API: Modify tool calling for wider compat

When revisiting tool calls, the formats have more or less become standard.
For greater compatibility with templates, primarily use the message.tools
parameter and remove the extra custom metadata that is no longer required.

However, unlike other backends, tabbyAPI still uses template metadata
to declare what the tool start string is. This allows for template-level
customization along with giving more power to the user while the server
exists to consume rather than work on a case-by-case basis.

Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>

common/templating.py

@@ -4,6 +4,8 @@ import traceback
 import aiofiles
 import json
 import pathlib
+from dataclasses import dataclass, field
+from datetime import datetime
 from importlib.metadata import version as package_version
 from typing import List, Optional
 from jinja2 import Template, TemplateError
@@ -11,7 +13,6 @@ from jinja2.ext import loopcontrols
 from jinja2.sandbox import ImmutableSandboxedEnvironment
 from loguru import logger
 from packaging import version
-from datetime import datetime
 
 
 from common.utils import unwrap
@@ -23,11 +24,12 @@ class TemplateLoadError(Exception):
     pass
 
 
+@dataclass
 class TemplateMetadata:
     """Represents the parsed metadata from a template."""
 
-    stop_strings: List[str] = []
+    stop_strings: List[str] = field(default_factory=list)
-    tool_starts: List[str] = []
+    tool_start: Optional[str] = None
 
 
 class PromptTemplate:
@@ -72,11 +74,7 @@ class PromptTemplate:
 
         if hasattr(template_module, "tool_start"):
             if isinstance(template_module.tool_start, str):
-                template_metadata.tool_starts.append(template_module.tool_start)
+                template_metadata.tool_start = template_module.tool_start
-
-        if hasattr(template_module, "tool_start_token"):
-            if isinstance(template_module.tool_start_token, int):
-                template_metadata.tool_starts.append(template_module.tool_start_token)
 
         self.metadata = template_metadata
         return template_metadata

endpoints/OAI/types/chat_completion.py

@@ -5,7 +5,7 @@ from typing import Literal, Union, List, Optional, Dict
 from uuid import uuid4
 
 from endpoints.OAI.types.common import UsageStats, CommonCompletionRequest
-from endpoints.OAI.types.tools import ToolSpec, ToolCall, tool_call_schema
+from endpoints.OAI.types.tools import ToolSpec, ToolCall
 
 
 class ChatCompletionLogprob(BaseModel):
@@ -73,15 +73,6 @@ class ChatCompletionRequest(CommonCompletionRequest):
     tools: Optional[List[ToolSpec]] = None
     functions: Optional[List[Dict]] = None
 
-    # Typically collected from Chat Template.
-    # Don't include this in the OpenAPI docs
-    # TODO: Use these custom parameters
-    tool_call_start: SkipJsonSchema[Optional[List[Union[str, int]]]] = Field(
-        default_factory=list
-    )
-    tool_call_end: SkipJsonSchema[Optional[str]] = None
-    tool_call_schema: SkipJsonSchema[Optional[dict]] = tool_call_schema
-
     # Chat completions requests do not have a BOS token preference. Backend
     # respects the tokenization config for the individual model.
     add_bos_token: Optional[bool] = None

endpoints/OAI/types/tools.py

@@ -1,30 +1,6 @@
 from pydantic import BaseModel
 from typing import Dict, Literal
 
-tool_call_schema = {
-    "$schema": "http://json-schema.org/draft-07/schema#",
-    "type": "array",
-    "items": {
-        "type": "object",
-        "properties": {
-            "id": {"type": "string"},
-            "function": {
-                "type": "object",
-                "properties": {
-                    "name": {"type": "string"},
-                    "arguments": {
-                        # Converted to OAI's string in post process
-                        "type": "object"
-                    },
-                },
-                "required": ["name", "arguments"],
-            },
-            "type": {"type": "string", "enum": ["function"]},
-        },
-        "required": ["id", "function", "type"],
-    },
-}
-
 
 class Function(BaseModel):
     """Represents a description of a tool function."""

endpoints/OAI/utils/chat_completion.py

@@ -1,7 +1,6 @@
 """Chat completion utilities for OAI server."""
 
