Test replit-code-v1-3b model

[abetlen]

Replit recently trained a 3B parameter Llama-style code model with some very promising results. Weights have been released here

[execveat]

The demo looks extremely underwhelming. Nowhere close to Codex/Copilot: https://huggingface.co/spaces/replit/replit-code-v1-3b-demo

[ggerganov]

Would be nice to get a breakdown of the differences with the LLaMA architecture to get a feeling of how big of a task this would be

[Heath123]

The demo looks extremely underwhelming. Nowhere close to Codex/Copilot: huggingface.co/spaces/replit/replit-code-v1-3b-demo

Do you expect it to be at 3B parameters? It might not be as good as Copilot but it's still very good

[abetlen]

Would be nice to get a breakdown of the differences with the LLaMA architecture to get a feeling of how big of a task this would be

Ah looks like I misread their press release, I took Llama-style to mean the exact architecture not that it was trained past Chinchilla optimallity. I'll try to look into this.

The demo looks extremely underwhelming. Nowhere close to Codex/Copilot: https://huggingface.co/spaces/replit/replit-code-v1-3b-demo

I think the better comparison is against the salesforce codegen models as they're the best option for self-hosted code completion, would be cool to build something like turbopilot but this may require a seperate ggml implementation for this model.

[Green-Sky]

replit-code-v1-3b is powered by state-of-the-art LLM techniques, such as: Flash Attention for fast training and inference, AliBi positional embeddings to support variable context length at inference time, LionW optimizer, etc.

alibi got merged recently

[v3ss0n]

The demo looks extremely underwhelming. Nowhere close to Codex/Copilot: https://huggingface.co/spaces/replit/replit-code-v1-3b-demo

Can you show your results? Mine looks quite good and already usable, this with chat tuned lora will be quite amazing

""" SQLAlchemy model for relationship betweek tasks and projects """ # < thats the only input i gave
class Project(Base):
    __tablename__ = 'project'

    id = Column(Integer, primary_key=True)
    name = Column(String, nullable=False)
    description = Column(String, nullable=False)
    tasks = relationship('Task', back_populates='project')

    def __init__(self, name, description):
        self.name = name
        self.description = description

    def __repr__(self):
        return f"<Project(name='{self.name}', description='{self.description}')>"

class Task(Base):
    __tablename__ = 'task'

    id = Column(Integer, primary_key=True)
    name = Column(String, nullable=False)
    description = Column(String, nullable=False)
    project_id = Column(Integer, ForeignKey('project.id'))
    project = relationship('Project', back_populates='tasks')

    def __init__(self, name, description, project_id):
        self.name = name
        self.description = description
        self.project_id = project_id

    def __repr__(self):
        return f"<Task(name='{self.name}', description='{self.description}', project_id='{self.project_id}')>"

[ElYaiko]

The demo looks extremely underwhelming. Nowhere close to Codex/Copilot: https://huggingface.co/spaces/replit/replit-code-v1-3b-demo

Can you show your results? Mine looks quite good and already usable, this with chat tuned lora will be quite amazing

""" SQLAlchemy model for relationship betweek tasks and projects """ # < thats the only input i gave
class Project(Base):
    __tablename__ = 'project'

    id = Column(Integer, primary_key=True)
    name = Column(String, nullable=False)
    description = Column(String, nullable=False)
    tasks = relationship('Task', back_populates='project')

    def __init__(self, name, description):
        self.name = name
        self.description = description

    def __repr__(self):
        return f"<Project(name='{self.name}', description='{self.description}')>"

class Task(Base):
    __tablename__ = 'task'

    id = Column(Integer, primary_key=True)
    name = Column(String, nullable=False)
    description = Column(String, nullable=False)
    project_id = Column(Integer, ForeignKey('project.id'))
    project = relationship('Project', back_populates='tasks')

    def __init__(self, name, description, project_id):
        self.name = name
        self.description = description
        self.project_id = project_id

    def __repr__(self):
        return f"<Task(name='{self.name}', description='{self.description}', project_id='{self.project_id}')>"

Yes, it gives ok results
Is there any chance of using this model with ggml?