dacorvo HF staff commited on
Commit
76e5e4b
1 Parent(s): d109adf

Synchronizing local compiler cache.

Browse files
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21/inference/llama/meta-llama/Llama-2-13b-chat-hf/2311cc196a0f70044562.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": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "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}