You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Im running on a Mac M3 Max, goal is to build the COG image here and then push it to replicate where I will be using Nvidia to run it.
Things I have tried: Use claude to help troubleshoot the problem but it keeps telling me: This error is occurring because Docker is trying to pull an image named "cog-mlcog" from Docker Hub, which doesn't exist.
But I have specified the model_path in predict.py so IDK why it's pulling from docker hub.
Here is the details of predict.py
from cog import BasePredictor, Input, Path
from llama_cpp import Llama
import os
class Predictor(BasePredictor):
def setup(self):
model_path = Path("./Users/username/Downloads/ML_COG/NeuralDaredevil-8B-abliterated.Q4_K_M.gguf")
# Check if CUDA is available
n_gpu_layers = -1 if "CUDA_VISIBLE_DEVICES" in os.environ else 0
self.model = Llama(
model_path=str(model_path),
n_ctx=2048,
n_gpu_layers=-1 # This will attempt to offload all layers to GPU
)
def predict(self, prompt: str = Input(description="Text prompt for generation")) -> str:
output = self.model(prompt, max_tokens=100, echo=True)
return output['choices'][0]['text']
So, im building a COG image for my LLM which I have downloaded from Huggingface: https://huggingface.co/QuantFactory/NeuralDaredevil-8B-abliterated-GGUF
Im running on a Mac M3 Max, goal is to build the COG image here and then push it to replicate where I will be using Nvidia to run it.
Things I have tried: Use claude to help troubleshoot the problem but it keeps telling me: This error is occurring because Docker is trying to pull an image named "cog-mlcog" from Docker Hub, which doesn't exist.
But I have specified the model_path in predict.py so IDK why it's pulling from docker hub.
Here is the details of predict.py
Here is the details of cog.yaml
Here is the error:
Tasks
The text was updated successfully, but these errors were encountered: