Azure LLM fix (#1227)

* azure embedding fix

* corrected embedding config

* fixed doc

Makefile

@@ -167,8 +167,8 @@ setup-config-prompts:
 		elif [ "$$llm_embedding_model" = "azureopenai" ]; then \
 			read -p "Enter the Azure endpoint URL (will overwrite LLM_BASE_URL): " llm_base_url; \
 				echo "LLM_BASE_URL=\"$$llm_base_url\"" >> $(CONFIG_FILE).tmp; \
-			read -p "Enter the Azure LLM Deployment Name: " llm_deployment_name; \
-				echo "LLM_DEPLOYMENT_NAME=\"$$llm_deployment_name\"" >> $(CONFIG_FILE).tmp; \
+			read -p "Enter the Azure LLM Embedding Deployment Name: " llm_embedding_deployment_name; \
+				echo "LLM_EMBEDDING_DEPLOYMENT_NAME=\"$$llm_embedding_deployment_name\"" >> $(CONFIG_FILE).tmp; \
 			read -p "Enter the Azure API Version: " llm_api_version; \
 				echo "LLM_API_VERSION=\"$$llm_api_version\"" >> $(CONFIG_FILE).tmp; \
 		fi

README.md

@@ -158,7 +158,7 @@ The following environment variables might be necessary for some LLMs:
 * `LLM_API_KEY`
 * `LLM_BASE_URL`
 * `LLM_EMBEDDING_MODEL`
-* `LLM_DEPLOYMENT_NAME`
+* `LLM_EMBEDDING_DEPLOYMENT_NAME`
 * `LLM_API_VERSION`
 
 We have a few guides for running OpenDevin with specific model providers:

agenthub/monologue_agent/utils/memory.py

@@ -32,7 +32,7 @@ elif embedding_strategy == 'azureopenai':
     from llama_index.embeddings.azure_openai import AzureOpenAIEmbedding
     embed_model = AzureOpenAIEmbedding(
         model='text-embedding-ada-002',
-        deployment_name=config.get('LLM_DEPLOYMENT_NAME', required=True),
+        deployment_name=config.get('LLM_EMBEDDING_DEPLOYMENT_NAME', required=True),
         api_key=config.get('LLM_API_KEY', required=True),
         azure_endpoint=config.get('LLM_BASE_URL', required=True),
         api_version=config.get('LLM_API_VERSION', required=True),

docs/README-zh.md

@@ -160,7 +160,7 @@ OpenDevin 可以与任何 LLM 后端配合使用。
 * `LLM_API_KEY`
 * `LLM_BASE_URL`
 * `LLM_EMBEDDING_MODEL`
-* `LLM_DEPLOYMENT_NAME`
+* `LLM_EMBEDDING_DEPLOYMENT_NAME`
 * `LLM_API_VERSION`
 
 **关于替代模型的说明：**

docs/guides/AzureLLMs.md

@@ -11,7 +11,7 @@ When running the OpenDevin Docker image, you'll need to set the following enviro
 LLM_BASE_URL="<azure-api-base-url>"          # e.g. "https://openai-gpt-4-test-v-1.openai.azure.com/"
 LLM_API_KEY="<azure-api-key>"
 LLM_MODEL="azure/<your-gpt-deployment-name>"
-AZURE_API_VERSION = "<api-version>"          # e.g. "2024-02-15-preview"
+LLM_API_VERSION = "<api-version>"          # e.g. "2024-02-15-preview"
 ```
 
 # 2. Embeddings
@@ -26,6 +26,6 @@ You need the correct deployment name for this model in your Azure account.
 When running OpenDevin in Docker, set the following environment variables using `-e`:
 ```
 LLM_EMBEDDING_MODEL="azureopenai"
-DEPLOYMENT_NAME = "<your-embedding-deployment-name>"        # e.g. "TextEmbedding...<etc>"
+LLM_EMBEDDING_DEPLOYMENT_NAME = "<your-embedding-deployment-name>"        # e.g. "TextEmbedding...<etc>"
 LLM_API_VERSION = "<api-version>"         # e.g. "2024-02-15-preview"
 ```

opendevin/config.py

@@ -18,7 +18,7 @@ DEFAULT_CONFIG: dict = {
     ConfigType.SANDBOX_CONTAINER_IMAGE: 'ghcr.io/opendevin/sandbox',
     ConfigType.RUN_AS_DEVIN: 'true',
     ConfigType.LLM_EMBEDDING_MODEL: 'local',
-    ConfigType.LLM_DEPLOYMENT_NAME: None,
+    ConfigType.LLM_EMBEDDING_DEPLOYMENT_NAME: None,
     ConfigType.LLM_API_VERSION: None,
     ConfigType.LLM_NUM_RETRIES: 1,
     ConfigType.LLM_COOLDOWN_TIME: 1,

opendevin/llm/llm.py

@@ -12,6 +12,7 @@ DEFAULT_BASE_URL = config.get('LLM_BASE_URL')
 DEFAULT_MODEL_NAME = config.get('LLM_MODEL')
 DEFAULT_LLM_NUM_RETRIES = config.get('LLM_NUM_RETRIES')
 DEFAULT_LLM_COOLDOWN_TIME = config.get('LLM_COOLDOWN_TIME')
+DEFAULT_API_VERSION = config.get('LLM_API_VERSION')
 
 
 class LLM:
@@ -21,14 +22,16 @@ class LLM:
                  base_url=DEFAULT_BASE_URL,
                  num_retries=DEFAULT_LLM_NUM_RETRIES,
                  cooldown_time=DEFAULT_LLM_COOLDOWN_TIME,
+                 api_version=DEFAULT_API_VERSION,
                  ):
         opendevin_logger.info(f'Initializing LLM with model: {model}')
-        self.model_name = model if model else DEFAULT_MODEL_NAME
-        self.api_key = api_key if api_key else DEFAULT_API_KEY
-        self.base_url = base_url if base_url else DEFAULT_BASE_URL
+        self.model_name = model
+        self.api_key = api_key
+        self.base_url = base_url
+        self.api_version = api_version
 
         self._completion = partial(
-            litellm_completion, model=self.model_name, api_key=self.api_key, base_url=self.base_url)
+            litellm_completion, model=self.model_name, api_key=self.api_key, base_url=self.base_url, api_version=self.api_version)
 
         completion_unwrapped = self._completion
 
@@ -64,4 +67,8 @@ class LLM:
         return self._completion
 
     def __str__(self):
-        return f'LLM(model={self.model_name}, base_url={self.base_url})'
+        if self.api_version:
+            return f'LLM(model={self.model_name}, api_version={self.api_version}, base_url={self.base_url})'
+        elif self.base_url:
+            return f'LLM(model={self.model_name}, base_url={self.base_url})'
+        return f'LLM(model={self.model_name})'

opendevin/schema/config.py

@@ -11,7 +11,7 @@ class ConfigType(str, Enum):
     SANDBOX_CONTAINER_IMAGE = 'SANDBOX_CONTAINER_IMAGE'
     RUN_AS_DEVIN = 'RUN_AS_DEVIN'
     LLM_EMBEDDING_MODEL = 'LLM_EMBEDDING_MODEL'
-    LLM_DEPLOYMENT_NAME = 'LLM_DEPLOYMENT_NAME'
+    LLM_EMBEDDING_DEPLOYMENT_NAME = 'LLM_EMBEDDING_DEPLOYMENT_NAME'
     LLM_API_VERSION = 'LLM_API_VERSION'
     LLM_NUM_RETRIES = 'LLM_NUM_RETRIES'
     LLM_COOLDOWN_TIME = 'LLM_COOLDOWN_TIME'