Hugging Face's logo

aws-neuron
/
optimum-neuron-cache

Model card Files Community
93
dacorvo HF staff commited on
Commit
5a79963
•
1 Parent(s): 136eae1

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 +34 -0
  2. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/5d4688c06ca1eea9cd2d.json +1 -0
  3. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/642817569926cd4530c8.json +1 -0
  4. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/c29571197f94c8546e45.json +1 -0
  5. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/d9b1eaeab430d169db57.json +1 -0
  6. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/57d08ed2158d3bc4efd8.json +1 -0
  7. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/be28f38f27aeb510e1c8.json +1 -0
  8. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/c9527c01253d9424170a.json +1 -0
  9. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/d6a72bcc14cde1f0364f.json +1 -0
  10. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/f6a3964311a50e56da2e.json +1 -0
  11. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/llama/defog/sqlcoder-7b-2/08b79e703c61b2f851cc.json +1 -0
  12. neuronxcc-2.13.66.0+6dfecc895/MODULE_01fe42b103324d50dd0f+2c2d707e/compile_flags.txt +1 -0
  13. neuronxcc-2.13.66.0+6dfecc895/MODULE_01fe42b103324d50dd0f+2c2d707e/model.hlo_module.pb +3 -0
  14. neuronxcc-2.13.66.0+6dfecc895/MODULE_01fe42b103324d50dd0f+2c2d707e/model.neff +3 -0
  15. neuronxcc-2.13.66.0+6dfecc895/MODULE_142005b3d93076706472+2c2d707e/model.hlo_module.pb +3 -0
  16. neuronxcc-2.13.66.0+6dfecc895/MODULE_21b1b0223e4fb6f5af41+2c2d707e/compile_flags.txt +1 -0
  17. neuronxcc-2.13.66.0+6dfecc895/MODULE_21b1b0223e4fb6f5af41+2c2d707e/model.hlo_module.pb +3 -0
  18. neuronxcc-2.13.66.0+6dfecc895/MODULE_21b1b0223e4fb6f5af41+2c2d707e/model.neff +3 -0
  19. neuronxcc-2.13.66.0+6dfecc895/MODULE_3020655465d2e0990c19+2c2d707e/compile_flags.txt +1 -0
  20. neuronxcc-2.13.66.0+6dfecc895/MODULE_3020655465d2e0990c19+2c2d707e/model.hlo_module.pb +3 -0
  21. neuronxcc-2.13.66.0+6dfecc895/MODULE_3020655465d2e0990c19+2c2d707e/model.neff +3 -0
  22. neuronxcc-2.13.66.0+6dfecc895/MODULE_312cb57ef1e4a82c10e2+2c2d707e/model.hlo_module.pb +3 -0
  23. neuronxcc-2.13.66.0+6dfecc895/MODULE_31bd89a484b00253cd84+2c2d707e/model.hlo_module.pb +3 -0
  24. neuronxcc-2.13.66.0+6dfecc895/MODULE_4051477874156d7450fe+2c2d707e/compile_flags.txt +1 -0
  25. neuronxcc-2.13.66.0+6dfecc895/MODULE_4051477874156d7450fe+2c2d707e/model.hlo_module.pb +3 -0
  26. neuronxcc-2.13.66.0+6dfecc895/MODULE_4051477874156d7450fe+2c2d707e/model.neff +3 -0
  27. neuronxcc-2.13.66.0+6dfecc895/MODULE_410c6d6a0c764608be0d+2c2d707e/compile_flags.txt +1 -0
  28. neuronxcc-2.13.66.0+6dfecc895/MODULE_410c6d6a0c764608be0d+2c2d707e/model.hlo_module.pb +3 -0
  29. neuronxcc-2.13.66.0+6dfecc895/MODULE_410c6d6a0c764608be0d+2c2d707e/model.neff +3 -0
  30. neuronxcc-2.13.66.0+6dfecc895/MODULE_426cd9312d26b5e0bf95+2c2d707e/compile_flags.txt +1 -0
  31. neuronxcc-2.13.66.0+6dfecc895/MODULE_426cd9312d26b5e0bf95+2c2d707e/model.hlo_module.pb +3 -0
  32. neuronxcc-2.13.66.0+6dfecc895/MODULE_426cd9312d26b5e0bf95+2c2d707e/model.neff +3 -0
  33. neuronxcc-2.13.66.0+6dfecc895/MODULE_443c7de81db84f0f078c+2c2d707e/model.neff +3 -0
  34. neuronxcc-2.13.66.0+6dfecc895/MODULE_44563e64a3dddda1db4d+2c2d707e/compile_flags.txt +1 -0
  35. neuronxcc-2.13.66.0+6dfecc895/MODULE_44563e64a3dddda1db4d+2c2d707e/model.hlo_module.pb +3 -0
  36. neuronxcc-2.13.66.0+6dfecc895/MODULE_44563e64a3dddda1db4d+2c2d707e/model.neff +3 -0
  37. neuronxcc-2.13.66.0+6dfecc895/MODULE_47eae16823191f315893+2c2d707e/compile_flags.txt +1 -0
  38. neuronxcc-2.13.66.0+6dfecc895/MODULE_47eae16823191f315893+2c2d707e/model.hlo_module.pb +3 -0
  39. neuronxcc-2.13.66.0+6dfecc895/MODULE_47eae16823191f315893+2c2d707e/model.neff +3 -0
  40. neuronxcc-2.13.66.0+6dfecc895/MODULE_4d18af62746967c79d82+2c2d707e/compile_flags.txt +1 -0
  41. neuronxcc-2.13.66.0+6dfecc895/MODULE_4d18af62746967c79d82+2c2d707e/model.hlo_module.pb +3 -0
  42. neuronxcc-2.13.66.0+6dfecc895/MODULE_4d18af62746967c79d82+2c2d707e/model.neff +3 -0
  43. neuronxcc-2.13.66.0+6dfecc895/MODULE_522ed20ba0b487552ad5+2c2d707e/model.hlo_module.pb +3 -0
  44. neuronxcc-2.13.66.0+6dfecc895/MODULE_539309b68144a4afde0b+2c2d707e/compile_flags.txt +1 -0
  45. neuronxcc-2.13.66.0+6dfecc895/MODULE_539309b68144a4afde0b+2c2d707e/model.hlo_module.pb +3 -0
  46. neuronxcc-2.13.66.0+6dfecc895/MODULE_539309b68144a4afde0b+2c2d707e/model.neff +3 -0
  47. neuronxcc-2.13.66.0+6dfecc895/MODULE_53f35b4110e7bebbcc45+2c2d707e/compile_flags.txt +1 -0
  48. neuronxcc-2.13.66.0+6dfecc895/MODULE_53f35b4110e7bebbcc45+2c2d707e/model.hlo_module.pb +3 -0
