Duplicate from mosaicml/mpt-7b-instruct

Co-authored-by: Sam <sam-mosaic@users.noreply.huggingface.co>

.gitattributes

@@ -0,0 +1,34 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,13 @@
+---
+title: MPT-7B-Instruct
+emoji: 💁
+colorFrom: yellow
+colorTo: purple
+sdk: gradio
+sdk_version: 3.28.0
+app_file: app.py
+pinned: false
+duplicated_from: mosaicml/mpt-7b-instruct
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,246 @@
+# Copyright 2023 MosaicML spaces authors
+# SPDX-License-Identifier: Apache-2.0
+# and
+# the https://huggingface.co/spaces/HuggingFaceH4/databricks-dolly authors
+import datetime
+import os
+from threading import Event, Thread
+from uuid import uuid4
+
+import gradio as gr
+import requests
+import torch
+from transformers import StoppingCriteria, StoppingCriteriaList, TextIteratorStreamer
+
+from quick_pipeline import InstructionTextGenerationPipeline as pipeline
+
+
+# Configuration
+HF_TOKEN = os.getenv("HF_TOKEN", None)
+
+examples = [
+    # to do: add coupled hparams so e.g. poem has higher temp
+    "Write a travel blog about a 3-day trip to Thailand.",
+    "Write a short story about a robot that has a nice day.",
+    "Convert the following to a single line of JSON:\n\n```name: John\nage: 30\naddress:\n  street:123 Main St.\n  city: San Francisco\n  state: CA\n  zip: 94101\n```",
+    "Write a quick email to congratulate MosaicML about the launch of their inference offering.",
+    "Explain how a candle works to a 6 year old in a few sentences.",
+    "What are some of the most common misconceptions about birds?",
+]
+
+# Initialize the model and tokenizer
+generate = pipeline(
+    "mosaicml/mpt-7b-instruct",
+    torch_dtype=torch.bfloat16,
+    trust_remote_code=True,
+    use_auth_token=HF_TOKEN,
+)
+stop_token_ids = generate.tokenizer.convert_tokens_to_ids(["<|endoftext|>"])
+
+
+# Define a custom stopping criteria
+class StopOnTokens(StoppingCriteria):
+    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
+        for stop_id in stop_token_ids:
+            if input_ids[0][-1] == stop_id:
+                return True
+        return False
+
+
+def log_conversation(session_id, instruction, response, generate_kwargs):
+    logging_url = os.getenv("LOGGING_URL", None)
+    if logging_url is None:
+        return
+
+    timestamp = datetime.datetime.now().strftime("%Y-%m-%dT%H:%M:%S")
+
+    data = {
+        "session_id": session_id,
+        "timestamp": timestamp,
+        "instruction": instruction,
+        "response": response,
+        "generate_kwargs": generate_kwargs,
+    }
+
+    try:
+        requests.post(logging_url, json=data)
+    except requests.exceptions.RequestException as e:
+        print(f"Error logging conversation: {e}")
+
+
+def process_stream(instruction, temperature, top_p, top_k, max_new_tokens, session_id):
+    # Tokenize the input
+    input_ids = generate.tokenizer(
+        generate.format_instruction(instruction), return_tensors="pt"
+    ).input_ids
+    input_ids = input_ids.to(generate.model.device)
+
+    # Initialize the streamer and stopping criteria
+    streamer = TextIteratorStreamer(
+        generate.tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True
+    )
+    stop = StopOnTokens()
+
+    if temperature < 0.1:
+        temperature = 0.0
+        do_sample = False
+    else:
+        do_sample = True
+
+    gkw = {
+        **generate.generate_kwargs,
+        **{
+            "input_ids": input_ids,
+            "max_new_tokens": max_new_tokens,
+            "temperature": temperature,
+            "do_sample": do_sample,
+            "top_p": top_p,
+            "top_k": top_k,
+            "streamer": streamer,
+            "stopping_criteria": StoppingCriteriaList([stop]),
+        },
+    }
+
+    response = ""
+    stream_complete = Event()
+
+    def generate_and_signal_complete():
+        generate.model.generate(**gkw)
+        stream_complete.set()
+
+    def log_after_stream_complete():
+        stream_complete.wait()
+        log_conversation(
+            session_id,
+            instruction,
+            response,
+            {
+                "top_k": top_k,
+                "top_p": top_p,
+                "temperature": temperature,
+            },
+        )
+
+    t1 = Thread(target=generate_and_signal_complete)
+    t1.start()
+
+    t2 = Thread(target=log_after_stream_complete)
+    t2.start()
+
+    for new_text in streamer:
+        response += new_text
+        yield response
+
+
+with gr.Blocks(
+    theme=gr.themes.Soft(),
+    css=".disclaimer {font-variant-caps: all-small-caps;}",
+) as demo:
+    session_id = gr.State(lambda: str(uuid4()))
+    gr.Markdown(
+        """<h1><center>MosaicML MPT-7B-Instruct</center></h1>
+
+        This demo is of [MPT-7B-Instruct](https://huggingface.co/mosaicml/mpt-7b-instruct). It is based on [MPT-7B](https://huggingface.co/mosaicml/mpt-7b) fine-tuned with approximately [60,000 instruction demonstrations](https://huggingface.co/datasets/sam-mosaic/dolly_hhrlhf)
+
+        If you're interested in [training](https://www.mosaicml.com/training) and [deploying](https://www.mosaicml.com/inference) your own MPT or LLMs, [sign up](https://forms.mosaicml.com/demo?utm_source=huggingface&utm_medium=referral&utm_campaign=mpt-7b) for MosaicML platform.
+
+        This is running on a smaller, shared GPU, so it may take a few seconds to respond. If you want to run it on your own GPU, you can [download the model from HuggingFace](https://huggingface.co/mosaicml/mpt-7b-instruct) and run it locally. Or [Duplicate the Space](https://huggingface.co/spaces/mosaicml/mpt-7b-instruct?duplicate=true) to skip the queue and run in a private space."""
+    )
+    with gr.Row():
+        with gr.Column():
+            with gr.Row():
+                instruction = gr.Textbox(
+                    placeholder="Enter your question here",
+                    label="Question/Instruction",
+                    elem_id="q-input",
+                )
+            with gr.Accordion("Advanced Options:", open=False):
+                with gr.Row():
+                    with gr.Column():
+                        with gr.Row():
+                            temperature = gr.Slider(
+                                label="Temperature",
+                                value=0.1,
+                                minimum=0.0,
+                                maximum=1.0,
+                                step=0.1,
+                                interactive=True,
+                                info="Higher values produce more diverse outputs",
+                            )
+                    with gr.Column():
+                        with gr.Row():
+                            top_p = gr.Slider(
+                                label="Top-p (nucleus sampling)",
+                                value=1.0,
+                                minimum=0.0,
+                                maximum=1,
+                                step=0.01,
+                                interactive=True,
+                                info=(
+                                    "Sample from the smallest possible set of tokens whose cumulative probability "
+                                    "exceeds top_p. Set to 1 to disable and sample from all tokens."
+                                ),
+                            )
+                    with gr.Column():
+                        with gr.Row():
+                            top_k = gr.Slider(
+                                label="Top-k",
+                                value=0,
+                                minimum=0.0,
+                                maximum=200,
+                                step=1,
+                                interactive=True,
+                                info="Sample from a shortlist of top-k tokens — 0 to disable and sample from all tokens.",
+                            )
+                    with gr.Column():
+                        with gr.Row():
+                            max_new_tokens = gr.Slider(
+                                label="Maximum new tokens",
+                                value=256,
+                                minimum=0,
+                                maximum=1664,
+                                step=5,
+                                interactive=True,
+                                info="The maximum number of new tokens to generate",
+                            )
+    with gr.Row():
+        submit = gr.Button("Submit")
+    with gr.Row():
+        with gr.Box():
+            gr.Markdown("**MPT-7B-Instruct**")
+            output_7b = gr.Markdown()
+
+    with gr.Row():
+        gr.Examples(
+            examples=examples,
+            inputs=[instruction],
+            cache_examples=False,
+            fn=process_stream,
+            outputs=output_7b,
+        )
+    with gr.Row():
+        gr.Markdown(
+            "Disclaimer: MPT-7B can produce factually incorrect output, and should not be relied on to produce "
+            "factually accurate information. MPT-7B was trained on various public datasets; while great efforts "
+            "have been taken to clean the pretraining data, it is possible that this model could generate lewd, "
+            "biased, or otherwise offensive outputs.",
+            elem_classes=["disclaimer"],
+        )
+    with gr.Row():
+        gr.Markdown(
+            "[Privacy policy](https://gist.github.com/samhavens/c29c68cdcd420a9aa0202d0839876dac)",
+            elem_classes=["disclaimer"],
+        )
+
+    submit.click(
+        process_stream,
+        inputs=[instruction, temperature, top_p, top_k, max_new_tokens, session_id],
+        outputs=output_7b,
+    )
+    instruction.submit(
+        process_stream,
+        inputs=[instruction, temperature, top_p, top_k, max_new_tokens, session_id],
+        outputs=output_7b,
+    )
+
+demo.queue(max_size=32, concurrency_count=4).launch(debug=True)

