Update tools/gemma_tool_parser.py
Browse files- tools/gemma_tool_parser.py +285 -291
tools/gemma_tool_parser.py
CHANGED
|
@@ -1,291 +1,285 @@
|
|
| 1 |
-
# SPDX-License-Identifier: Apache-2.0
|
| 2 |
-
|
| 3 |
-
import json
|
| 4 |
-
import re
|
| 5 |
-
from collections.abc import Sequence
|
| 6 |
-
from json import JSONDecoder
|
| 7 |
-
from typing import Union
|
| 8 |
-
|
| 9 |
-
import partial_json_parser
|
| 10 |
-
from partial_json_parser.core.options import Allow
|
| 11 |
-
from transformers import PreTrainedTokenizerBase
|
| 12 |
-
|
| 13 |
-
from vllm.entrypoints.openai.protocol import (
|
| 14 |
-
ChatCompletionRequest,
|
| 15 |
-
DeltaFunctionCall,
|
| 16 |
-
DeltaMessage,
|
| 17 |
-
DeltaToolCall,
|
| 18 |
-
ExtractedToolCallInformation,
|
| 19 |
-
FunctionCall,
|
| 20 |
-
ToolCall,
|
| 21 |
-
)
|
| 22 |
-
from vllm.entrypoints.openai.tool_parsers.abstract_tool_parser import (
|
| 23 |
-
ToolParser,
|
| 24 |
-
ToolParserManager,
|
| 25 |
-
)
|
| 26 |
-
from vllm.entrypoints.openai.tool_parsers.utils import (
|
| 27 |
-
find_common_prefix,
|
| 28 |
-
is_complete_json,
|
| 29 |
-
partial_json_loads,
|
| 30 |
-
)
|
| 31 |
-
from vllm.logger import init_logger
|
| 32 |
-
from vllm.utils import random_uuid
|
| 33 |
-
|
| 34 |
-
logger = init_logger(__name__)
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
@ToolParserManager.register_module("
|
| 38 |
-
class GemmaJsonToolParser(ToolParser):
|
| 39 |
-
"""
|
| 40 |
-
Tool call parser for Gemma 3 models intended for use with the
|
| 41 |
-
appropriate Gemma chat template.
|
| 42 |
-
|
| 43 |
-
Used when --enable-auto-tool-choice --tool-call-parser gemma_json
|
| 44 |
-
are all set
|
| 45 |
-
"""
|
| 46 |
-
|
| 47 |
-
def __init__(self, tokenizer: PreTrainedTokenizerBase):
|
| 48 |
-
super().__init__(tokenizer)
|
| 49 |
-
|
| 50 |
-
# initialize properties used for state when parsing tool calls in
|
| 51 |
-
# streaming mode
|
| 52 |
-
self.prev_tool_call_arr: list[dict] = []
|
| 53 |
-
self.current_tool_id: int = -1
|
| 54 |
-
self.current_tool_name_sent: bool = False
|
| 55 |
-
self.streamed_args_for_tool: list[str] = []
|
| 56 |
-
|
| 57 |
-
# Gemma specific tokens
|
| 58 |
-
self.bos_token = "<bos>"
|
| 59 |
-
self.model_token = "<start_of_turn>model"
|
| 60 |
-
self.user_token = "<start_of_turn>user"
|
| 61 |
-
self.end_turn_token = "<end_of_turn>"
|
| 62 |
-
|
| 63 |
-
# For JSON detection
|
| 64 |
-
self.tool_call_regex = re.compile(r"\[{.*?}\]", re.DOTALL)
|
| 65 |
-
|
| 66 |
-
def extract_tool_calls(
|
| 67 |
-
self, model_output: str, request: ChatCompletionRequest
|
| 68 |
-
) -> ExtractedToolCallInformation:
|
| 69 |
-
"""
|
| 70 |
-
Extract the tool calls from a complete model response.
|
| 71 |
-
"""
|
| 72 |
-
# case -- if the response doesn't contain JSON, return a text response
|
| 73 |
-
if not model_output.startswith("{"):
|
| 74 |
-
return ExtractedToolCallInformation(
|
| 75 |
-
tools_called=False, tool_calls=[], content=model_output
|
| 76 |
-
)
|
| 77 |
-
|
| 78 |
-
try:
|
| 79 |
-
# load the JSON, and then use it to build the Function and
|
| 80 |
-
# Tool Call
|
| 81 |
-
dec = JSONDecoder()
|
| 82 |
-
function_call_arr = []
|
| 83 |
-
|
| 84 |
-
start_idx = 0
|
| 85 |
-
while start_idx < len(model_output):
|
| 86 |
-
try:
|
| 87 |
-
(obj, end_idx) = dec.raw_decode(model_output[start_idx:])
|
| 88 |
-
start_idx += end_idx
|
| 89 |
-
# Skip any separators like semicolons or commas
|
| 90 |
-
while start_idx < len(model_output) and model_output[start_idx] in [
|
| 91 |
-
";",
|
| 92 |
-
",",
|
| 93 |
-
" ",
|
| 94 |
-
]:
|
| 95 |
-
start_idx += 1
|
| 96 |
-
function_call_arr.append(obj)
|
| 97 |
-
except json.JSONDecodeError:
|
| 98 |
-
break
|
| 99 |
-
|
| 100 |
-
tool_calls: list[ToolCall] = [
|
| 101 |
-
ToolCall(
|
| 102 |
-
type="function",
|
| 103 |
-
function=FunctionCall(
|
| 104 |
-
name=raw_function_call["name"],
|
| 105 |
-
# function call args are JSON but as a string
|
| 106 |
-
arguments=json.dumps(
|
| 107 |
-
raw_function_call["arguments"]
|
| 108 |
-
if "arguments" in raw_function_call
|
| 109 |
-
else raw_function_call["parameters"]
|
| 110 |
-
),
|
| 111 |
-
),
|
| 112 |
-
)
|
| 113 |
-
for raw_function_call in function_call_arr
|
| 114 |
-
]
|
| 115 |
-
|
| 116 |
-
return ExtractedToolCallInformation(
|
| 117 |
-
tools_called=True, tool_calls=tool_calls, content=None
|
| 118 |
-
)
|
| 119 |
-
|
| 120 |
-
except Exception:
|
| 121 |
-
logger.exception("Error in extracting tool call from response.")
|
| 122 |
-
# return information to just treat the tool call as regular JSON
|
| 123 |
-
return ExtractedToolCallInformation(
|
| 124 |
-
tools_called=False, tool_calls=[], content=model_output
|
| 125 |
-
)
|
| 126 |
-
|
| 127 |
-
def extract_tool_calls_streaming(
|
| 128 |
-
self,
|
| 129 |
-
previous_text: str,
|
| 130 |
-
current_text: str,
|
| 131 |
-
delta_text: str,
|
| 132 |
-
previous_token_ids: Sequence[int],
|
| 133 |
-
current_token_ids: Sequence[int],
|
| 134 |
-
delta_token_ids: Sequence[int],
|
| 135 |
-
request: ChatCompletionRequest,
|
| 136 |
-
) -> Union[DeltaMessage, None]:
|
| 137 |
-
|
| 138 |
-
# Skip if not JSON format
|
| 139 |
-
if not current_text.startswith("{"):
|
| 140 |
-
return DeltaMessage(content=delta_text)
|
| 141 |
-
|
| 142 |
-
# bit mask flags for partial JSON parsing
|
| 143 |
-
flags = Allow.ALL if self.current_tool_name_sent else Allow.ALL & ~Allow.STR
|
| 144 |
-
try:
|
| 145 |
-
tool_call_arr = []
|
| 146 |
-
is_complete = []
|
| 147 |
-
try:
|
| 148 |
-
start_idx = 0
|
| 149 |
-
while start_idx < len(current_text):
|
| 150 |
-
(obj, end_idx) = partial_json_loads(current_text[start_idx:], flags)
|
| 151 |
-
is_complete.append(
|
| 152 |
-
is_complete_json(current_text[start_idx : start_idx + end_idx])
|
| 153 |
-
)
|
| 154 |
-
start_idx += end_idx
|
| 155 |
-
# Skip any separators like semicolons or commas
|
| 156 |
-
while start_idx < len(current_text) and current_text[start_idx] in [
|
| 157 |
-
";",
|
| 158 |
-
",",
|
| 159 |
-
" ",
|
| 160 |
-
]:
|
| 161 |
-
start_idx += 1
|
| 162 |
-
|
| 163 |
-
# Handle parameters field as arguments if needed
|
| 164 |
-
if "parameters" in obj:
|
| 165 |
-
assert (
|
| 166 |
-
"arguments" not in obj
|
| 167 |
-
), "model generated both parameters and arguments"
|
| 168 |
-
obj["arguments"] = obj["parameters"]
|
| 169 |
-
tool_call_arr.append(obj)
|
| 170 |
-
except partial_json_parser.core.exceptions.MalformedJSON:
|
| 171 |
-
logger.debug("not enough tokens to parse into JSON yet")
|
| 172 |
-
return None
|
| 173 |
-
|
| 174 |
-
# select as the current tool call the one we're on the state at
|
| 175 |
-
current_tool_call: dict = (
|
| 176 |
-
tool_call_arr[self.current_tool_id] if len(tool_call_arr) > 0 else {}
|
| 177 |
-
)
|
| 178 |
-
|
| 179 |
-
# case -- if no tokens have been streamed for the tool, e.g.
|
| 180 |
-
# only the array brackets, stream nothing
|
| 181 |
-
if len(tool_call_arr) == 0:
|
| 182 |
-
return None
|
| 183 |
-
|
| 184 |
-
# case: we are starting a new tool in the array
|
| 185 |
-
# -> array has > 0 length AND length has moved past cursor
|
| 186 |
-
elif (
|
| 187 |
-
len(tool_call_arr) > 0 and len(tool_call_arr) > self.current_tool_id + 1
|
| 188 |
-
):
|
| 189 |
-
|
| 190 |
-
|
| 191 |
-
|
| 192 |
-
|
| 193 |
-
|
| 194 |
-
|
| 195 |
-
|
| 196 |
-
|
| 197 |
-
|
| 198 |
-
|
| 199 |
-
|
| 200 |
-
|
| 201 |
-
|
| 202 |
-
|
| 203 |
-
|
| 204 |
-
|
| 205 |
-
|
| 206 |
-
|
| 207 |
-
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
| 213 |
-
|
| 214 |
-
|
| 215 |
-
|
| 216 |
-
|
| 217 |
-
|
| 218 |
-
|
| 219 |
-
|
| 220 |
-
|
| 221 |
-
|
| 222 |
-
|
| 223 |
-
|
| 224 |
-
|
| 225 |
-
|
| 226 |
-
|
| 227 |
-
|
| 228 |
-
|
| 229 |
-
|
| 230 |
-
|
| 231 |
-
|
| 232 |
-
|
| 233 |
-
|
| 234 |
-
|
| 235 |
-
|
| 236 |
-
|
| 237 |
-
|
| 238 |
-
|
| 239 |
-
|
| 240 |
-
|
| 241 |
-
|
| 242 |
-
|
| 243 |
-
|
| 244 |
-
|
| 245 |
-
|
| 246 |
-
|
| 247 |
-
|
| 248 |
-
|
| 249 |
-
|
| 250 |
-
|
| 251 |
-
|
| 252 |
-
|
| 253 |
-
|
| 254 |
-
|
| 255 |
-
|
| 256 |
-
|
| 257 |
-
|
| 258 |
-
|
| 259 |
-
|
| 260 |
-
|
| 261 |
-
|
| 262 |
-
|
| 263 |
-
|
| 264 |
-
|
| 265 |
-
|
| 266 |
-
|
| 267 |
-
|
| 268 |
-
|
| 269 |
-
|
| 270 |
-
|
| 271 |
-
|
| 272 |
-
|
| 273 |
-
|
| 274 |
-
|
| 275 |
-
|
| 276 |
-
|
| 277 |
-
|
| 278 |
-
|
| 279 |
-
|
| 280 |
-
|
| 281 |
-
|
| 282 |
-
|
| 283 |
-
|
| 284 |
-
|
| 285 |
-
|
| 286 |
-
except Exception:
|
| 287 |
-
logger.exception("Error trying to handle streaming tool call.")
|
| 288 |
-
logger.debug(
|
| 289 |
-
"Skipping chunk as a result of tool streaming extraction error"
|
| 290 |
-
)
|
| 291 |
-
return None
|
|
|
|
| 1 |
+
# SPDX-License-Identifier: Apache-2.0
|
| 2 |
+
|
| 3 |
+
import json
|
| 4 |
+
import re
|
| 5 |
+
from collections.abc import Sequence
|
| 6 |
+
from json import JSONDecoder
|
| 7 |
+
from typing import Union
|
| 8 |
+
|
| 9 |
+
import partial_json_parser
|
| 10 |
+
from partial_json_parser.core.options import Allow
|
| 11 |
+
from transformers import PreTrainedTokenizerBase
|
| 12 |
+
|
| 13 |
+
from vllm.entrypoints.openai.protocol import (
|
| 14 |
+
ChatCompletionRequest,
|
| 15 |
+
DeltaFunctionCall,
|
| 16 |
+
DeltaMessage,
|
| 17 |
+
DeltaToolCall,
|
| 18 |
+
ExtractedToolCallInformation,
|
| 19 |
+
FunctionCall,
|
| 20 |
+
ToolCall,
|
| 21 |
+
)
|
| 22 |
+
from vllm.entrypoints.openai.tool_parsers.abstract_tool_parser import (
|
| 23 |
+
ToolParser,
|
| 24 |
+
ToolParserManager,
|
| 25 |
+
)
|
| 26 |
+
from vllm.entrypoints.openai.tool_parsers.utils import (
|
| 27 |
+
find_common_prefix,
|
| 28 |
+
is_complete_json,
|
| 29 |
+
partial_json_loads,
|
| 30 |
+
)
|
| 31 |
+
from vllm.logger import init_logger
|
| 32 |
+
from vllm.utils import random_uuid
|
| 33 |
+
|
| 34 |
+
logger = init_logger(__name__)
|
| 35 |
+
|
| 36 |
+
|
| 37 |
+
@ToolParserManager.register_module("gemma")
|
| 38 |
+
class GemmaJsonToolParser(ToolParser):
|
| 39 |
+
"""
|
| 40 |
+
Tool call parser for Gemma 3 models intended for use with the
|
| 41 |
+
appropriate Gemma chat template.
|
| 42 |
+
|
| 43 |
+
Used when --enable-auto-tool-choice --tool-call-parser gemma_json
|
| 44 |
+
are all set
|
| 45 |
+
"""
|
| 46 |
+
|
| 47 |
+
def __init__(self, tokenizer: PreTrainedTokenizerBase):
|
| 48 |
+
super().__init__(tokenizer)
|
| 49 |
+
|
| 50 |
+
# initialize properties used for state when parsing tool calls in
|
| 51 |
+
# streaming mode
|
| 52 |
+
self.prev_tool_call_arr: list[dict] = []
|
| 53 |
+
self.current_tool_id: int = -1
|
| 54 |
+
self.current_tool_name_sent: bool = False
|
| 55 |
+
self.streamed_args_for_tool: list[str] = []
|
| 56 |
+
|
| 57 |
+
# Gemma specific tokens
|
| 58 |
+
self.bos_token = "<bos>"
|
| 59 |
+
self.model_token = "<start_of_turn>model"
|
| 60 |
+
self.user_token = "<start_of_turn>user"
|
| 61 |
+
self.end_turn_token = "<end_of_turn>"
|
| 62 |
+
|
| 63 |
+
# For JSON detection
|
| 64 |
+
self.tool_call_regex = re.compile(r"\[{.*?}\]", re.DOTALL)
|
| 65 |
+
|
| 66 |
+
def extract_tool_calls(
|
| 67 |
+
self, model_output: str, request: ChatCompletionRequest
|
| 68 |
+
) -> ExtractedToolCallInformation:
|
| 69 |
+
"""
|
| 70 |
+
Extract the tool calls from a complete model response.
