
  43. neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff +0 -0
  44. neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff +0 -0
  45. neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff +0 -0
  46. neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff +0 -0
  47. neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff +0 -0
  48. neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff +0 -0
  49. neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff +0 -0
  50. neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff +0 -0
.gitattributes CHANGED
@@ -16792,3 +16792,15 @@ neuronxcc-2.21.33363.0+82129205/MODULE_fe491aea85c492a9d7ed+7e8048af/model.neff
16792
  neuronxcc-2.21.33363.0+82129205/MODULE_fe491aea85c492a9d7ed+7e8048af/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
16793
  neuronxcc-2.21.33363.0+82129205/MODULE_fe6b8b40a8e1d622e71e+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
16794
  neuronxcc-2.21.33363.0+82129205/MODULE_fe6b8b40a8e1d622e71e+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
16792
  neuronxcc-2.21.33363.0+82129205/MODULE_fe491aea85c492a9d7ed+7e8048af/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
16793
  neuronxcc-2.21.33363.0+82129205/MODULE_fe6b8b40a8e1d622e71e+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
16794
  neuronxcc-2.21.33363.0+82129205/MODULE_fe6b8b40a8e1d622e71e+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
16795
+ neuronxcc-2.21.33363.0+82129205/MODULE_246e55a7b5fd9a448d11+665fc8af/model.neff filter=lfs diff=lfs merge=lfs -text
16796
+ neuronxcc-2.21.33363.0+82129205/MODULE_246e55a7b5fd9a448d11+665fc8af/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
16797
+ neuronxcc-2.21.33363.0+82129205/MODULE_2f83a5c5770b55577746+aca16090/model.neff filter=lfs diff=lfs merge=lfs -text
16798
+ neuronxcc-2.21.33363.0+82129205/MODULE_2f83a5c5770b55577746+aca16090/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
16799
+ neuronxcc-2.21.33363.0+82129205/MODULE_31944d1660ea5c0c01d4+ba0e29f2/model.neff filter=lfs diff=lfs merge=lfs -text
16800
+ neuronxcc-2.21.33363.0+82129205/MODULE_31944d1660ea5c0c01d4+ba0e29f2/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
16801
+ neuronxcc-2.21.33363.0+82129205/MODULE_4fbc0195c41b496f58f8+b4b79373/model.neff filter=lfs diff=lfs merge=lfs -text
16802
+ neuronxcc-2.21.33363.0+82129205/MODULE_4fbc0195c41b496f58f8+b4b79373/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
16803
+ neuronxcc-2.21.33363.0+82129205/MODULE_57424899c5118ea4aa3f+c3339a48/model.neff filter=lfs diff=lfs merge=lfs -text
16804
+ neuronxcc-2.21.33363.0+82129205/MODULE_57424899c5118ea4aa3f+c3339a48/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
16805
+ neuronxcc-2.21.33363.0+82129205/MODULE_66b8b1d207a139e87593+cf3fc371/model.neff filter=lfs diff=lfs merge=lfs -text
16806
+ neuronxcc-2.21.33363.0+82129205/MODULE_66b8b1d207a139e87593+cf3fc371/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev2/gemma3_text/unsloth/gemma-3-270m-it/7fde6971cc66d8965351.json ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "unsloth/gemma-3-270m-it",
4
+ "_sliding_window_pattern": 6,
5
+ "_task": "text-generation",
6
+ "architectures": [
7
+ "Gemma3ForCausalLM"
8
+ ],
9
+ "attention_bias": false,
10
+ "attention_dropout": 0.0,
11
+ "attn_logit_softcapping": null,
12
+ "dtype": "bfloat16",