 import asyncio
-import json
 import pathlib
 from asyncio import CancelledError
 from typing import List, Optional
@@ -30,7 +29,7 @@ from endpoints.OAI.types.chat_completion import (
 )
 from endpoints.OAI.types.common import UsageStats
 from endpoints.OAI.utils.completion import _parse_gen_request_id, _stream_collector
-from endpoints.OAI.utils.tools import ToolCallProcessor
+from endpoints.OAI.utils.tools import ToolCallProcessor, TOOL_CALL_SCHEMA
 
 
 def _create_response(
@@ -209,12 +208,9 @@ async def _append_template_metadata(data: ChatCompletionRequest, template_vars:
     else:
         data.stop.extend(template_metadata.stop_strings)
 
-    # Tool call start strings
+    # if a tool start is present, append it to stopping strings
-    if template_metadata.tool_starts:
+    if template_metadata.tool_start:
-        data.tool_call_start.extend(template_metadata.tool_starts)
+        data.stop.append(template_metadata.tool_start)
-
-        # Append to stop strings to halt for a tool call generation
-        data.stop.extend(template_metadata.tool_starts)
 
 
 async def format_messages_with_template(
@@ -255,9 +251,7 @@ async def format_messages_with_template(
     return prompt, mm_embeddings, template_vars
 
 
-async def apply_chat_template(
+async def apply_chat_template(data: ChatCompletionRequest):
-    data: ChatCompletionRequest, tool_precursor: Optional[str] = None
-):
     """
     Compile the prompt and get any additional stop strings from the template.
     Template stop strings can be overriden by sampler overrides if force is true.
@@ -271,10 +265,7 @@ async def apply_chat_template(
             {
                 "add_generation_prompt": data.add_generation_prompt,
                 "tools": tools,
-                "tools_json": json.dumps(tools, indent=2),
                 "functions": data.functions,
-                "functions_json": json.dumps(data.functions, indent=2),
-                "tool_precursor": tool_precursor,
             }
         )
 
@@ -332,6 +323,7 @@ async def stream_generate_chat_completion(
     abort_event = asyncio.Event()
     gen_queue = asyncio.Queue()
     gen_tasks: List[asyncio.Task] = []
+    tool_start = model.container.prompt_template.metadata.tool_start
     disconnect_task = asyncio.create_task(request_disconnect_loop(request))
 
     try:
@@ -355,7 +347,7 @@ async def stream_generate_chat_completion(
 
             gen_tasks.append(gen_task)
 
-        # We need to keep track of the text generated so we can resume the tool calls
+        # Text accumulation for tool calls
         current_generation_text = ""
 
         # Consumer loop
@@ -367,19 +359,21 @@ async def stream_generate_chat_completion(
                 )
 
             generation = await gen_queue.get()
-            # lets only append the text if we need it for tool calls later
-            if data.tool_call_start and "text" in generation:
-                current_generation_text += generation["text"]
 
-            # check if we are running a tool model, and that we are at stop
+            # Handle options if a tool model is present
-            if data.tool_call_start and "stop_str" in generation:
+            if tool_start:
-                generations = await generate_tool_calls(
+                if "stop_str" in generation:
-                    data,
+                    generations = await generate_tool_calls(
-                    [generation],
+                        data,
-                    request,
+                        [generation],
-                    current_generations=current_generation_text,
+                        request,
-                )
+                        current_generation_text=current_generation_text,
-                generation = generations[0]  # We only have one generation in this case
+                    )
+
+                    # Only one generation present in this case
+                    generation = generations[0]
+                elif "text" in generation:
+                    current_generation_text += generation["text"]
 
             # Stream collector will push an exception to the queue if it fails
             if isinstance(generation, Exception):
@@ -428,6 +422,7 @@ async def generate_chat_completion(
     model_path: pathlib.Path,
 ):
     gen_tasks: List[asyncio.Task] = []
+    tool_start = model.container.prompt_template.metadata.tool_start
 
     try:
         logger.info(f"Received chat completion request {request.state.id}")
@@ -448,8 +443,8 @@ async def generate_chat_completion(
 
         generations = await asyncio.gather(*gen_tasks)
 
