Delete neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/4b3ca94bb4445bc28bc8.json

@@ -1 +0,0 @@
-{"apply_residual_connection_post_layernorm": false, "architectures": ["BloomForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 1, "dtype": "float32", "eos_token_id": 2, "gradient_checkpointing": false, "hidden_dropout": 0.1, "hidden_size": 32, "id2label": {"0": "LABEL_0", "1": "LABEL_1", "2": "LABEL_2"}, "initializer_range": 0.02, "is_decoder": true, "label2id": {"LABEL_0": 0, "LABEL_1": 1, "LABEL_2": 2}, "layer_norm_epsilon": 1e-05, "model_type": "bloom", "n_head": 4, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM", "checkpoint_revision": "92b07e9b7b4f986fa7c54e2ac3b9201b4ba5212e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 3, "pretraining_tp": 1, "seq_length": 7, "slow_but_exact": true, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 1024}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/4eccd2e66477ff5ef5b3.json

@@ -1 +0,0 @@
-{"vocab_size": 1024, "hidden_size": 32, "n_layer": 5, "n_head": 4, "layer_norm_epsilon": 1e-05, "initializer_range": 0.02, "use_cache": true, "pretraining_tp": 1, "apply_residual_connection_post_layernorm": false, "hidden_dropout": 0.1, "attention_dropout": 0.1, "bos_token_id": 1, "eos_token_id": 2, "slow_but_exact": true, "torch_dtype": "float32", "is_decoder": true, "architectures": ["BloomForCausalLM"], "id2label": {"0": "LABEL_0", "1": "LABEL_1", "2": "LABEL_2"}, "label2id": {"LABEL_0": 0, "LABEL_1": 1, "LABEL_2": 2}, "pad_token_id": 3, "dtype": "float32", "gradient_checkpointing": false, "model_type": "bloom", "n_positions": 512, "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM", "checkpoint_revision": "0f4f06f162cd67d34d03ee156484e4001d468500", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "seq_length": 7, "type_vocab_size": 16}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/d635309efcd921a3a3f6.json

@@ -1 +0,0 @@
-{"vocab_size": 1024, "hidden_size": 32, "n_layer": 5, "n_head": 4, "layer_norm_epsilon": 1e-05, "initializer_range": 0.02, "use_cache": true, "pretraining_tp": 1, "apply_residual_connection_post_layernorm": false, "hidden_dropout": 0.1, "attention_dropout": 0.1, "bos_token_id": 1, "eos_token_id": 2, "slow_but_exact": true, "torch_dtype": "float32", "is_decoder": true, "architectures": ["BloomForCausalLM"], "id2label": {"0": "LABEL_0", "1": "LABEL_1", "2": "LABEL_2"}, "label2id": {"LABEL_0": 0, "LABEL_1": 1, "LABEL_2": 2}, "pad_token_id": 3, "dtype": "float32", "gradient_checkpointing": false, "model_type": "bloom", "n_positions": 512, "seq_length": 7, "type_vocab_size": 16, "neuron": {"task": "text-generation", "batch_size": 2, "num_cores": 2, "auto_cast_type": "fp32", "sequence_length": 100, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM", "checkpoint_revision": "0f4f06f162cd67d34d03ee156484e4001d468500"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/gpt2/gpt2/2c3a47e6fca2fd6ac3cf.json

@@ -1 +0,0 @@
-{"vocab_size": 50257, "n_positions": 1024, "n_embd": 768, "n_layer": 12, "n_head": 12, "n_inner": null, "activation_function": "gelu_new", "resid_pdrop": 0.1, "embd_pdrop": 0.1, "attn_pdrop": 0.1, "layer_norm_epsilon": 1e-05, "initializer_range": 0.02, "summary_type": "cls_index", "summary_use_proj": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "scale_attn_weights": true, "use_cache": true, "scale_attn_by_inverse_layer_idx": false, "reorder_and_upcast_attn": false, "bos_token_id": 50256, "eos_token_id": 50256, "architectures": ["GPT2LMHeadModel"], "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "model_type": "gpt2", "n_ctx": 1024, "neuron": {"task": "text-generation", "batch_size": 4, "num_cores": 2, "auto_cast_type": "fp32", "sequence_length": 1024, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/gpt2/gpt2/8ae9e463111b0fa38673.json