  49. neuronxcc-2.13.66.0+6dfecc895/MODULE_53f35b4110e7bebbcc45+2c2d707e/model.neff +3 -0
  50. neuronxcc-2.13.66.0+6dfecc895/MODULE_5ab5b6d9ff18b8896d5a+2c2d707e/compile_flags.txt +1 -0
.gitattributes CHANGED
@@ -2654,3 +2654,37 @@ neuronxcc-2.13.66.0+6dfecc895/MODULE_5f931e8bf304c87202cd+2c2d707e/model.neff fi
2654
  neuronxcc-2.13.66.0+6dfecc895/MODULE_a4fe1c700a9abdb20641+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2655
  neuronxcc-2.13.66.0+6dfecc895/MODULE_b54a08c18ad776940cd0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2656
  neuronxcc-2.13.66.0+6dfecc895/MODULE_b59ff3b7e545f85152c8+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2654
  neuronxcc-2.13.66.0+6dfecc895/MODULE_a4fe1c700a9abdb20641+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2655
  neuronxcc-2.13.66.0+6dfecc895/MODULE_b54a08c18ad776940cd0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2656
  neuronxcc-2.13.66.0+6dfecc895/MODULE_b59ff3b7e545f85152c8+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2657
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_01fe42b103324d50dd0f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2658
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_21b1b0223e4fb6f5af41+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2659
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_3020655465d2e0990c19+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2660
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_4051477874156d7450fe+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2661
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_410c6d6a0c764608be0d+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2662
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_426cd9312d26b5e0bf95+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2663
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_443c7de81db84f0f078c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2664
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_44563e64a3dddda1db4d+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2665
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_47eae16823191f315893+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2666
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_4d18af62746967c79d82+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2667
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_539309b68144a4afde0b+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2668
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_53f35b4110e7bebbcc45+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2669
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_5ab5b6d9ff18b8896d5a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2670
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_62cafa382ec618cff287+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2671
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_639cfa0c302f955a704e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2672
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_64d9d7ab844b064ddcff+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2673
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_72f780796470e2d45102+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2674
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_7943455ee983f978e58c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2675
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_7c67afcb8bd12bae0123+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2676
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_85a47887d4fc41e26097+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2677
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_8f8b48a08a0c68320dbf+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2678
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_98b888cee6ab04043987+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2679
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_9db2a3438f041d4e467e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2680
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_a29e3f3cc47445e1c231+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2681
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_a97de47fb0d0f47ea4c0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2682
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_b7517ea4b617cd73f3ad+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2683
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_cda75e05b1d55681d8eb+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2684
