dacorvo HF staff commited on
Commit
dd36392
1 Parent(s): 61e83f1

Synchronizing local compiler cache.

Browse files
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.19/inference/llama/meta-llama/Llama-2-13b-chat-hf/4b83e8f7a9fb417d14c0.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 5120, "intermediate_size": 13824, "num_hidden_layers": 40, "num_attention_heads": 40, "num_key_value_heads": 40, "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": 8, "auto_cast_type": "fp16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "c2f3ec81aac798ae26dcc57799a994dfbf521496"}}