HF and agent miners

.gitignore

@@ -3,6 +3,8 @@ __pycache__/
 *.py[cod]
 *$py.class
 .DS_Store
+**/.DS_Store
+
 
 # C extensions
 *.so

neurons/miners/huggingface/README.md

@@ -0,0 +1,47 @@
+# Hugging Face Bittensor Miner
+This repository contains a Bittensor Miner integrated with 🤗 Hugging Face pipelines. The miner connects to the Bittensor network, registers its wallet, and serves a hugging face model to the network.
+
+## Prerequisites
+
+- Python 3.8+
+- OpenAI Python API (https://github.com/openai/openai)
+
+## Installation
+1. Clone the repository 
+```bash
+git clone https://github.com/opentensor/prompting.git
+```
+2. Install the required packages for the [repository requirements](../../../requirements.txt) with `pip install -r requirements.txt`
+
+
+For more configuration options related to the wallet, axon, subtensor, logging, and metagraph, please refer to the Bittensor documentation.
+
+## Example Usage
+
+Here are some model examples that could be leveraged by the HuggingFace Miner, alongside suggested GPU footprint to run the models comfortably:
+| model_id | Default GPU footprint | 8bits quantization GPU footprint | 4bits quantization GPU footprint |
+| --- | ---- | ---- | ---- |  
+| HuggingFaceH4/zephyr-7b-beta | 18 GB | 12 GB | 7 GB |
+| teknium/OpenHermes-2.5-Mistral-7B | 30 GB | 10 GB | 7 GB |
+| upstage/SOLAR-10.7B-Instruct-v1.0 | 42 GB | 14 GB| 8 GB |
+| mistralai/Mixtral-8x7B-Instruct-v0.1 | 92 GB* | 64 GB* | 30 GB* |
+
+> \* Big models such as mixtral are very costly to run and optimize, so always bear in mind the trade-offs between model speed, model quality and infra cost.
+
+
+To run the Hugging Face Bittensor Miner with default settings, use the following command:
+```bash
+python3 neurons/miners/huggingface/miner.py \
+    --wallet.name <<your-wallet-name>> \
+    --wallet.hotkey <<your-hotkey>>
+    --neuron.model_id <<model_id>>
+```
+
+You can also run automatic quantization by adding the flag `--neuron.load_in_8bit` for 8bits quantization and `--neuron.load_in_4bit` for 4 bits quantization:
+```bash
+python3 neurons/miners/huggingface/miner.py \
+    --wallet.name <<your-wallet-name>> \
+    --wallet.hotkey <<your-hotkey>>
+    --neuron.model_id <<model_id>>
+    --neuron.load_in_8bit True
+```

neurons/miners/huggingface/miner.py

@@ -0,0 +1,31 @@
+# The MIT License (MIT)
+# Copyright © 2024 Yuma Rao
+
+# Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated
+# documentation files (the “Software”), to deal in the Software without restriction, including without limitation
+# the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software,
+# and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
+
+# The above copyright notice and this permission notice shall be included in all copies or substantial portions of
+# the Software.
+
+# THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
+# THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+# OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+# DEALINGS IN THE SOFTWARE.
+import time
+import bittensor as bt
+from prompting.miners import HuggingFaceMiner
+
+
+# This is the main function, which runs the miner.
+if __name__ == "__main__":
+    with HuggingFaceMiner() as miner:
+        while True:
+            miner.log_status()
+            time.sleep(5)
+
+            if miner.should_exit:
+                bt.logging.warning("Ending miner...")
+                break