-        # Let's not waste our time if we arn't running a tool model
+        # Check all the generations and see if a tool call is required
-        if data.tool_call_start:
+        if tool_start:
             generations = await generate_tool_calls(data, generations, request)
 
         response = _create_response(request.state.id, generations, model_path.name)
@@ -472,51 +467,52 @@ async def generate_tool_calls(
     data: ChatCompletionRequest,
     generations: List[str],
     request: Request,
-    current_generations: str = None,
+    current_generation_text: str = None,
 ):
     gen_tasks: List[asyncio.Task] = []
+    tool_start = model.container.prompt_template.metadata.tool_start
+
+    # Tracks which generations asked for a tool call
     tool_idx: List[int] = []
 
     # Copy to make sure the parent JSON schema doesn't get modified
-    # FIXME: May not be necessary depending on how the codebase evolves
     tool_data = data.model_copy(deep=True)
-    tool_data.json_schema = tool_data.tool_call_schema
+    tool_data.json_schema = TOOL_CALL_SCHEMA
 
     for idx, gen in enumerate(generations):
-        if gen["stop_str"] in tool_data.tool_call_start:
+        if gen["stop_str"] != tool_start:
-            logger.info(
+            continue
-                f"Detected tool call in chat completion request {request.state.id}"
-            )
 
-            if "text" in gen:
+        logger.info(f"Detected tool call in chat completion request {request.state.id}")
-                # non streaming, all generations will have the text they generated
-                pre_tool_prompt, embeddings = await apply_chat_template(
-                    data, gen["text"]
-                )
-            elif current_generations is not None:
-                # streaming, we wont have text in the generation,
-                # we'll have to use the current_generations
-                pre_tool_prompt, embeddings = await apply_chat_template(
-                    data, current_generations
-                )
 
-            request_id = _parse_gen_request_id(data.n, request.state.id, idx)
+        # Append the existing generation as part of the response prefix
+        precursor_text = current_generation_text or gen.get("text")
+        if precursor_text:
+            tool_data.response_prefix = precursor_text
 
-            gen_tasks.append(
+        pre_tool_prompt, embeddings = await apply_chat_template(tool_data)
-                asyncio.create_task(
+
-                    model.container.generate(
+        gen_request_id = _parse_gen_request_id(data.n, request.state.id, idx)
-                        request_id,
+        tool_request_id = f"{gen_request_id}-tool"
-                        pre_tool_prompt,
+
-                        tool_data,
+        gen_tasks.append(
-                        mm_embeddings=embeddings,
+            asyncio.create_task(
-                    )
+                model.container.generate(
+                    tool_request_id,
+                    pre_tool_prompt,
+                    tool_data,
+                    mm_embeddings=embeddings,
                 )
             )
-            tool_idx.append(idx)
+        )
 
-    tool_calls = await asyncio.gather(*gen_tasks)
+        tool_idx.append(idx)
-    for outer_idx in range(0, len(tool_idx)):
+
-        gen_idx = tool_idx[outer_idx]
+    if len(tool_idx) > 0:
-        generations[gen_idx]["tool_calls"] = tool_calls[outer_idx]["text"]
+        tool_calls = await asyncio.gather(*gen_tasks)
+
+        # Map tool calls to their appropriate generation
+        for gen_idx, tool_call in zip(tool_idx, tool_calls, strict=True):
+            generations[gen_idx]["tool_calls"] = tool_call["text"]
 
     return generations

endpoints/OAI/utils/tools.py

@@ -5,6 +5,31 @@ from typing import List
 from endpoints.OAI.types.tools import ToolCall
 
 
+TOOL_CALL_SCHEMA = {
+    "$schema": "http://json-schema.org/draft-07/schema#",
+    "type": "array",
+    "items": {
+        "type": "object",
+        "properties": {
+            "id": {"type": "string"},
+            "function": {
+                "type": "object",
+                "properties": {
+                    "name": {"type": "string"},
+                    "arguments": {
+                        # Converted to OAI's string in post process
+                        "type": "object"
+                    },
+                },
+                "required": ["name", "arguments"],
+            },
+            "type": {"type": "string", "enum": ["function"]},
+        },
+        "required": ["id", "function", "type"],
+    },
+}
+
+
 class ToolCallProcessor:
     @staticmethod
     def from_json(tool_calls_str: str) -> List[ToolCall]: