from deepsparse import Pipeline
from transformers import LlamaConfig, LlamaTokenizerFast

tokenizer = LlamaTokenizerFast.from_pretrained("hf-internal-testing/llama-tokenizer")
config = {
   "_name_or_path": None,
   "architectures": [
      "LlamaForCausalLM"
   ],
   "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",
   "num_attention_heads": 40,
   "num_hidden_layers": 40,
   "num_key_value_heads": 40,
   "pretraining_tp": 1,
   "rms_norm_eps": 1e-05,
   "rope_scaling": None,
   "tie_word_embeddings": "false",
   "torch_dtype": "float16",
   "transformers_version": "4.31.0.dev0",
   "use_cache": "true",
   "vocab_size": 32000
}

llama = Pipeline.create(
   task="text-generation",
   model_path="/home/dsikka/models_llama/deployment_13",
   engine_type="onnxruntime",
   deterministic=False,
   config=config,
   tokenizer=tokenizer
)

inference = llama(sequences=["Hello?"])
for s in inference.sequences:
   print(s)