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_d4cc88cd9d2d907c99f6+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2685
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_d66f5802e5ad890e3919+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2686
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_e334ed8a755e5380cbbd+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2687
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_ea017e5effcd92c4e6ff+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2688
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_ee55eab3eac6b30f23a0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2689
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_f2145a17aa8d689f7038+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
2690
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_f74f07a635721468c2ae+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/5d4688c06ca1eea9cd2d.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "a2cb7a712bb6e5e736ca7f8cd98167f81a0b5bd8", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/642817569926cd4530c8.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 16, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "a2cb7a712bb6e5e736ca7f8cd98167f81a0b5bd8", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/c29571197f94c8546e45.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 8, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "a2cb7a712bb6e5e736ca7f8cd98167f81a0b5bd8", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/d9b1eaeab430d169db57.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 32, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "a2cb7a712bb6e5e736ca7f8cd98167f81a0b5bd8", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/57d08ed2158d3bc4efd8.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 16, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "f5db02db724555f92da89c216ac04704f23d4590", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/be28f38f27aeb510e1c8.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "f5db02db724555f92da89c216ac04704f23d4590", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/c9527c01253d9424170a.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "f5db02db724555f92da89c216ac04704f23d4590", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/d6a72bcc14cde1f0364f.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 32, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "f5db02db724555f92da89c216ac04704f23d4590", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/f6a3964311a50e56da2e.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 8, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "f5db02db724555f92da89c216ac04704f23d4590", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/llama/defog/sqlcoder-7b-2/08b79e703c61b2f851cc.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 16384, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "defog/sqlcoder-7b-2", "checkpoint_revision": "7e5b6f7981c0aa7d143f6bec6fa26625bdfcbe66", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 1000000, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32016}
neuronxcc-2.13.66.0+6dfecc895/MODULE_01fe42b103324d50dd0f+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_01fe42b103324d50dd0f+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9651ac6a82101e2da0eaa54eeb71671927d1cd35f7031e1118c9c8e3ce902b75
3
+ size 292188
neuronxcc-2.13.66.0+6dfecc895/MODULE_01fe42b103324d50dd0f+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9aae69ecb4a3764efc750d860ab1879542a56e8aafd867b24207fadd57030ee8
3
+ size 8244224
neuronxcc-2.13.66.0+6dfecc895/MODULE_142005b3d93076706472+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2816b10dcc5107592b5bd8fbb9b828cef19ae6a6dc281edfce5c134ea8620073
3
+ size 301369
neuronxcc-2.13.66.0+6dfecc895/MODULE_21b1b0223e4fb6f5af41+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_21b1b0223e4fb6f5af41+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:26525b90f7655fdf8a1366ab186317f8ef87a940102130a131edd3b776c64afa
3
+ size 299610
neuronxcc-2.13.66.0+6dfecc895/MODULE_21b1b0223e4fb6f5af41+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:647d4971554b115074a5d1efb73a71b547215bc89bc5ac3af9683f91af042454
3
+ size 2161664
neuronxcc-2.13.66.0+6dfecc895/MODULE_3020655465d2e0990c19+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_3020655465d2e0990c19+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c1a4e9a556d295e43b93fa54987cb08b21f05a761fa322beaf0d839587b5dc0f
3
+ size 299610
neuronxcc-2.13.66.0+6dfecc895/MODULE_3020655465d2e0990c19+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:13f64c745785649059fd2bf9f760387b9aa29ed2a436e808878ac9dcc8190b53
