added mistral

app.py

@@ -11,6 +11,7 @@ from llama_index.llms.openai import OpenAI
 
 from llama_index.embeddings.openai import OpenAIEmbedding
 from llama_index.embeddings.huggingface import HuggingFaceEmbedding, HuggingFaceInferenceAPIEmbedding
+from llama_index.embeddings.mistralai import MistralAIEmbedding
 
 from llama_index.core import (
     VectorStoreIndex,
@@ -42,7 +43,7 @@ def main():
         elif provider == 'huggingface':
             llm_list = []
         elif provider == 'mistralai':
-            llm_list =[]
+            llm_list = ["mistral-small-latest", "mistral-medium-latest", "mistral-large-latest", "open-mistral-nemo"]
         elif provider == 'openai':
             llm_list = ['gpt-3.5-turbo', 'gpt-4', 'gpt-4-turbo', 'gpt-4o', 'gpt-4o-mini']
         else:
@@ -91,25 +92,14 @@ def main():
         # Global tokenization needs to be consistent with LLM for token counting
         # https://docs.llamaindex.ai/en/stable/module_guides/models/llms/
         if llm_key is not None:
-            if provider == 'openai':
-                os.environ["OPENAI_API_KEY"] = str(llm_key)
-                Settings.llm = OpenAI(
-                    model=llm_name, 
-                    temperature=temperature,
-                    # max_tokens=MAX_OUTPUT_TOKENS
-                )
-                Settings.tokenizer = tiktoken.encoding_for_model(llm_name).encode
-                Settings.num_output = MAX_OUTPUT_TOKENS
-                Settings.embed_model = OpenAIEmbedding()
-                Settings.context_window = 4096 # max possible
-            elif provider == 'huggingface':
+            if provider == 'huggingface':
                 if llm_name is not None and embed_name is not None:
                     os.environ['HFTOKEN'] = str(llm_key)
                     Settings.llm = HuggingFaceInferenceAPI(
                         model_name=llm_name, 
                         token=os.environ.get("HFTOKEN"),
                         temperature=temperature,
-                        # max_tokens=MAX_OUTPUT_TOKENS
+                        max_tokens=MAX_OUTPUT_TOKENS
                     )
                     Settings.tokenizer = AutoTokenizer.from_pretrained(
                         llm_name,
@@ -120,6 +110,33 @@ def main():
                         model_name=embed_name
                     )
                     Settings.context_window = 4096 
+            elif provider == 'mistralai':
+                os.environ['MISTRAL_API_KEY'] = str(llm_key)
+                Settings.llm = MistralAI(
+                    model=llm_name, 
+                    temperature=temperature,
+                    max_tokens=MAX_OUTPUT_TOKENS,
+                    random_seed=42,
+                    safe_mode=True
+                )
+                # Settings.tokenizer = tiktoken.encoding_for_model(llm_name).encode
+                Settings.num_output = MAX_OUTPUT_TOKENS
+                Settings.embed_model = MistralAIEmbedding(
+                    model_name="mistral-embed", 
+                    api_key=os.environ.get("MISTRAL_API_KEY")
+                )
+                Settings.context_window = 4096 # max possible
+            elif provider == 'openai':
+                os.environ["OPENAI_API_KEY"] = str(llm_key)
+                Settings.llm = OpenAI(
+                    model=llm_name, 
+                    temperature=temperature,
+                    max_tokens=MAX_OUTPUT_TOKENS
+                )
+                Settings.tokenizer = tiktoken.encoding_for_model(llm_name).encode
+                Settings.num_output = MAX_OUTPUT_TOKENS
+                Settings.embed_model = OpenAIEmbedding()
+                Settings.context_window = 4096 # max possible
             else:
                 raise NotImplementedError(f"{provider} is not supported yet")
 

requirements.txt

@@ -10,4 +10,5 @@ tiktoken
 llama-parse
 llama-index-callbacks-langfuse
 llama-index-embeddings-huggingface
-llama-index-embeddings-openai
+llama-index-embeddings-openai
+llama-index-embeddings-mistralai