neurons/miners/openai/miner.py

@@ -14,140 +14,9 @@
 # THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
 # OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 # DEALINGS IN THE SOFTWARE.
-
-import os
 import time
 import bittensor as bt
-import argparse
-
-# Bittensor Miner Template:
-import prompting
-from prompting.protocol import PromptingSynapse
-
-# import base miner class which takes care of most of the boilerplate
-from neurons.miner import Miner
-
-from langchain.prompts import ChatPromptTemplate
-from langchain_core.output_parsers import StrOutputParser
-from langchain.chat_models import ChatOpenAI
-from dotenv import load_dotenv, find_dotenv
-from langchain.callbacks import get_openai_callback
-
-
-class OpenAIMiner(Miner):
-    """Langchain-based miner which uses OpenAI's API as the LLM.
-
-    You should also install the dependencies for this miner, which can be found in the requirements.txt file in this directory.
-    """
-
-    @classmethod
-    def add_args(cls, parser: argparse.ArgumentParser):
-        """
-        Adds OpenAI-specific arguments to the command line parser.
-        """
-        super().add_args(parser)
-
-    def __init__(self, config=None):
-        super().__init__(config=config)
-
-        bt.logging.info(f"Initializing with model {self.config.neuron.model_id}...")
-
-        if self.config.wandb.on:
-            self.identity_tags = ("openai_miner",) + (self.config.neuron.model_id,)
-
-        _ = load_dotenv(find_dotenv())
-        api_key = os.environ.get("OPENAI_API_KEY")
-
-        # Set openai key and other args
-        self.model = ChatOpenAI(
-            api_key=api_key,
-            model_name=self.config.neuron.model_id,
-            max_tokens=self.config.neuron.max_tokens,
-            temperature=self.config.neuron.temperature,
-        )
-
-        self.system_prompt = "You are a friendly chatbot who always responds concisely and helpfully. You are honest about things you don't know."
-        self.accumulated_total_tokens = 0
-        self.accumulated_prompt_tokens = 0
-        self.accumulated_completion_tokens = 0
-        self.accumulated_total_cost = 0
-
-    def get_cost_logging(self, cb):
-        bt.logging.info(f"Total Tokens: {cb.total_tokens}")
-        bt.logging.info(f"Prompt Tokens: {cb.prompt_tokens}")
-        bt.logging.info(f"Completion Tokens: {cb.completion_tokens}")
-        bt.logging.info(f"Total Cost (USD): ${round(cb.total_cost,4)}")
-
-        self.accumulated_total_tokens += cb.total_tokens
-        self.accumulated_prompt_tokens += cb.prompt_tokens
-        self.accumulated_completion_tokens += cb.completion_tokens
-        self.accumulated_total_cost += cb.total_cost
-
-        return {
-            "total_tokens": cb.total_tokens,
-            "prompt_tokens": cb.prompt_tokens,
-            "completion_tokens": cb.completion_tokens,
-            "total_cost": cb.total_cost,
-            "accumulated_total_tokens": self.accumulated_total_tokens,
-            "accumulated_prompt_tokens": self.accumulated_prompt_tokens,
-            "accumulated_completion_tokens": self.accumulated_completion_tokens,
-            "accumulated_total_cost": self.accumulated_total_cost,
-        }
-
-    async def forward(self, synapse: PromptingSynapse) -> PromptingSynapse:
-        """
-        Processes the incoming synapse by performing a predefined operation on the input data.
-        This method should be replaced with actual logic relevant to the miner's purpose.
-
-        Args:
-            synapse (PromptingSynapse): The synapse object containing the 'dummy_input' data.
-
-        Returns:
-            PromptingSynapse: The synapse object with the 'dummy_output' field set to twice the 'dummy_input' value.
-
-        The 'forward' function is a placeholder and should be overridden with logic that is appropriate for
-        the miner's intended operation. This method demonstrates a basic transformation of input data.
-        """
-        try:
-            with get_openai_callback() as cb:
-                t0 = time.time()
-                bt.logging.debug(f"📧 Message received, forwarding synapse: {synapse}")
-
-                prompt = ChatPromptTemplate.from_messages(
-                    [("system", self.system_prompt), ("user", "{input}")]
-                )
-                chain = prompt | self.model | StrOutputParser()
-
-                role = synapse.roles[-1]
-                message = synapse.messages[-1]
-
-                bt.logging.debug(f"💬 Querying openai: {prompt}")
-                response = chain.invoke({"role": role, "input": message})
-
-                synapse.completion = response
-                synapse_latency = time.time() - t0
-
-                if self.config.wandb.on:
-                    self.log_event(
-                        timing=synapse_latency,
-                        prompt=message,
-                        completion=response,
-                        system_prompt=self.system_prompt,
-                        extra_info=self.get_cost_logging(cb),
-                    )
-
-            bt.logging.debug(f"✅ Served Response: {response}")
-            self.step += 1
-
-            return synapse
-        except Exception as e:
-            bt.logging.error(f"Error in forward: {e}")
-            synapse.completion = "Error: " + str(e)
-        finally:
-            if self.config.neuron.stop_on_forward_exception:
-                self.should_exit = True
-            return synapse
-
+from prompting.miners import OpenAIMiner
 
