fix: temp ..

Author: numb3r3

README.md

@@ -2,10 +2,24 @@
 
 OpenGPT is an open-source cloud-native of large multi-modal models (LMMs).
 
+OpenGPT is an open-source cloud-native large multi-modal models (LMMs) serving solution. 
+It is designed to simplify the deployment and management of large language models, on a distributed cluster of GPUs.
+
 **Warning**: This is an idea that I had and I wanted to try it out. 
 The design has not been implemented yet. 
 **The content of README.md is just a placeholder to remind me of what I want to do.**
 
+## Features
+
+OpenGPT provides the following features to make it easy to deploy and serve large multi-modal models (LMMs) in production:
+
+- Support for multi-modal models
+- Scalable architecture for handling high traffic loads
+- Optimized for low-latency inference
+- Automatic model partitioning and distribution across multiple GPUs
+- Centralized model management and monitoring
+- REST API for easy integration with existing applications
+
 You can learn more about OpenGPT’s [architecture in our documentation](https://opengpt.readthedocs.io/en/latest/).
 
 
@@ -142,6 +156,51 @@ response = requests.post(
 )
 ```
 
+## Kubernetes
+
+To deploy OpenGPT on your Kubernetes cluster, follow these steps:
+
+1. Install the OpenGPT operator on your Kubernetes cluster using Helm:
+
+    ```bash
+    helm install opengpt ./helm/opengpt --namespace opengpt
+    ```
+
+2. Create a custom resource for your GPT model:
+    
+    ```YAML
+    apiVersion: opengpt.io/v1alpha1
+    kind: GptModel
+    metadata:
+      name: my-gpt-model
+      namespace: opengpt
+    spec:
+      modelPath: s3://my-bucket/my-model
+      modelName: my-model
+      maxBatchSize: 16
+      inputShape:
+        - 1024
+        - 1024
+        - 3
+      outputShape:
+        - 1024
+        - 1024
+        - 3
+
+    ```
+   
+3. Apply the custom resource to your cluster:
+
+    ```bash
+   kubectl apply -f my-gpt-model.yaml
+    ```
+
+4. Monitor the status of your GPT model using the OpenGPT dashboard:
+
+    ```bash
+   kubectl port-forward -n opengpt svc/opengpt-dashboard 8080:80
+    ```
+
 ## Accessing models via API
 
 You can also access the online models via API. To do so, you can use the `inference_client` package:
@@ -209,4 +268,13 @@ Specifically, we implement the following fine-tuning methods:
 
 ## Documentation
 
-For more information, check out the [documentation](https://opengpt.readthedocs.io/en/latest/).
+For more information, check out the [documentation](https://opengpt.readthedocs.io/en/latest/).
+
+
+## Contributing
+
+We welcome contributions from the community! To contribute, please submit a pull request following our contributing guidelines.
+
+## License
+
+OpenGPT is licensed under the Apache License, Version 2.0. See LICENSE for the full license text.

opengpt/factory.py

@@ -40,8 +40,8 @@ def create_model_and_transforms(
 
         model_config = {
             'clip_model_name': 'ViT-L-14::openai',
-            'lang_model_name_or_path': 'facebook/opt-1.3b',
-            'tokenizer_name_or_path': 'facebook/opt-1.3b',
+            'lang_model_name_or_path': 'llama_7B',
+            'tokenizer_name_or_path': 'llama_7B',
         }
         return load_model_and_transforms(**model_config)
     else:

opengpt/helper.py

@@ -0,0 +1,4 @@
+def get_envs():
+    from torch.utils import collect_env
+
+    return collect_env.get_pretty_env_info()

pyproject.toml

@@ -38,16 +38,17 @@ build-backend = "poetry.core.masonry.api"
 [tool.poetry.dependencies]
 # Compatible Python versions
 python = ">=3.8"
-torch = ">=1.9.0" # a meta device requires torch >= 1.9.0
+torch = ">=1.9.0,<2.0.0" # a meta device requires torch >= 1.9.0
 click = "^8.1.3"
 numpy = "^1.21.2"
 einops = "^0.6.0"
 transformers = "^4.12.5"
-open_clip_torch = "^2.16.0"
+open_clip_torch = "~2.16.0"
+accelerate = "^0.18.0"
 
 # A list of all of the optional dependencies, some of which are included in the
 # below `extras`. They can be opted into by apps.
-open-flamingo = { version = "^0.0.2", optional = true }
+open-flamingo = { version = "^0.0", optional = true }
 
 [tool.poetry.extras]
 flamingo = ["open-flamingo"]