Add MiniMax external LLM API provider and model paths

api.py

@@ -1,8 +1,50 @@
 import os
 from pathlib import Path
 
+import requests
 from huggingface_hub import InferenceClient
 
+MINIMAX_MODELS = ("MiniMax-M3", "MiniMax-M2.7")
+
+MINIMAX_BASE_URLS = {
+    "global_en": "https://api.minimax.io/v1",
+    "cn_zh": "https://api.minimaxi.com/v1",
+}
+
+def minimax_api(prompt, history, verbose=True, return_dict=True, api_model="MiniMax-M3", region="global_en"):
+    """
+    Example:
+    --------
+    agent = Agent(toolchain = "responses, history = minimax_api(prompt, history)")
+    agent('Write a neurology ICU admission note')
+    """
+    history = [] if history is None else history
+    history = history + [{"role": "user", "content": prompt}]
+    base_url = MINIMAX_BASE_URLS.get(region, MINIMAX_BASE_URLS["global_en"])
+    response = requests.post(
+        f"{base_url}/chat/completions",
+        headers={
+            "Authorization": f"Bearer {os.environ.get('MINIMAX_API_KEY', '')}",
+            "Content-Type": "application/json",
+        },
+        json=dict(
+            model=api_model,
+            messages=history,
+            temperature=0.9,
+            top_p=0.95,
+            max_tokens=8192,
+            stream=False,
+        ),
+    )
+    response.raise_for_status()
+    result = response.json()["choices"][0]["message"]["content"].strip()
+    history = history + [{"role": "assistant", "content": result}]
+    if verbose:
+        print(f'### Prompt ###\n{prompt}\n### Output ###\n{result}')
+    if return_dict:
+        return {'responses':result, 'history':history}
+    return result
+
 def mistral_api(prompt, history, verbose=True, return_dict=True, api_model="mistralai/Mistral-Nemo-Instruct-2407"):
     """
     Example:

examples.py

@@ -73,6 +73,16 @@
 agent("The patient's prognosis?")
 agent.end()
 
+## LLM Backend Hotswap (MiniMax)
+agent = pv.Agent(toolchain = "responses, history = minimax_api(prompt, history)")
+agent('Write a neurology ICU admission note')
+agent.end()
+
+### Selecting a MiniMax model (MiniMax-M3 is the default; MiniMax-M2.7 is also available)
+agent = pv.Agent(toolchain = 'responses, history = minimax_api(prompt, history, api_model="MiniMax-M2.7")')
+agent('Write a neurology ICU admission note')
+agent.end()
+
 # Misc
 
 ## ICL

phi_3_vision_mlx.py

@@ -26,7 +26,7 @@
 from mlx.utils import tree_flatten, tree_unflatten
 from PIL import Image
 
-from api import bark_api, mistral_api
+from api import bark_api, minimax_api, mistral_api
 from gte import VDB, GteModel
 from phi import (LoRALinear, Phi3ForCausalLM, Phi3FProcessor, Phi3VForCausalLM,
                  Phi3VProcessor, Tic, TrainingCallback)