 # This is the main function, which runs the miner.
 if __name__ == "__main__":

neurons/miners/test/echo.py

@@ -14,37 +14,9 @@
 # THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
 # OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 # DEALINGS IN THE SOFTWARE.
-
 import time
-import typing
 import bittensor as bt
-
-# Bittensor Miner Template:
-import prompting
-from prompting.protocol import PromptingSynapse
-
-# import base miner class which takes care of most of the boilerplate
-from neurons.miner import Miner
-
-
-class EchoMiner(Miner):
-    """
-    This little fella just repeats the last message it received.
-    """
-
-    def __init__(self, config=None):
-        super().__init__(config=config)
-
-    async def forward(self, synapse: PromptingSynapse) -> PromptingSynapse:
-        synapse.completion = synapse.messages[-1]
-        self.step += 1
-        return synapse
-
-    async def blacklist(self, synapse: PromptingSynapse) -> typing.Tuple[bool, str]:
-        return False, "All good here"
-
-    async def priority(self, synapse: PromptingSynapse) -> float:
-        return 1e6
+from prompting.miners import EchoMiner
 
 
 # This is the main function, which runs the miner.

neurons/miners/test/mock.py

@@ -14,37 +14,9 @@
 # THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
 # OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 # DEALINGS IN THE SOFTWARE.
-
 import time
-import typing
 import bittensor as bt
-
-# Bittensor Miner Template:
-import prompting
-from prompting.protocol import PromptingSynapse
-
-# import base miner class which takes care of most of the boilerplate
-from neurons.miner import Miner
-
-
-class MockMiner(Miner):
-    """
-    This little fella responds with a static message.
-    """
-
-    def __init__(self, config=None):
-        super().__init__(config=config)
-
-    async def forward(self, synapse: PromptingSynapse) -> PromptingSynapse:
-        synapse.completion = f"Hey you reached mock miner {self.config.wallet.hotkey!r}. Please leave a message after the tone.. Beep!"
-        self.step += 1
-        return synapse
-
-    async def blacklist(self, synapse: PromptingSynapse) -> typing.Tuple[bool, str]:
-        return False, "All good here"
-
-    async def priority(self, synapse: PromptingSynapse) -> float:
-        return 1e6
+from prompting.miners import MockMiner
 
 
 # This is the main function, which runs the miner.

neurons/miners/test/phrase.py

@@ -14,49 +14,9 @@
 # THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
 # OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 # DEALINGS IN THE SOFTWARE.
-
 import time
-import typing
-import argparse
 import bittensor as bt
-
-# Bittensor Miner Template:
-import prompting
-from prompting.protocol import PromptingSynapse
-
-# import base miner class which takes care of most of the boilerplate
-from neurons.miner import Miner
-
-
-class PhraseMiner(Miner):
-    """
-    This little fella responds with whatever phrase you give it.
-    """
-
-    @classmethod
-    def add_args(cls, parser: argparse.ArgumentParser):
-        super().add_args(parser)
-
-        parser.add_argument(
-            "--neuron.phrase",
-            type=str,
-            help="The phrase to use when running a phrase (test) miner.",
-            default="Can you please repeat that?",
-        )
-
-    def __init__(self, config=None):
-        super().__init__(config=config)
-
-    async def forward(self, synapse: PromptingSynapse) -> PromptingSynapse:
-        synapse.completion = self.config.neuron.phrase
-        self.step += 1
-        return synapse
-
-    async def blacklist(self, synapse: PromptingSynapse) -> typing.Tuple[bool, str]:
-        return False, "All good here"
-
-    async def priority(self, synapse: PromptingSynapse) -> float:
-        return 1e6
+from prompting.miners import PhraseMiner
 
 
 # This is the main function, which runs the miner.