@@ -1 +0,0 @@
-{"vocab_size": 50257, "n_positions": 1024, "n_embd": 768, "n_layer": 12, "n_head": 12, "n_inner": null, "activation_function": "gelu_new", "resid_pdrop": 0.1, "embd_pdrop": 0.1, "attn_pdrop": 0.1, "layer_norm_epsilon": 1e-05, "initializer_range": 0.02, "summary_type": "cls_index", "summary_use_proj": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "scale_attn_weights": true, "use_cache": true, "scale_attn_by_inverse_layer_idx": false, "reorder_and_upcast_attn": false, "bos_token_id": 50256, "eos_token_id": 50256, "architectures": ["GPT2LMHeadModel"], "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "model_type": "gpt2", "n_ctx": 1024, "neuron": {"auto_cast_type": "fp32", "batch_size": 4, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 1024, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/165740fba2c72c56c43d.json

@@ -1 +0,0 @@
-{"activation_function": "gelu_new", "attention_probs_dropout_prob": 0.1, "attn_pdrop": 0.1, "bos_token_id": 98, "embd_pdrop": 0.1, "eos_token_id": 98, "gradient_checkpointing": false, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 512, "n_embd": 32, "n_head": 4, "n_inner": null, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-gpt2", "checkpoint_revision": "91c0fe31d692dd8448d9bc06e8d1877345009e3b", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 98, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "type_vocab_size": 16, "use_cache": true, "vocab_size": 1000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/2037d1be0146ceb8d639.json

@@ -1 +0,0 @@
-{"vocab_size": 1000, "n_positions": 512, "n_embd": 32, "n_layer": 5, "n_head": 4, "n_inner": null, "activation_function": "gelu_new", "resid_pdrop": 0.1, "embd_pdrop": 0.1, "attn_pdrop": 0.1, "layer_norm_epsilon": 1e-05, "initializer_range": 0.02, "summary_type": "cls_index", "summary_use_proj": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "scale_attn_weights": true, "use_cache": true, "scale_attn_by_inverse_layer_idx": false, "reorder_and_upcast_attn": false, "bos_token_id": 98, "eos_token_id": 98, "pad_token_id": 98, "attention_probs_dropout_prob": 0.1, "gradient_checkpointing": false, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "intermediate_size": 37, "model_type": "gpt2", "n_ctx": 512, "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-gpt2", "checkpoint_revision": "91c0fe31d692dd8448d9bc06e8d1877345009e3b", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "type_vocab_size": 16}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/3edc0caf8d3805a471bd.json

@@ -1 +0,0 @@
-{"vocab_size": 1000, "n_positions": 512, "n_embd": 32, "n_layer": 5, "n_head": 4, "n_inner": null, "activation_function": "gelu_new", "resid_pdrop": 0.1, "embd_pdrop": 0.1, "attn_pdrop": 0.1, "layer_norm_epsilon": 1e-05, "initializer_range": 0.02, "summary_type": "cls_index", "summary_use_proj": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "scale_attn_weights": true, "use_cache": true, "scale_attn_by_inverse_layer_idx": false, "reorder_and_upcast_attn": false, "bos_token_id": 98, "eos_token_id": 98, "pad_token_id": 98, "attention_probs_dropout_prob": 0.1, "gradient_checkpointing": false, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "intermediate_size": 37, "model_type": "gpt2", "n_ctx": 512, "type_vocab_size": 16, "neuron": {"task": "text-generation", "batch_size": 2, "num_cores": 2, "auto_cast_type": "fp32", "sequence_length": 100, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "hf-internal-testing/tiny-random-gpt2", "checkpoint_revision": "91c0fe31d692dd8448d9bc06e8d1877345009e3b"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/NousResearch/Llama-2-7b-chat-hf/6c7cfec071642e5fecee.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "pad_token_id": 0, "eos_token_id": 2, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "NousResearch/Llama-2-7b-chat-hf", "checkpoint_revision": "37892f30c23786c0d5367d80481fa0d9fba93cf8", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/NousResearch/Llama-2-7b-chat-hf/b9904e0c46d338b623d2.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "pad_token_id": 0, "eos_token_id": 2, "model_type": "llama", "neuron": {"task": "text-generation", "batch_size": 4, "num_cores": 24, "auto_cast_type": "fp16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "NousResearch/Llama-2-7b-chat-hf", "checkpoint_revision": "37892f30c23786c0d5367d80481fa0d9fba93cf8"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/dacorvo/tiny-random-llama/62a76db84304b34ae305.json

@@ -1 +0,0 @@
-{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 128, "initializer_range": 0.02, "intermediate_size": 256, "max_position_embeddings": 512, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "dacorvo/tiny-random-llama", "checkpoint_revision": "7fdafd2fe6a2d31c6abb72ae60db606d8bb23196", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 1, "num_hidden_layers": 1, "num_key_value_heads": 1, "pretraining_tp": 1, "rms_norm_eps": 1e-06, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float32", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/dacorvo/tiny-random-llama/7c595f241fa311d7b227.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 512, "hidden_size": 128, "intermediate_size": 256, "num_hidden_layers": 1, "num_attention_heads": 1, "num_key_value_heads": 1, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-06, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float32", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "dacorvo/tiny-random-llama", "checkpoint_revision": "7fdafd2fe6a2d31c6abb72ae60db606d8bb23196", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/dacorvo/tiny-random-llama/b650b9d41c4386d36a89.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 512, "hidden_size": 128, "intermediate_size": 256, "num_hidden_layers": 1, "num_attention_heads": 1, "num_key_value_heads": 1, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-06, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float32", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "llama", "neuron": {"task": "text-generation", "batch_size": 2, "num_cores": 2, "auto_cast_type": "fp32", "sequence_length": 100, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "dacorvo/tiny-random-llama", "checkpoint_revision": "7fdafd2fe6a2d31c6abb72ae60db606d8bb23196"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/04a286f25641fb37c0da.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/09bd40b23d9644abeb4e.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/13cbb9ec2234e7ca3c26.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/22d316bd8b3cdfb5514b.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/2e95f36ce181921d7a7b.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/39695bb575ebdb0bdfc1.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/3a328b55c6445b520b8e.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "llama", "neuron": {"task": "text-generation", "batch_size": 16, "num_cores": 24, "auto_cast_type": "fp16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/3cad45de92cf81a0a16f.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/3fa0092e4157c2ddb47b.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/44c81784d74cd4713969.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "llama", "neuron": {"task": "text-generation", "batch_size": 8, "num_cores": 24, "auto_cast_type": "fp16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/4d28ec49befb4e87ec23.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/5424b76f5346f48c87dc.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/58eec5854a9f564f8a27.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 16, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/6a314b3d69ca329d2711.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "llama", "neuron": {"task": "text-generation", "batch_size": 1, "num_cores": 24, "auto_cast_type": "fp16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/7a0a7caf91982c3c67b0.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/7a9166d6f50e7c014961.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/7e2416095368bdd0a5cd.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/ad1f7e0ca3ab366f91d4.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 8, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/d3acdf9ff1795080fd1c.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/d49c97f13a4593173a49.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "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/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/dd0ae8e9d982a2ed293b.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "llama", "neuron": {"task": "text-generation", "batch_size": 4, "num_cores": 24, "auto_cast_type": "fp16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/efa4119dabd9c1f8277f.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/ff6b3f58fac88c2f96c4.json

@@ -1 +0,0 @@
-{"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": "92011f62d7604e261f748ec0cfe6329f31193e33", "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/0.0.21.dev0/inference/llama/princeton-nlp/Sheared-LLaMA-1.3B/059827c299e8d9043f57.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 2048, "intermediate_size": 5504, "num_hidden_layers": 24, "num_attention_heads": 16, "num_key_value_heads": 16, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float32", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "pad_token_id": 0, "eos_token_id": 2, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 4, "checkpoint_id": "princeton-nlp/Sheared-LLaMA-1.3B", "checkpoint_revision": "a4b76938edbf571ea7d7d9904861cbdca08809b4", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 1024, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/princeton-nlp/Sheared-LLaMA-1.3B/ece87a51a12bdc2169c6.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 2048, "intermediate_size": 5504, "num_hidden_layers": 24, "num_attention_heads": 16, "num_key_value_heads": 16, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float32", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "pad_token_id": 0, "eos_token_id": 2, "model_type": "llama", "neuron": {"task": "text-generation", "batch_size": 4, "num_cores": 2, "auto_cast_type": "fp32", "sequence_length": 1024, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "princeton-nlp/Sheared-LLaMA-1.3B", "checkpoint_revision": "a4b76938edbf571ea7d7d9904861cbdca08809b4"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/1cfca4036d7b607639ea.json

@@ -1 +0,0 @@
-{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.1, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "hidden_size": 32, "initializer_range": 0.02, "intermediate_size": 37, "is_decoder": true, "max_position_embeddings": 512, "model_type": "mistral", "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM", "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 2, "pad_token_id": 0, "rms_norm_eps": 1e-06, "rope_theta": 10000.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/43c8baaa3d89767fe7f2.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 512, "hidden_size": 32, "intermediate_size": 37, "num_hidden_layers": 2, "num_attention_heads": 4, "sliding_window": 4096, "num_key_value_heads": 2, "hidden_act": "gelu", "initializer_range": 0.02, "rms_norm_eps": 1e-06, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "float32", "tie_word_embeddings": false, "is_decoder": true, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "pad_token_id": 0, "eos_token_id": 2, "attention_probs_dropout_prob": 0.1, "hidden_dropout_prob": 0.1, "model_type": "mistral", "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM", "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "type_vocab_size": 16}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/f9b607c62a5393e0b42f.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 512, "hidden_size": 32, "intermediate_size": 37, "num_hidden_layers": 2, "num_attention_heads": 4, "sliding_window": 4096, "num_key_value_heads": 2, "hidden_act": "gelu", "initializer_range": 0.02, "rms_norm_eps": 1e-06, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "float32", "tie_word_embeddings": false, "is_decoder": true, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "pad_token_id": 0, "eos_token_id": 2, "attention_probs_dropout_prob": 0.1, "hidden_dropout_prob": 0.1, "model_type": "mistral", "type_vocab_size": 16, "neuron": {"task": "text-generation", "batch_size": 2, "num_cores": 2, "auto_cast_type": "fp32", "sequence_length": 100, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM", "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.1/7a99245071db1d4e48e6.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.1", "checkpoint_revision": "73068f3702d050a2fd5aa2ca1e612e5036429398", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.1/b0a5a2e1e8dd02dbe923.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"task": "text-generation", "batch_size": 1, "num_cores": 24, "auto_cast_type": "fp16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.1", "checkpoint_revision": "73068f3702d050a2fd5aa2ca1e612e5036429398"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/1c4540cc29dcb3b99ff8.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": null, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 1000000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/2d6c1d9513debbcc5f7f.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": null, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 1000000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"task": "text-generation", "batch_size": 1, "num_cores": 8, "auto_cast_type": "bf16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/41284a607429208347c0.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": null, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 1000000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"task": "text-generation", "batch_size": 16, "num_cores": 8, "auto_cast_type": "bf16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/4620b76c4f39e71a8459.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": null, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 1000000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"task": "text-generation", "batch_size": 1, "num_cores": 24, "auto_cast_type": "fp16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/4fcf5530fe99f6f4f3e6.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": null, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 1000000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 8, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/5c9bfea69bb260a59b0c.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": null, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 1000000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 16, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/698b844f07e07829c78e.json

@@ -1 +0,0 @@
-{"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": null, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 1000000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}}