3
+ size 2427904
neuronxcc-2.13.66.0+6dfecc895/MODULE_312cb57ef1e4a82c10e2+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:069bc855215002ad4f29ae20d3c0cd082bbcda7d16f8c8ec12634a88a46afba7
3
+ size 298010
neuronxcc-2.13.66.0+6dfecc895/MODULE_31bd89a484b00253cd84+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fffb92a5dc964361ba99f5becaf80f146e749dc2982eb1fc859f504258bd6807
3
+ size 337486
neuronxcc-2.13.66.0+6dfecc895/MODULE_4051477874156d7450fe+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_4051477874156d7450fe+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0f92f7c0539766a36e06f93be6b5b8c26736837fbb84bef8e733e5c120d68248
3
+ size 420713
neuronxcc-2.13.66.0+6dfecc895/MODULE_4051477874156d7450fe+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:faadd4e50cafb354ca4100bda46e5f0d49ad82431050ed22da69790a3d1088e7
3
+ size 30024704
neuronxcc-2.13.66.0+6dfecc895/MODULE_410c6d6a0c764608be0d+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_410c6d6a0c764608be0d+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a39002c742b526c1d25080f085d4404e696f5067844da317a4403bdba477a12
3
+ size 295835
neuronxcc-2.13.66.0+6dfecc895/MODULE_410c6d6a0c764608be0d+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f7c9a4d8ad38fcaf1fa8e44ac96793d75451f4867d4565c278d4d087246fb207
3
+ size 8909824
neuronxcc-2.13.66.0+6dfecc895/MODULE_426cd9312d26b5e0bf95+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_426cd9312d26b5e0bf95+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:54fa5b9b20bbb94ac7dd18e82683af038f246e41bfb3288999d178cb3344e384
3
+ size 296219
neuronxcc-2.13.66.0+6dfecc895/MODULE_426cd9312d26b5e0bf95+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7f087f3482370f029c3910b2ebce46016ca33dec12fa2b2b11dbfab59ba8acb3
3
+ size 2960384
neuronxcc-2.13.66.0+6dfecc895/MODULE_443c7de81db84f0f078c+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2710534bde9e62b9aa96153376baf77e2ef7e50d21728ab651f5b97c616c836a
3
+ size 30024704
neuronxcc-2.13.66.0+6dfecc895/MODULE_44563e64a3dddda1db4d+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_44563e64a3dddda1db4d+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2a2f181f04beaac9d096d3aebf89aa6ea883d5c6db15fe4a428cf5cffbdd5df7
3
+ size 299610
neuronxcc-2.13.66.0+6dfecc895/MODULE_44563e64a3dddda1db4d+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:04731055ba71bc04552825fc869ee2fc20dea0efe8e56b5ce5aac95584ec347a
3
+ size 2212864
neuronxcc-2.13.66.0+6dfecc895/MODULE_47eae16823191f315893+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_47eae16823191f315893+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74fd95bec34f2c116fa9bca9f99add961a4c24981b58125584a6ac5b1043c244
3
+ size 299610
neuronxcc-2.13.66.0+6dfecc895/MODULE_47eae16823191f315893+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:036d173fe684811ba65e05a405f930106563744240bfaf8d18160825adc5d652
3
+ size 2274304
neuronxcc-2.13.66.0+6dfecc895/MODULE_4d18af62746967c79d82+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_4d18af62746967c79d82+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c0fde127dfb0a3f1949f1d87211ddacd4c9d874cb52202b18fd3018b93864c91
3
+ size 299226
neuronxcc-2.13.66.0+6dfecc895/MODULE_4d18af62746967c79d82+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ab29489f59000a5075367e42c5edd166c342f5cd6a8abb3f879482db038fffe6
3
+ size 7619584
neuronxcc-2.13.66.0+6dfecc895/MODULE_522ed20ba0b487552ad5+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b41ed0c1d8390f2103dc0517b13b748bc779e5644c02efab33444c5f81b76640
3
+ size 382458
neuronxcc-2.13.66.0+6dfecc895/MODULE_539309b68144a4afde0b+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_539309b68144a4afde0b+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f22e2dbe61f23faa7225930e9ede6e502e603385b8e6ad00802b87ec9013f2cc
3
+ size 476771
neuronxcc-2.13.66.0+6dfecc895/MODULE_539309b68144a4afde0b+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ba405706383c6fcafbf7a0879e2579e7b558e76e216b5a6ee6524b8cc0b0f14
3
+ size 14285824
neuronxcc-2.13.66.0+6dfecc895/MODULE_53f35b4110e7bebbcc45+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_53f35b4110e7bebbcc45+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be6313039d8693ed78d1a25acff373312bf0b3c837edff815caec3f70d2bfabd
3
+ size 299226
neuronxcc-2.13.66.0+6dfecc895/MODULE_53f35b4110e7bebbcc45+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:23af3682bd096103b870f32821c8aaea0a03a2fe3c3554f8cb76014c6516d21c
3
+ size 8182784
neuronxcc-2.13.66.0+6dfecc895/MODULE_5ab5b6d9ff18b8896d5a+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none