quick_pipeline.py

@@ -0,0 +1,85 @@
+from typing import Any, Dict, Tuple
+import warnings
+
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+
+INSTRUCTION_KEY = "### Instruction:"
+RESPONSE_KEY = "### Response:"
+END_KEY = "### End"
+INTRO_BLURB = "Below is an instruction that describes a task. Write a response that appropriately completes the request."
+PROMPT_FOR_GENERATION_FORMAT = """{intro}
+
+{instruction_key}
+{instruction}
+
+{response_key}
+""".format(
+    intro=INTRO_BLURB,
+    instruction_key=INSTRUCTION_KEY,
+    instruction="{instruction}",
+    response_key=RESPONSE_KEY,
+)
+
+
+class InstructionTextGenerationPipeline:
+    def __init__(
+        self,
+        model_name,
+        torch_dtype=torch.bfloat16,
+        trust_remote_code=True,
+        use_auth_token=None,
+    ) -> None:
+        self.model = AutoModelForCausalLM.from_pretrained(
+            model_name,
+            torch_dtype=torch_dtype,
+            trust_remote_code=trust_remote_code,
+            use_auth_token=use_auth_token,
+        )
+
+        tokenizer = AutoTokenizer.from_pretrained(
+            model_name,
+            trust_remote_code=trust_remote_code,
+            use_auth_token=use_auth_token,
+        )
+        if tokenizer.pad_token_id is None:
+            warnings.warn(
+                "pad_token_id is not set for the tokenizer. Using eos_token_id as pad_token_id."
+            )
+            tokenizer.pad_token = tokenizer.eos_token
+        tokenizer.padding_side = "left"
+        self.tokenizer = tokenizer
+
+        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        self.model.eval()
+        self.model.to(device=device, dtype=torch_dtype)
+
+        self.generate_kwargs = {
+            "temperature": 0.5,
+            "top_p": 0.92,
+            "top_k": 0,
+            "max_new_tokens": 512,
+            "use_cache": True,
+            "do_sample": True,
+            "eos_token_id": self.tokenizer.eos_token_id,
+            "pad_token_id": self.tokenizer.pad_token_id,
+            "repetition_penalty": 1.1,  # 1.0 means no penalty, > 1.0 means penalty, 1.2 from CTRL paper
+        }
+
+    def format_instruction(self, instruction):
+        return PROMPT_FOR_GENERATION_FORMAT.format(instruction=instruction)
+
+    def __call__(
+        self, instruction: str, **generate_kwargs: Dict[str, Any]
+    ) -> Tuple[str, str, float]:
+        s = PROMPT_FOR_GENERATION_FORMAT.format(instruction=instruction)
+        input_ids = self.tokenizer(s, return_tensors="pt").input_ids
+        input_ids = input_ids.to(self.model.device)
+        gkw = {**self.generate_kwargs, **generate_kwargs}
+        with torch.no_grad():
+            output_ids = self.model.generate(input_ids, **gkw)
+        # Slice the output_ids tensor to get only new tokens
+        new_tokens = output_ids[0, len(input_ids[0]) :]
+        output_text = self.tokenizer.decode(new_tokens, skip_special_tokens=True)
+        return output_text

requirements.txt

@@ -0,0 +1,4 @@
+-e git+https://github.com/samhavens/just-triton-flash.git#egg=flash_attn
+einops
+torch
+transformers