feat: Implement self-instruct and evolve-instruct synthetic data generation pipeline #720
Conversation
|
Important Review skippedAuto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the You can disable this status message by setting the Thank you for using CodeRabbit. We offer it for free to the OSS community and would appreciate your support in helping us grow. If you find it useful, would you consider giving us a shout-out on your favorite social media? TipsChatThere are 3 ways to chat with CodeRabbit:
Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments. CodeRabbit Commands (invoked as PR comments)
Additionally, you can add CodeRabbit Configuration File (
|
implementations/example.py
Outdated
There was a problem hiding this comment.
May be better to put the logic in a main() function and then run
if __name__ == "__main__":
main()
in the end
There was a problem hiding this comment.
Add a way to pass in tasks rather than always using the whole seed_tasks file?
implementations/synthetic_datagen/self_instruct/self_instruct_generator.py
Outdated
Show resolved
Hide resolved
Wendong-Fan
changed the title
andrei3131
changed the title
andrei3131
changed the title
implementations/synthetic_datagen/evolve_instruct/seed_files/alpaca_data.json
Outdated
Show resolved
Hide resolved
| """ | ||
|
|
||
| self.prompt_templates[Mutation.FRESH_START] = ( | ||
| self.prompt_templates['base'] |
There was a problem hiding this comment.
We could move the prompts to a separate file, similar to templates.py, in self-instruct
There was a problem hiding this comment.
We could break down the implementation into smaller methods for ease of readability
implementations/synthetic_datagen/evolve_instruct/evolve_instruct_generator.py
Outdated
Show resolved
Hide resolved
|
|
||
| import os | ||
|
|
||
| if isinstance(self.seed_data, str) and os.path.exists(self.seed_data): |
There was a problem hiding this comment.
let's try to use more descriptive variable names
implementations/synthetic_datagen/evolve_instruct/evolve_instruct_generator.py
Outdated
Show resolved
Hide resolved
implementations/example_evolve.py
Outdated
| type=str, | ||
| default="implementations/data/seed_files/alpaca_data.json", | ||
| ) | ||
| parser.add_argument("--column_names", nargs='+', default="instruction") |
There was a problem hiding this comment.
Let's move this to the jupyter notebook camel_demo_instruct.ipynb
| return ret | ||
|
|
||
|
|
||
| class GradioClientPipeline: |
There was a problem hiding this comment.
Is this necessary?
| return ret | ||
|
|
||
|
|
||
| class HFPipeline: |
There was a problem hiding this comment.
Is this necessary?
There was a problem hiding this comment.
If these pipelines are needed let's try to integrate them by creating specific agent systems
There was a problem hiding this comment.
Synthetic data does not need to be inside the package. Best to download it from huggingface.
| Union[ChatCompletion, Stream[ChatCompletionChunk]]: | ||
| `ChatCompletion` in the non-stream mode. | ||
| """ | ||
| print(messages) |
There was a problem hiding this comment.
IMO, you can use logging info to instead print func
| if isinstance(choice.message, list): | ||
| # If choice.message is a list, handle accordingly | ||
| # It's a check to fit with Nemotron model integration. | ||
| content = "".join( | ||
| [msg.content for msg in choice.message if msg.content] | ||
| ) | ||
| else: | ||
| content = choice.message.content or "" |
There was a problem hiding this comment.
Please help explain the reason why choice.message will be a list
| class NemotronRewardEvalAgent(BaseEvalAgentSystem): | ||
| def __init__(self) -> None: | ||
| self.nemotron = NvidiaModelV2(model_type=ModelType.NEMOTRON_4_REWARD) | ||
|
|
||
| def run_eval( | ||
| self, synthetic_datum: SyntheticDatum | ||
| ) -> Optional[Dict[str, Any]]: | ||
| message = [ | ||
| { | ||
| "role": "user", | ||
| "content": USER_PROMPT_TEMPLATE.format( | ||
| instruction=synthetic_datum.instruction, | ||
| input=synthetic_datum.input, | ||
| ), | ||
| }, | ||
| { | ||
| "role": "assistant", | ||
| "content": synthetic_datum.output, | ||
| }, | ||
| ] | ||
| response = self.nemotron.run(message) | ||
| print(response) | ||
| match = re.search(r"content='(.*?)'", str(response)) | ||
| if not match: | ||
| logger.error("Failed to parse response from Nemotron. {response}") | ||
| return None | ||
|
|
||
| scores_str = match.group(1) | ||
| logger.info(f"Scores: {scores_str}") | ||
| scores = { | ||
| item.split(":")[0]: float(item.split(":")[1]) | ||
| for item in scores_str.split(",") | ||
| } | ||
|
|
||
| return scores |
There was a problem hiding this comment.
need code comments in here
| @dataclass | ||
| class SyntheticDatum: | ||
| instruction: str | ||
| input: str |
There was a problem hiding this comment.
we are moving from dataclass to pydantic BaseModel, maybe better to use pydantic BaseModel?
There was a problem hiding this comment.
Is there an example in camel of using pydantic BaseModel?
There was a problem hiding this comment.
I feel it's not necessary to set this SingleAgent class, in this class the model type is hard coded as ollama, why not calling agent from ChatAgent from camel.agents?
| Returns: | ||
| str: The response generated by the agent. | ||
| """ | ||
| for agent in self.agents: |
There was a problem hiding this comment.
how could it work? no self.agents
There was a problem hiding this comment.
what's this MultiAgent class used for?
There was a problem hiding this comment.
to initialize a multi-agent system with chat/eval agents etc. inside
| import re | ||
| from typing import Any, Dict, Optional | ||
|
|
||
| from camel.models import NvidiaModelV2 |
There was a problem hiding this comment.
if only NvidiaModelV2 is used, we can remove other models like NvidiaModelV3
| match = re.search(r"content='(.*?)'", str(response)) | ||
| if not match: | ||
| logger.error("Failed to parse response from Nemotron. {response}") | ||
| return None | ||
|
|
||
| scores_str = match.group(1) |
There was a problem hiding this comment.
we can get the score from response without using regex matching
There was a problem hiding this comment.
But in that way, it will give the error that we were discussing during the call
| raise NotImplementedError( | ||
| "This method must be implemented by subclasses." | ||
| ) | ||
|
|
There was a problem hiding this comment.
normally we use pass rather than rasing error
| @property | ||
| def is_self_instruct(self) -> bool: | ||
| return self is SyntheticDataGeneratorMethodType.SELFINSTRUCT |
There was a problem hiding this comment.
why we need to check this? I think we can do the check by using .value
Description
Implementation of a generic pipeline for synthetic data generation designed to support multiple methods, starting with self-instruct (credits to https://github.com/yizhongw/self-instruct).
Motivation and Context
New project: synthetic data generation
Types of changes
What types of changes does your code introduce? Put an
xin all the boxes that apply:Implemented Tasks
Checklist
Go over all the following points, and put an
xin all the boxes that apply.If you are unsure about any of these, don't hesitate to ask. We are here to help!