| Compiled with: |
| ```json |
| compiler_args = {"num_cores": 16, "auto_cast_type": auto_cast_type} |
| input_shapes = {"batch_size": 8, "sequence_length": 32} |
| ``` |
|
|
| Usage: |
| ```python |
| import torch |
| from optimum.neuron import NeuronModelForCausalLM |
| from transformers import AutoTokenizer |
| model_id="deepseek-ai/DeepSeek-R1-Distill-Llama-70B" |
| prompt="What is is the capital of France?" |
| tokenizer = AutoTokenizer.from_pretrained(model_id) |
| tokenizer.pad_token_id = tokenizer.eos_token_id if tokenizer.pad_token_id is None else tokenizer.pad_token_id |
| inputs = tokenizer(prompt, return_tensors="pt") |
| model = NeuronModelForCausalLM.from_pretrained("yahavb/DeepSeek-R1-Distill-Llama-70B-Neuron") |
| outputs = model.generate(**inputs,max_new_tokens=512,do_sample=True,use_cache=True,temperature=0.7,top_k=50,top_p=0.9) |
| outputs=outputs[0, inputs.input_ids.size(-1):] |
| response=tokenizer.decode(outputs, skip_special_tokens=True) |
| print(response) |
| |
| |
