added support for custom openai api

[sebaxzero]

Support for a custom openai api, using default llama_index.llms.openai (no requirement change)

• OpenAI compatable API #39 (comment)
• Request: Add OPENAI_API_URL #1 (comment)

backend:

• created "CUSTOM" model mapping with "gpt-4" as its name for llama_index to use 8k context.
• added elif statement to chat, related queries and validator for ChatModel.CUSTOM

frontend:

• added Custom API as selectable model.

docker:

• added CUSTOM_HOST and CUSTOM_API_KEY env variables with lm-studio url set by default (http://localhost:1234/v1)

enviromental variables:

CUSTOM_HOST=your-custom-host
CUSTOM_API_KEY=your-custom-api-key

.env example for lm-studio server:

CUSTOM_HOST=http://localhost:1234/v1
CUSTOM_API_KEY=local

api-key in most cases is not nedded.

why use custom instead of openai naming?

llama_index uses OPENAI_API_BASE while openai uses OPENAI_BASE_URL
adding the new variable make sure it does not conflict with cloud openai usage.

tested with lm-studio local server using:

• llama3 8B (also some finetunes versions like Herme 2 Theta)
• mistral v0.3 7B
• phi 3 mini

preview:

[vercel]

@sebaxzero is attempting to deploy a commit to the rashadphil's projects Team on Vercel.

A member of the Team first needs to authorize it.

[ramanveerji]

Support for a custom openai api, using default llama_index.llms.openai (no requirement change)

• OpenAI compatable API #39 (comment)
• Request: Add OPENAI_API_URL #1 (comment)

backend:

• created "CUSTOM" model mapping with "gpt-4" as its name for llama_index to use 8k context.
• added elif statement to chat, related queries and validator for ChatModel.CUSTOM

frontend:

• added Custom API as selectable model.

docker:

• added CUSTOM_HOST and CUSTOM_API_KEY env variables with lm-studio url set by default (http://localhost:1234/v1)

enviromental variables:

CUSTOM_HOST=your-custom-host
CUSTOM_API_KEY=your-custom-api-key

.env example for lm-studio server:

CUSTOM_HOST=http://localhost:1234/v1
CUSTOM_API_KEY=local

api-key in most cases is not nedded.

why use custom instead of openai naming?

llama_index uses OPENAI_API_BASE while openai uses OPENAI_BASE_URL adding the new variable make sure it does not conflict with cloud openai usage.

tested with lm-studio local server using:

• llama3 8B (also some finetunes versions like Herme 2 Theta)
• mistral v0.3 7B
• phi 3 mini

preview:

can you add cohere support for search also?