13
+ "final_logit_softcapping": null,
14
+ "head_dim": 256,
15
+ "hidden_activation": "gelu_pytorch_tanh",
16
+ "hidden_size": 640,
17
+ "initializer_range": 0.02,
18
+ "intermediate_size": 2048,
19
+ "layer_types": [
20
+ "sliding_attention",
21
+ "sliding_attention",
22
+ "sliding_attention",
23
+ "sliding_attention",
24
+ "sliding_attention",
25
+ "full_attention",
26
+ "sliding_attention",
27
+ "sliding_attention",
28
+ "sliding_attention",
29
+ "sliding_attention",
30
+ "sliding_attention",
31
+ "full_attention",
32
+ "sliding_attention",
33
+ "sliding_attention",
34
+ "sliding_attention",
35
+ "sliding_attention",
36
+ "sliding_attention",
37
+ "full_attention"
38
+ ],
39
+ "max_position_embeddings": 32768,
40
+ "model_type": "gemma3_text",
41
+ "neuron": {
42
+ "_serialized_key": "NxDNeuronConfig",
43
+ "batch_size": 1,
44
+ "capacity_factor": null,
45
+ "checkpoint_id": "unsloth/gemma-3-270m-it",
46
+ "checkpoint_revision": "23cf460f6bb16954176b3ddcc8d4f250501458a9",
47
+ "continuous_batching": false,
48
+ "ep_degree": 1,
49
+ "fused_qkv": true,
50
+ "glu_mlp": true,
51
+ "local_ranks_size": 2,
52
+ "max_batch_size": 1,
53
+ "max_context_length": 1024,
54
+ "max_topk": 256,
55
+ "n_active_tokens": 1024,
56
+ "neuronxcc_version": "2.21.33363.0+82129205",
57
+ "on_device_sampling": true,
58
+ "optimum_neuron_version": "0.4.6.dev2",
59
+ "output_logits": false,
60
+ "pp_degree": 1,
61
+ "sequence_length": 1024,
62
+ "speculation_length": 0,
63
+ "start_rank_id": 0,
64
+ "target": "trn1",
65
+ "torch_dtype": "bfloat16",
66
+ "tp_degree": 2
67
+ },
68
+ "num_attention_heads": 4,
69
+ "num_hidden_layers": 18,
70
+ "num_key_value_heads": 1,
71
+ "query_pre_attn_scalar": 256,
72
+ "rms_norm_eps": 1e-06,
73
+ "rope_local_base_freq": 10000.0,
74
+ "rope_scaling": null,
75
+ "rope_theta": 1000000.0,
76
+ "sliding_window": 512,
77
+ "unsloth_fixed": true,
78
+ "use_bidirectional_attention": false,
79
+ "use_cache": true,
80
+ "vocab_size": 262144
81
+ }
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev2/llama/unsloth/Llama-3.2-1B-Instruct/237f3427f0ffd5bd2890.json ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "unsloth/Llama-3.2-1B-Instruct",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "LlamaForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "dtype": "bfloat16",
11
+ "head_dim": 64,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 2048,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 8192,
16
+ "max_position_embeddings": 131072,
17
+ "mlp_bias": false,
18
+ "model_type": "llama",
19
+ "neuron": {
20
+ "_serialized_key": "NxDNeuronConfig",
21
+ "batch_size": 1,
22
+ "capacity_factor": null,
23
+ "checkpoint_id": "unsloth/Llama-3.2-1B-Instruct",
24
+ "checkpoint_revision": "5a8abab4a5d6f164389b1079fb721cfab8d7126c",
25
+ "continuous_batching": false,
26
+ "ep_degree": 1,
27
+ "fused_qkv": true,
28
+ "glu_mlp": true,
29
+ "local_ranks_size": 2,
30
+ "max_batch_size": 1,
31
+ "max_context_length": 4096,
32
+ "max_topk": 256,
33
+ "n_active_tokens": 4096,
34
+ "neuronxcc_version": "2.21.33363.0+82129205",
35
+ "on_device_sampling": true,
36
+ "optimum_neuron_version": "0.4.6.dev2",
37
+ "output_logits": false,
38
+ "pp_degree": 1,
39
+ "sequence_length": 4096,
40
+ "speculation_length": 0,
41
+ "start_rank_id": 0,
42
+ "target": "trn1",
43
+ "torch_dtype": "bfloat16",
