feat: Added extra env variable to control DALLE additional parameters

README.md

@@ -68,6 +68,7 @@ The following variables cluster all deployments into the groups of deployments w
 |---|---|---|
 |DALLE3_DEPLOYMENTS|``|Comma-separated list of deployments that support DALL-E 3 API. Example: `dall-e-3,dalle3,dall-e`|
 |DALLE3_AZURE_API_VERSION|2024-02-01|The version API for requests to Azure DALL-E-3 API|
+|DALLE_EXTRA_PARAMETERS|`{}`|JSON formated list of extra parameters, example: '{"size": "1024x1024", "quality": "hd", "style": "vivid"}'|
 |GPT4_VISION_DEPLOYMENTS|``|Comma-separated list of deployments that support GPT-4V API. Example: `gpt-4-vision-preview,gpt-4-vision`|
 |GPT4_VISION_MAX_TOKENS|1024|Default value of `max_tokens` parameter for GPT-4V when it wasn't provided in the request|
 |MISTRAL_DEPLOYMENTS|``|Comma-separated list of deployments that support Mistral Large Azure API. Example: `mistral-large-azure,mistral-large`|

aidial_adapter_openai/dalle3.py

@@ -1,3 +1,5 @@
+import json
+import os
 from typing import Any, AsyncIterator, Optional
 
 import aiohttp
@@ -19,10 +21,24 @@
 async def generate_image(
     api_url: str, creds: OpenAICreds, user_prompt: str
 ) -> JSONResponse | Any:
+
+    extra_params = {}
+    try:
+        extra_params = json.loads(os.getenv("DALLE_EXTRA_PARAMETERS", "{}"))
+    except json.JSONDecodeError:
+        # Log or handle the case where the environment variable is not valid JSON
+        extra_params = {}
+
+    payload = {
+        "prompt": user_prompt,
+        "response_format": "b64_json",
+        **extra_params,  # Add extra parameters
+    }
+
     async with aiohttp.ClientSession() as session:
         async with session.post(
             api_url,
-            json={"prompt": user_prompt, "response_format": "b64_json"},
+            json=payload,
             headers=get_auth_headers(creds),
         ) as response:
             status_code = response.status

aidial_adapter_openai/exception_handlers.py

@@ -9,7 +9,6 @@
     ResponseWrapper,
     parse_adapter_exception,
 )
-from aidial_adapter_openai.utils.log_config import logger
 
 
 def to_adapter_exception(exc: Exception) -> AdapterException:
@@ -33,7 +32,7 @@ def to_adapter_exception(exc: Exception) -> AdapterException:
 
         return parse_adapter_exception(
             status_code=r.status_code,
-            headers=httpx_headers,
+            headers=dict(httpx_headers.items()),
             content=r.text,
         )
 
@@ -72,12 +71,6 @@ def to_adapter_exception(exc: Exception) -> AdapterException:
 
 
 def adapter_exception_handler(
-    request: FastAPIRequest, e: Exception
+    request: FastAPIRequest, exc: Exception
 ) -> FastAPIResponse:
-    adapter_exception = to_adapter_exception(e)
-
-    logger.error(
-        f"Caught exception: {type(e).__module__}.{type(e).__name__}. "
-        f"Converted to the adapter exception: {adapter_exception!r}"
-    )
-    return adapter_exception.to_fastapi_response()
+    return to_adapter_exception(exc).to_fastapi_response()

aidial_adapter_openai/gpt.py

@@ -8,18 +8,12 @@
 from aidial_adapter_openai.utils.auth import OpenAICreds
 from aidial_adapter_openai.utils.parsers import chat_completions_parser
 from aidial_adapter_openai.utils.reflection import call_with_extra_body
-from aidial_adapter_openai.utils.streaming import (
-    chunk_to_dict,
-    debug_print,
-    generate_stream,
-    map_stream,
-)
+from aidial_adapter_openai.utils.streaming import (chunk_to_dict, debug_print,
+                                                   generate_stream, map_stream)
 from aidial_adapter_openai.utils.tokenizer import PlainTextTokenizer
-from aidial_adapter_openai.utils.truncate_prompt import (
-    DiscardedMessages,
-    TruncatedTokens,
-    truncate_prompt,
-)
+from aidial_adapter_openai.utils.truncate_prompt import (DiscardedMessages,
+                                                         TruncatedTokens,
+                                                         truncate_prompt)
 
 
 def plain_text_truncate_prompt(

aidial_adapter_openai/gpt4_multi_modal/chat_completion.py

@@ -1,49 +1,31 @@
 import os
-from typing import (
-    Any,
-    AsyncIterator,
-    Callable,
-    Dict,
-    List,
-    Optional,
-    Tuple,
-    TypeVar,
-    cast,
-)
+from typing import (Any, AsyncIterator, Callable, Dict, List, Optional, Tuple,
+                    TypeVar, cast)
 
 import aiohttp
 from aidial_sdk.exceptions import HTTPException as DialException
 from aidial_sdk.exceptions import RequestValidationError
 from fastapi.responses import JSONResponse, Response
 
