run_interactive_multiple_host.py

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import os
import random
import time
from typing import List

import jax
from absl import app, flags
from jetstream.engine import token_utils
from jetstream_pt import ray_engine
from jetstream_pt.config import FLAGS

_NUM_HOSTS = flags.DEFINE_integer(
    "num_hosts", 0, "Number of TPU host", required=False
)

_WORKER_CHIPS = flags.DEFINE_integer(
    "worker_chips", 4, "Number of TPU chips per worker", required=False
)

_TPU_CHIPS = flags.DEFINE_integer(
    "tpu_chips", 4, "All devices TPU chips", required=False
)


def create_engine():
  """create a pytorch engine"""
  jax.config.update("jax_default_prng_impl", "unsafe_rbg")
  os.environ["TF_CPP_MIN_LOG_LEVEL"] = "0"

  start = time.perf_counter()
  engine = ray_engine.create_pytorch_ray_engine(
      model_name=FLAGS.model_name,
      tokenizer_path=FLAGS.tokenizer_path,
      ckpt_path=FLAGS.checkpoint_path,
      bf16_enable=FLAGS.bf16_enable,
      param_size=FLAGS.size,
      context_length=FLAGS.context_length,
      batch_size=FLAGS.batch_size,
      quantize_weights=FLAGS.quantize_weights,
      quantize_kv=FLAGS.quantize_kv_cache,
      max_cache_length=FLAGS.max_cache_length,
      sharding_config=FLAGS.sharding_config,
      num_hosts=_NUM_HOSTS.value,
      worker_chips=_WORKER_CHIPS.value,
      tpu_chips=_TPU_CHIPS.value,
  )

  print("Initialize engine", time.perf_counter() - start)
  return engine


# pylint: disable-next=all
def main(argv):

  engine = create_engine()

  start = time.perf_counter()
  engine.load_params()
  print("Load params ", time.perf_counter() - start)

  metadata = engine.get_tokenizer()
  vocab = token_utils.load_vocab(metadata.path, metadata.extra_ids)
  stop_tokens = [vocab.eos_id, vocab.pad_id]
  max_output_length = 1024

  profiling_output = FLAGS.profiling_output
  if profiling_output:
    jax.profiler.start_trace(profiling_output)

  engine.init_decode_state()
  prompts: List[str] = [
      "I believe the meaning of life is",
      # pylint: disable-next=all
      "To add an element to an ArrayList of a specific class type in Java, you can follow the following steps:\n\n1. Create an instance of the class to be added.\n2. Get a reference to the ArrayList.\n3. Call the `add()` method on the ArrayList, passing the instance of the class as the argument.\n\nHere's an example of how to add an object of type `Person` to an ArrayList of type `ArrayList<Person>`:\n```csharp\n// Create a new instance of the Person class\nPerson person = new Person(\"John\", 25);\n\n// Get a reference to the ArrayList\nArrayList<Person> peopleList = new ArrayList<>();\n\n// Add the person object to the ArrayList\npeopleList.add(person);\n```\nIn this example, the `Person` class is assumed to have a constructor that takes two arguments: a String for the person's name, and an int for their age. You can substitute your own class and constructor as necessary.",
      # pylint: disable-next=all
      "<s>[INST] <<SYS>>\nYou are an AI assistant. User will you give you a task. Your goal is to complete the task as faithfully as you can. While performing the task think step-by-step and justify your steps.\n<</SYS>>\n\nQuestion 1: What is commercial real estate finance?\nQuestion 2: What are Commercial Real Estate services?\nOptions are:\n[a]. no.\n[b]. yes.\nWould the answer to these two questions be the same? [/INST]",
      # pylint: disable-next=all
      "<s>[INST] <<SYS>>\nYou are an AI assistant that helps people find information. Provide a detailed answer so user don\u2019t need to search outside to understand the answer.\n<</SYS>>\n\nUse reasoning to lead to the answer of the following question:\nWhere are you likely to find water underneath?\nOptions:\n- toilet\n- sink\n- jar\n- bridge\n- house\n Reasoning process: [/INST",
      # pylint: disable-next=all
      "<s>[INST] <<SYS>>\nYou are an AI assistant. You will be given a task. You must generate a detailed and long answer.\n<</SYS>>\n\nContinue the following story.\n\nKay didn't have shoes that fit her feet properly. She only wore sneakers, because the \nChoose from: [I] shoes  fitted badly. [II] sneakers  fitted badly. [/INST]",
  ]
  for prompt in prompts:
    slot = random.randint(0, FLAGS.batch_size - 1)
    tokens, true_length = token_utils.tokenize_and_pad(
        prompt, vocab, is_bos=True, jax_padding=False
    )
    print(f"---- Input prompts are: {prompt}")
    print(f"---- Encoded tokens are: {tokens}")

    # pylint: disable-next=all
    prefill_result, _ = engine.prefill(
        params=None, padded_tokens=tokens, true_length=true_length
    )
    # pylint: disable-next=all
    decode_state = engine.insert(prefill_result, None, slot=slot)
    sampled_tokens_list = []
    while True:
      # pylint: disable-next=all
      decode_state, result_tokens = engine.generate(None, decode_state)
      result_tokens = result_tokens.convert_to_numpy()

      slot_data = result_tokens.get_result_at_slot(slot)
      slot_tokens = slot_data.tokens
      slot_lengths = slot_data.lengths

      token_id = slot_tokens[slot, 0].item()
      if slot_lengths > max_output_length or token_id in stop_tokens:
        break

      sampled_tokens_list.append(token_id)

    print("---- All output tokens.")
    print(sampled_tokens_list)
    print("---- All output text.")
    print(vocab.tokenizer.decode(sampled_tokens_list))

  if profiling_output:
    jax.profiler.stop_trace()


if __name__ == "__main__":
  os.environ["TF_CPP_MIN_LOG_LEVEL"] = "0"
  app.run(main)