Productionize Chat demo (#1235)

* Add chatbot example rebased from master

* Update history to task
Add persistent session_id
Delete init

* Delete init.py
Update fire condition

* Move to src
Add cli callable

setup.py

@@ -298,6 +298,7 @@ def _setup_entry_points() -> Dict:
         "console_scripts": [
             f"deepsparse.transformers.run_inference={data_api_entrypoint}",
             f"deepsparse.transformers.eval_downstream={eval_downstream}",
+            "deepsparse.infer=deepsparse.transformers.infer:main",
             "deepsparse.debug_analysis=deepsparse.debug_analysis:main",
             "deepsparse.analyze=deepsparse.analyze:main",
             "deepsparse.check_hardware=deepsparse.cpu:print_hardware_capability",

src/deepsparse/tasks.py

@@ -230,7 +230,7 @@ def is_chat(cls, task: str) -> bool:
         :param task: the name of the task to check whether it is a chat task
         :return: True if it is a chat task, False otherwise
         """
-        return any([chat_task.matches(task) for chat_task in cls.chat])
+        return any(chat_task.matches(task) for chat_task in cls.chat)
 
     @classmethod
     def is_text_generation(cls, task: str) -> bool:

src/deepsparse/transformers/infer.py

@@ -0,0 +1,162 @@
+# Copyright (c) 2021 - present / Neuralmagic, Inc. All Rights Reserved.
+#
+# 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.
+
+"""
+Usage: deepsparse.infer [OPTIONS] MODEL_PATH
+
+  Command Line utility to interact with a text genration LLM in a chatbot
+  style
+
+  Example usage:
+
+  deepsparse.infer [OPTIONS] <MODEL_PATH>
+
+Options:
+  --sequence_length INTEGER       Sequence length to compile model and
+                                  tokenizer for.This controls the maximum
+                                  context length of the pipeline.  [default:
+                                  512]
+  --sampling_temperature FLOAT    The temperature to use when samplingfrom the
+                                  probability distribution computed from the
+                                  logits.Higher values will result in more
+                                  random samples. Shouldbe greater than 0.0.
+                                  [default: 1.0]
+  --prompt_sequence_length INTEGER
+                                  Processed prompt in chunks of this length.
+                                  This is to maximize the inference speed
+                                  [default: 64]
+  --show_tokens_per_sec / --no_show_tokens_per_sec
+                                  Whether to display the token generation
+                                  speed or not  [default:
+                                  no_show_tokens_per_sec]
+  --task TEXT                     The task to use for the pipeline. Choose any
+                                  of `chat`, `codegen`, `text-generation`
+                                  [default: chat]
+  --help                          Show this message and exit.
+
+Installation: pip install deepsparse[transformers]
+Examples:
+
+1) Use a local deployment directory
+deepsparse.infer models/llama/deployment
+
+2) Use a SparseZoo stub
+deepsparse.infer \
+    zoo:nlg/text_generation/codegen_mono-350m/pytorch/huggingface/bigpython_bigquery_thepile/base-none # noqa: E501
+
+3) Display token generation speed
+deepsparse.infer models/llama/deployment \
+    --show_tokens_per_sec
+
+4) Disable history
+deepsparse.infer models/llama/deployment \
+    --task text-generation
+"""
+import click
+
+from deepsparse import Pipeline
+from deepsparse.tasks import SupportedTasks
+
+
+@click.command(
+    context_settings=dict(
+        token_normalize_func=lambda x: x.replace("-", "_"), show_default=True
+    )
+)
+@click.argument("model_path", type=str)
+@click.option(
+    "--sequence_length",
+    type=int,
+    default=512,
+    help="Sequence length to compile model and tokenizer for."
+    "This controls the maximum context length of the pipeline.",
+)
+@click.option(
+    "--sampling_temperature",
+    type=float,
+    default=1.0,
+    help="The temperature to use when sampling"
+    "from the probability distribution computed from the logits."
+    "Higher values will result in more random samples. Should"
+    "be greater than 0.0.",
+)
+@click.option(
+    "--prompt_sequence_length",
+    type=int,
+    default=64,
+    help="Processed prompt in chunks of this length. "
+    "This is to maximize the inference speed",
+)
+@click.option(
+    "--show_tokens_per_sec/--no_show_tokens_per_sec",
+    default=False,
+    help="Whether to display the token generation speed or not",
+)
+@click.option(
+    "--task",
+    default="chat",
+    type=str,
+    help="The task to use for the pipeline. Choose any of "
+    "`chat`, `codegen`, `text-generation`",
+)
+def main(
+    model_path: str,
+    sequence_length: int,
+    sampling_temperature: float,
+    prompt_sequence_length: int,
+    show_tokens_per_sec: bool,
+    task: str,
+):
+    """
+    Command Line utility to interact with a text genration LLM in a chatbot style
+
+    Example usage:
+
+    deepsparse.infer [OPTIONS] <MODEL_PATH>
+    """
+    session_ids = "chatbot_cli_session"
+
+    pipeline = Pipeline.create(
+        task=task,  # let pipeline determine if task is supported
+        model_path=model_path,
+        sequence_length=sequence_length,
+        sampling_temperature=sampling_temperature,
+        prompt_sequence_length=prompt_sequence_length,
+    )
+
+    # continue prompts until a keyboard interrupt
+    while True:
+        input_text = input("User: ")
+        pipeline_inputs = {"prompt": [input_text]}
+
+        if SupportedTasks.is_chat(task):
+            pipeline_inputs["session_ids"] = session_ids
+
+        response = pipeline(**pipeline_inputs)
+        print("Bot: ", response.generations[0].text)
+        if show_tokens_per_sec:
+            times = pipeline.timer_manager.times
+            prefill_speed = (
+                1.0 * prompt_sequence_length / times["engine_prompt_prefill_single"]
+            )
+            generation_speed = 1.0 / times["engine_token_generation_single"]
+            print(
+                f"[prefill: {prefill_speed:.2f} tokens/sec]",
+                f"[decode: {generation_speed:.2f} tokens/sec]",
+                sep="\n",
+            )
+
+
+if __name__ == "__main__":
+    main()