 from aidial_adapter_openai.dial_api.storage import FileStorage
-from aidial_adapter_openai.gpt4_multi_modal.gpt4_vision import (
-    convert_gpt4v_to_gpt4_chunk,
-)
+from aidial_adapter_openai.gpt4_multi_modal.gpt4_vision import \
+    convert_gpt4v_to_gpt4_chunk
 from aidial_adapter_openai.gpt4_multi_modal.transformation import (
-    SUPPORTED_FILE_EXTS,
-    ResourceProcessor,
-)
+    SUPPORTED_FILE_EXTS, ResourceProcessor)
 from aidial_adapter_openai.utils.auth import OpenAICreds, get_auth_headers
-from aidial_adapter_openai.utils.chat_completion_response import (
-    ChatCompletionBlock,
-)
+from aidial_adapter_openai.utils.chat_completion_response import \
+    ChatCompletionBlock
 from aidial_adapter_openai.utils.log_config import logger
 from aidial_adapter_openai.utils.multi_modal_message import MultiModalMessage
 from aidial_adapter_openai.utils.sse_stream import parse_openai_sse_stream
-from aidial_adapter_openai.utils.streaming import (
-    create_response_from_chunk,
-    create_stage_chunk,
-    generate_stream,
-    map_stream,
-    prepend_to_stream,
-)
+from aidial_adapter_openai.utils.streaming import (create_response_from_chunk,
+                                                   create_stage_chunk,
+                                                   generate_stream, map_stream,
+                                                   prepend_to_stream)
 from aidial_adapter_openai.utils.tokenizer import MultiModalTokenizer
-from aidial_adapter_openai.utils.truncate_prompt import (
-    DiscardedMessages,
-    TruncatedTokens,
-    truncate_prompt,
-)
+from aidial_adapter_openai.utils.truncate_prompt import (DiscardedMessages,
+                                                         TruncatedTokens,
+                                                         truncate_prompt)
 
 # The built-in default max_tokens is 16 tokens,
 # which is too small for most image-to-text use cases.

aidial_adapter_openai/utils/adapter_exception.py

@@ -1,5 +1,5 @@
 import json
-from typing import Any, MutableMapping
+from typing import Any, Dict
 
 from aidial_sdk.exceptions import HTTPException as DialException
 from fastapi.responses import Response as FastAPIResponse
@@ -8,14 +8,14 @@
 class ResponseWrapper(Exception):
     content: Any
     status_code: int
-    headers: MutableMapping[str, str] | None
+    headers: Dict[str, str] | None
 
     def __init__(
         self,
         *,
         content: Any,
         status_code: int,
-        headers: MutableMapping[str, str] | None,
+        headers: Dict[str, str] | None,
     ) -> None:
         super().__init__(str(content))
         self.content = content
@@ -51,7 +51,7 @@ def json_error(self) -> dict:
 
 
 def _parse_dial_exception(
-    *, status_code: int, headers: MutableMapping[str, str], content: Any
+    *, status_code: int, headers: Dict[str, str], content: Any
 ) -> DialException | None:
     if isinstance(content, str):
         try:
@@ -61,11 +61,6 @@ def _parse_dial_exception(
     else:
         obj = content
 
-    # The content length is invalidated as soon as
-    # the original content is lost
-    if "Content-Length" in headers:
-        del headers["Content-Length"]
-
     if (
         isinstance(obj, dict)
         and (error := obj.get("error"))
@@ -84,14 +79,14 @@ def _parse_dial_exception(
             param=param,
             code=code,
             display_message=display_message,
-            headers=dict(headers.items()),
+            headers=headers,
         )
 
     return None
 
 
 def parse_adapter_exception(
-    *, status_code: int, headers: MutableMapping[str, str], content: Any
+    *, status_code: int, headers: Dict[str, str], content: Any
 ) -> AdapterException:
     return _parse_dial_exception(
         status_code=status_code, headers=headers, content=content

aidial_adapter_openai/utils/sse_stream.py

@@ -60,11 +60,13 @@ async def to_openai_sse_stream(
         async for chunk in stream:
             yield format_chunk(chunk)
     except Exception as e:
-        adapter_exception = to_adapter_exception(e)
-
         logger.exception(
-            f"Caught exception while streaming: {type(e).__module__}.{type(e).__name__}. "
-            f"Converted to the adapter exception: {adapter_exception!r}"
+            f"caught exception while streaming: {type(e).__module__}.{type(e).__name__}"
+        )
+
+        adapter_exception = to_adapter_exception(e)
+        logger.error(
+            f"converted to the adapter exception: {adapter_exception!r}"
         )
 
         yield format_chunk(adapter_exception.json_error())

aidial_adapter_openai/utils/streaming.py

@@ -10,9 +10,7 @@
 from pydantic import BaseModel
 
 from aidial_adapter_openai.utils.chat_completion_response import (
-    ChatCompletionResponse,
-    ChatCompletionStreamingChunk,
-)
+    ChatCompletionResponse, ChatCompletionStreamingChunk)
 from aidial_adapter_openai.utils.log_config import logger
 from aidial_adapter_openai.utils.sse_stream import to_openai_sse_stream
 

aidial_adapter_openai/utils/tokenizer.py

@@ -9,9 +9,8 @@
 from aidial_sdk.exceptions import InternalServerError
 from tiktoken import Encoding, encoding_for_model
 
-from aidial_adapter_openai.utils.chat_completion_response import (
-    ChatCompletionResponse,
-)
+from aidial_adapter_openai.utils.chat_completion_response import \
+    ChatCompletionResponse
 from aidial_adapter_openai.utils.image_tokenizer import ImageTokenizer
 from aidial_adapter_openai.utils.multi_modal_message import MultiModalMessage
 from aidial_adapter_openai.utils.text import truncate_string