diff --git a/README.md b/README.md index ff8a3fa2..dc2da278 100644 --- a/README.md +++ b/README.md @@ -71,7 +71,7 @@ The configuration `configs/config.lite.yaml` does not require any expert models ### Quick Start -First replace `openai.key` and `huggingface.token` in `server/configs/config.default.yaml` with **your personal OpenAI Key** and **your Hugging Face Token**, or put them in the environment variables `OPENAI_API_KEY` and `HUGGINGFACE_ACCESS_TOKEN` respectively. Then run the following commands: +First replace `openai.key` and `huggingface.token` in `server/configs/config.default.yaml` with **your personal OpenAI Key** and **your Hugging Face Token**, or put them in the environment variables `OPENAI_API_KEY` and `HUGGINGFACE_ACCESS_TOKEN` respectively. Alternatively, you can use [MiniMax](https://www.minimaxi.com/) as the LLM provider by running with `--config configs/config.minimax.yaml` (set your MiniMax API key first). Then run the following commands: @@ -182,7 +182,7 @@ Welcome to Jarvis! A collaborative system that consists of an LLM as the control The server-side configuration file is `server/configs/config.default.yaml`, and some parameters are presented as follows: -+ `model`: LLM, currently supports `text-davinci-003`. We are working on integrating more open-source LLMs. ++ `model`: LLM, currently supports `text-davinci-003`, `gpt-4`, and [MiniMax](https://www.minimaxi.com/) models (`MiniMax-M2.7`, `MiniMax-M2.7-highspeed`, `MiniMax-M2.5`, `MiniMax-M2.5-highspeed`). We are working on integrating more open-source LLMs. + `inference_mode`: mode of inference endpoints + `local`: only use the local inference endpoints + `huggingface`: only use the Hugging Face Inference Endpoints **(free of local inference endpoints)** @@ -192,6 +192,14 @@ The server-side configuration file is `server/configs/config.default.yaml`, and + `standard` (RAM>16GB, ControlNet + Standard Pipelines) + `full` (RAM>42GB, All registered models) +#### LLM Provider + +Jarvis supports multiple LLM providers as the backbone controller. Configure the provider in the YAML config file: + ++ **OpenAI** (default): Set `openai.api_key` in config or the `OPENAI_API_KEY` environment variable. ++ **Azure OpenAI**: Set `azure.api_key`, `azure.base_url`, `azure.deployment_name`, and `azure.api_version` in config. ++ **MiniMax**: Set `minimax.api_key` in config or the `MINIMAX_API_KEY` environment variable. Use `model: MiniMax-M2.7` and `use_completion: false`. A ready-to-use config is provided at `server/configs/config.minimax.yaml`. MiniMax models offer a 204K token context window. Get your API key at [MiniMax Platform](https://www.minimaxi.com/). + On a personal laptop, we recommend the configuration of `inference_mode: hybrid `and `local_deployment: minimal`. But the available models under this setting may be limited due to the instability of remote Hugging Face Inference Endpoints. ### NVIDIA Jetson Embedded Device Support diff --git a/hugginggpt/server/awesome_chat.py b/hugginggpt/server/awesome_chat.py index 70c72053..ddd4a682 100644 --- a/hugginggpt/server/awesome_chat.py +++ b/hugginggpt/server/awesome_chat.py @@ -80,11 +80,13 @@ api_name = "chat/completions" API_TYPE = None -# priority: local > azure > openai +# priority: local > azure > minimax > openai if "dev" in config and config["dev"]: API_TYPE = "local" elif "azure" in config: API_TYPE = "azure" +elif "minimax" in config: + API_TYPE = "minimax" elif "openai" in config: API_TYPE = "openai" else: @@ -100,6 +102,14 @@ elif API_TYPE == "azure": API_ENDPOINT = f"{config['azure']['base_url']}/openai/deployments/{config['azure']['deployment_name']}/{api_name}?api-version={config['azure']['api_version']}" API_KEY = config["azure"]["api_key"] +elif API_TYPE == "minimax": + API_ENDPOINT = f"{config['minimax'].get('base_url', 'https://api.minimax.io/v1')}/{api_name}" + if config["minimax"]["api_key"] and config["minimax"]["api_key"] != "REPLACE_WITH_YOUR_MINIMAX_API_KEY_HERE": + API_KEY = config["minimax"]["api_key"] + elif "MINIMAX_API_KEY" in os.environ: + API_KEY = os.getenv("MINIMAX_API_KEY") + else: + raise ValueError(f"Incorrect MiniMax key. Please check your {args.config} file or set the MINIMAX_API_KEY environment variable.") elif API_TYPE == "openai": API_ENDPOINT = f"https://api.openai.com/v1/{api_name}" if config["openai"]["api_key"].startswith("sk-"): # Check for valid OpenAI key in config file @@ -190,9 +200,12 @@ def send_request(data): api_key = data.pop("api_key") api_type = data.pop("api_type") api_endpoint = data.pop("api_endpoint") + # MiniMax requires temperature in (0.0, 1.0]; adjust zero values + if api_type == "minimax" and data.get("temperature", 1) == 0: + data["temperature"] = 0.01 if use_completion: data = convert_chat_to_completion(data) - if api_type == "openai": + if api_type in ("openai", "minimax"): HEADER = { "Authorization": f"Bearer {api_key}" } diff --git a/hugginggpt/server/configs/config.minimax.yaml b/hugginggpt/server/configs/config.minimax.yaml new file mode 100644 index 00000000..5cc17f1e --- /dev/null +++ b/hugginggpt/server/configs/config.minimax.yaml @@ -0,0 +1,45 @@ +minimax: + api_key: REPLACE_WITH_YOUR_MINIMAX_API_KEY_HERE + base_url: https://api.minimax.io/v1 +# openai: +# api_key: REPLACE_WITH_YOUR_OPENAI_API_KEY_HERE +# azure: +# api_key: REPLACE_WITH_YOUR_AZURE_API_KEY_HERE +# base_url: REPLACE_WITH_YOUR_ENDPOINT_HERE +# deployment_name: REPLACE_WITH_YOUR_DEPLOYMENT_NAME_HERE +# api_version: "2022-12-01" +huggingface: + token: REPLACE_WITH_YOUR_HUGGINGFACE_TOKEN_HERE # required: huggingface token @ https://huggingface.co/settings/tokens +dev: false +debug: false +log_file: logs/debug.log +model: MiniMax-M2.7 # MiniMax models: MiniMax-M2.7, MiniMax-M2.7-highspeed, MiniMax-M2.5, MiniMax-M2.5-highspeed (204K context) +use_completion: false # MiniMax uses chat/completions endpoint +inference_mode: huggingface # local, huggingface or hybrid, prefer hybrid +local_deployment: minimal # minimal, standard or full, prefer full +num_candidate_models: 5 +max_description_length: 100 +proxy: # optional: your proxy server "http://ip:port" +http_listen: + host: 0.0.0.0 + port: 8004 +logit_bias: + parse_task: 0.1 + choose_model: 5 +tprompt: + parse_task: >- + #1 Task Planning Stage: The AI assistant can parse user input to several tasks: [{"task": task, "id": task_id, "dep": dependency_task_id, "args": {"text": text or -dep_id, "image": image_url or -dep_id, "audio": audio_url or -dep_id}}]. The special tag "-dep_id" refer to the one generated text/image/audio in the dependency task (Please consider whether the dependency task generates resources of this type.) and "dep_id" must be in "dep" list. The "dep" field denotes the ids of the previous prerequisite tasks which generate a new resource that the current task relies on. The "args" field must in ["text", "image", "audio"], nothing else. The task MUST be selected from the following options: "token-classification", "text2text-generation", "summarization", "translation", "question-answering", "conversational", "text-generation", "sentence-similarity", "tabular-classification", "object-detection", "image-classification", "image-to-image", "image-to-text", "text-to-image", "text-to-video", "visual-question-answering", "document-question-answering", "image-segmentation", "depth-estimation", "text-to-speech", "automatic-speech-recognition", "audio-to-audio", "audio-classification", "canny-control", "hed-control", "mlsd-control", "normal-control", "openpose-control", "canny-text-to-image", "depth-text-to-image", "hed-text-to-image", "mlsd-text-to-image", "normal-text-to-image", "openpose-text-to-image", "seg-text-to-image". There may be multiple tasks of the same type. Think step by step about all the tasks needed to resolve the user's request. Parse out as few tasks as possible while ensuring that the user request can be resolved. Pay attention to the dependencies and order among tasks. If the user input can't be parsed, you need to reply empty JSON []. + choose_model: >- + #2 Model Selection Stage: Given the user request and the parsed tasks, the AI assistant helps the user to select a suitable model from a list of models to process the user request. The assistant should focus more on the description of the model and find the model that has the most potential to solve requests and tasks. Also, prefer models with local inference endpoints for speed and stability. + response_results: >- + #4 Response Generation Stage: With the task execution logs, the AI assistant needs to describe the process and inference results. +demos_or_presteps: + parse_task: demos/demo_parse_task.json + choose_model: demos/demo_choose_model.json + response_results: demos/demo_response_results.json +prompt: + parse_task: The chat log [ {{context}} ] may contain the resources I mentioned. Now I input { {{input}} }. Pay attention to the input and output types of tasks and the dependencies between tasks. + choose_model: >- + Please choose the most suitable model from {{metas}} for the task {{task}}. The output must be in a strict JSON format: {"id": "id", "reason": "your detail reasons for the choice"}. + response_results: >- + Yes. Please first think carefully and directly answer my request based on the inference results. Some of the inferences may not always turn out to be correct and require you to make careful consideration in making decisions. Then please detail your workflow including the used models and inference results for my request in your friendly tone. Please filter out information that is not relevant to my request. Tell me the complete path or urls of files in inference results. If there is nothing in the results, please tell me you can't make it. } diff --git a/hugginggpt/server/get_token_ids.py b/hugginggpt/server/get_token_ids.py index 2e6c9e37..7cc87701 100644 --- a/hugginggpt/server/get_token_ids.py +++ b/hugginggpt/server/get_token_ids.py @@ -15,6 +15,10 @@ "curie": tiktoken.get_encoding("r50k_base"), "babbage": tiktoken.get_encoding("r50k_base"), "ada": tiktoken.get_encoding("r50k_base"), + "MiniMax-M2.7": tiktoken.get_encoding("cl100k_base"), + "MiniMax-M2.7-highspeed": tiktoken.get_encoding("cl100k_base"), + "MiniMax-M2.5": tiktoken.get_encoding("cl100k_base"), + "MiniMax-M2.5-highspeed": tiktoken.get_encoding("cl100k_base"), } max_length = { @@ -31,7 +35,11 @@ "davinci": 2049, "curie": 2049, "babbage": 2049, - "ada": 2049 + "ada": 2049, + "MiniMax-M2.7": 204800, + "MiniMax-M2.7-highspeed": 204800, + "MiniMax-M2.5": 204800, + "MiniMax-M2.5-highspeed": 204800, } def count_tokens(model_name, text):