44
+ "tp_degree": 2
45
+ },
46
+ "num_attention_heads": 32,
47
+ "num_hidden_layers": 16,
48
+ "num_key_value_heads": 8,
49
+ "pretraining_tp": 1,
50
+ "rms_norm_eps": 1e-05,
51
+ "rope_scaling": {
52
+ "factor": 32.0,
53
+ "high_freq_factor": 4.0,
54
+ "low_freq_factor": 1.0,
55
+ "original_max_position_embeddings": 8192,
56
+ "rope_type": "llama3"
57
+ },
58
+ "rope_theta": 500000.0,
59
+ "tie_word_embeddings": true,
60
+ "unsloth_fixed": true,
61
+ "use_cache": true,
62
+ "vocab_size": 128256
63
+ }
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev2/qwen2/Qwen/Qwen2.5-0.5B/682142914fb2798c2038.json ADDED
@@ -0,0 +1,82 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "dtype": "bfloat16",
10
+ "hidden_act": "silu",
11
+ "hidden_size": 896,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 4864,
14
+ "layer_types": [
15
+ "full_attention",
16
+ "full_attention",
17
+ "full_attention",
18
+ "full_attention",
19
+ "full_attention",
20
+ "full_attention",
21
+ "full_attention",
22
+ "full_attention",
23
+ "full_attention",
24
+ "full_attention",
25
+ "full_attention",
26
+ "full_attention",
27
+ "full_attention",
28
+ "full_attention",
29
+ "full_attention",
30
+ "full_attention",
31
+ "full_attention",
32
+ "full_attention",
33
+ "full_attention",
34
+ "full_attention",
35
+ "full_attention",
36
+ "full_attention",
37
+ "full_attention",
38
+ "full_attention"
39
+ ],
40
+ "max_position_embeddings": 32768,
41
+ "max_window_layers": 24,
42
+ "model_type": "qwen2",
43
+ "neuron": {
44
+ "_serialized_key": "NxDNeuronConfig",
45
+ "batch_size": 2,
46
+ "capacity_factor": null,
47
+ "checkpoint_id": "Qwen/Qwen2.5-0.5B",
48
+ "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987",
49
+ "continuous_batching": true,
50
+ "ep_degree": 1,
51
+ "fused_qkv": false,
52
+ "glu_mlp": true,
53
+ "local_ranks_size": 2,
54
+ "max_batch_size": 2,
55
+ "max_context_length": 1024,
56
+ "max_topk": 256,
57
+ "n_active_tokens": 1024,
58
+ "neuronxcc_version": "2.21.33363.0+82129205",
59
+ "on_device_sampling": false,
60
+ "optimum_neuron_version": "0.4.6.dev2",
61
+ "output_logits": false,
62
+ "pp_degree": 1,
63
+ "sequence_length": 1024,
64
+ "speculation_length": 0,
65
+ "start_rank_id": 0,
66
+ "target": "trn1",
67
+ "torch_dtype": "bfloat16",
68
+ "tp_degree": 2
69
+ },
70
+ "num_attention_heads": 14,
71
+ "num_hidden_layers": 24,
72
+ "num_key_value_heads": 2,
73
+ "rms_norm_eps": 1e-06,
74
+ "rope_scaling": null,
75
+ "rope_theta": 1000000.0,
76
+ "sliding_window": null,
77
+ "tie_word_embeddings": true,
78
+ "use_cache": true,
79
+ "use_mrope": false,
80
+ "use_sliding_window": false,
81
+ "vocab_size": 151936
82
+ }
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev2/smollm3/HuggingFaceTB/SmolLM3-3B/8adf13bbd08271a796c8.json ADDED
@@ -0,0 +1,134 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "HuggingFaceTB/SmolLM3-3B",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "SmolLM3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "dtype": "bfloat16",
11
+ "hidden_act": "silu",
12
+ "hidden_size": 2048,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 11008,
15
+ "layer_types": [
16
+ "full_attention",
17
+ "full_attention",
18
+ "full_attention",
19
+ "full_attention",
20
+ "full_attention",
21
+ "full_attention",
22
+ "full_attention",
23
+ "full_attention",
24
+ "full_attention",
25
+ "full_attention",
26
+ "full_attention",
27
+ "full_attention",
28
+ "full_attention",
29
+ "full_attention",
30
+ "full_attention",
31
+ "full_attention",
32
+ "full_attention",
33
+ "full_attention",
34
+ "full_attention",
35
+ "full_attention",
36
+ "full_attention",
37
+ "full_attention",
38
+ "full_attention",
39
+ "full_attention",
40
+ "full_attention",
41
+ "full_attention",
42
+ "full_attention",
43
+ "full_attention",
44
+ "full_attention",
45
+ "full_attention",
46
+ "full_attention",
47
+ "full_attention",
48
+ "full_attention",
49
+ "full_attention",
50
+ "full_attention",
51
+ "full_attention"
52
+ ],
53
+ "max_position_embeddings": 65536,
54
+ "max_window_layers": 28,
55
+ "mlp_bias": false,
56
+ "model_type": "smollm3",
57
+ "neuron": {
58
+ "_serialized_key": "NxDNeuronConfig",
59
+ "batch_size": 1,
60
+ "capacity_factor": null,
61
+ "checkpoint_id": "HuggingFaceTB/SmolLM3-3B",
62
+ "checkpoint_revision": "a07cc9a04f16550a088caea529712d1d335b0ac1",
63
+ "continuous_batching": false,
64
+ "ep_degree": 1,
65
+ "fused_qkv": true,
66
+ "glu_mlp": true,
67
+ "local_ranks_size": 2,
68
+ "max_batch_size": 1,
69
+ "max_context_length": 1024,
70
+ "max_topk": 256,
71
+ "n_active_tokens": 1024,
72
+ "neuronxcc_version": "2.21.33363.0+82129205",
73
+ "on_device_sampling": true,
74
+ "optimum_neuron_version": "0.4.6.dev2",
75
+ "output_logits": false,
76
+ "pp_degree": 1,
77
+ "sequence_length": 1024,
78
+ "speculation_length": 0,
79
+ "start_rank_id": 0,
80
+ "target": "trn1",
81
+ "torch_dtype": "bfloat16",
82
+ "tp_degree": 2
83
+ },
84
+ "no_rope_layer_interval": 4,
85
+ "no_rope_layers": [
86
+ 1,
87
+ 1,
88
+ 1,
89
+ 0,
90
+ 1,
91
+ 1,
92
+ 1,
93
+ 0,
94
+ 1,
95
+ 1,
96
+ 1,
97
+ 0,
98
+ 1,
99
+ 1,
100
+ 1,
101
+ 0,
102
+ 1,
103
+ 1,
104
+ 1,
105
+ 0,
106
+ 1,
107
+ 1,
108
+ 1,
109
+ 0,
110
+ 1,
111
+ 1,
112
+ 1,
113
+ 0,
114
+ 1,
115
+ 1,
116
+ 1,
117
+ 0,
118
+ 1,
119
+ 1,
120
+ 1,
121
+ 0
122
+ ],
123
+ "num_attention_heads": 16,
124
+ "num_hidden_layers": 36,
125
+ "num_key_value_heads": 4,
126
+ "pretraining_tp": 2,
127
+ "rms_norm_eps": 1e-06,
128
+ "rope_scaling": null,
129
+ "rope_theta": 5000000.0,
130
+ "sliding_window": null,
131
+ "use_cache": false,
132
+ "use_sliding_window": false,
133
+ "vocab_size": 128256
134
+ }
neuronxcc-2.21.33363.0+82129205/MODULE_0007bf645cd8f44068db+a02c3a36/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f605a954b114104fa81403712a2efbe24d99be08d5bcfe37bb5b9a565639ad02
3
  size 443127
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c33e25476c481f8a4f2f2b91c32193eb39fbb2a46c25d95214e5f798dc90626f
3
  size 443127
neuronxcc-2.21.33363.0+82129205/MODULE_0007bf645cd8f44068db+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4959586c29d7dbe597ae03fad6b1c0ffc0f7bc0a0b574257c2a43c418ff8469c
3
  size 2171904
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6b54396cc0ffbbce49b7cdeed4dba797622c9dbfd6e766134dfc0d9767b9d794
3
  size 2171904
neuronxcc-2.21.33363.0+82129205/MODULE_0007bf645cd8f44068db+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:82d721b25142da6db296f0e7a48219cfb200220310600e0214afb4872bff2ad5
3
  size 2245991
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c27a9a0b39627623582007b60c7942709d32254cf53587d8f84895ae5861aaa
3
  size 2245991
neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff differ