|
| 71 |
+
"""
|
| 72 |
+
# case -- if the response doesn't contain JSON, return a text response
|
| 73 |
+
if not model_output.startswith("{"):
|
| 74 |
+
return ExtractedToolCallInformation(
|
| 75 |
+
tools_called=False, tool_calls=[], content=model_output
|
| 76 |
+
)
|
| 77 |
+
|
| 78 |
+
try:
|
| 79 |
+
# load the JSON, and then use it to build the Function and
|
| 80 |
+
# Tool Call
|
| 81 |
+
dec = JSONDecoder()
|
| 82 |
+
function_call_arr = []
|
| 83 |
+
|
| 84 |
+
start_idx = 0
|
| 85 |
+
while start_idx < len(model_output):
|
| 86 |
+
try:
|
| 87 |
+
(obj, end_idx) = dec.raw_decode(model_output[start_idx:])
|
| 88 |
+
start_idx += end_idx
|
| 89 |
+
# Skip any separators like semicolons or commas
|
| 90 |
+
while start_idx < len(model_output) and model_output[start_idx] in [
|
| 91 |
+
";",
|
| 92 |
+
",",
|
| 93 |
+
" ",
|
| 94 |
+
]:
|
| 95 |
+
start_idx += 1
|
| 96 |
+
function_call_arr.append(obj)
|
| 97 |
+
except json.JSONDecodeError:
|
| 98 |
+
break
|
| 99 |
+
|
| 100 |
+
tool_calls: list[ToolCall] = [
|
| 101 |
+
ToolCall(
|
| 102 |
+
type="function",
|
| 103 |
+
function=FunctionCall(
|
| 104 |
+
name=raw_function_call["name"],
|
| 105 |
+
# function call args are JSON but as a string
|
| 106 |
+
arguments=json.dumps(
|
| 107 |
+
raw_function_call["arguments"]
|
| 108 |
+
if "arguments" in raw_function_call
|
| 109 |
+
else raw_function_call["parameters"]
|
| 110 |
+
),
|
| 111 |
+
),
|
| 112 |
+
)
|
| 113 |
+
for raw_function_call in function_call_arr
|
| 114 |
+
]
|
| 115 |
+
|
| 116 |
+
return ExtractedToolCallInformation(
|
| 117 |
+
tools_called=True, tool_calls=tool_calls, content=None
|
| 118 |
+
)
|
| 119 |
+
|
| 120 |
+
except Exception:
|
| 121 |
+
logger.exception("Error in extracting tool call from response.")
|
| 122 |
+
# return information to just treat the tool call as regular JSON
|
| 123 |
+
return ExtractedToolCallInformation(
|
| 124 |
+
tools_called=False, tool_calls=[], content=model_output
|
| 125 |
+
)
|
| 126 |
+
|
| 127 |
+
def extract_tool_calls_streaming(
|
| 128 |
+
self,
|
| 129 |
+
previous_text: str,
|
| 130 |
+
current_text: str,
|
| 131 |
+
delta_text: str,
|
| 132 |
+
previous_token_ids: Sequence[int],
|
| 133 |
+
current_token_ids: Sequence[int],
|
| 134 |
+
delta_token_ids: Sequence[int],
|
| 135 |
+
request: ChatCompletionRequest,
|
| 136 |
+
) -> Union[DeltaMessage, None]:
|
| 137 |
+
|
| 138 |
+
# Skip if not JSON format
|
| 139 |
+
if not current_text.startswith("{"):
|
| 140 |
+
return DeltaMessage(content=delta_text)
|
| 141 |
+
|
| 142 |
+
# bit mask flags for partial JSON parsing
|
| 143 |
+
flags = Allow.ALL if self.current_tool_name_sent else Allow.ALL & ~Allow.STR
|
| 144 |
+
try:
|
| 145 |
+
tool_call_arr = []
|
| 146 |
+
is_complete = []
|
| 147 |
+
try:
|
| 148 |
+
start_idx = 0
|
| 149 |
+
while start_idx < len(current_text):
|
| 150 |
+
(obj, end_idx) = partial_json_loads(current_text[start_idx:], flags)
|
| 151 |
+
is_complete.append(
|
| 152 |
+
is_complete_json(current_text[start_idx : start_idx + end_idx])
|
| 153 |
+
)
|
| 154 |
+
start_idx += end_idx
|
| 155 |
+
# Skip any separators like semicolons or commas
|
| 156 |
+
while start_idx < len(current_text) and current_text[start_idx] in [
|
| 157 |
+
";",
|
| 158 |
+
",",
|
| 159 |
+
" ",
|
| 160 |
+
]:
|
| 161 |
+
start_idx += 1
|
| 162 |
+
|
| 163 |
+
# Handle parameters field as arguments if needed
|
| 164 |
+
if "parameters" in obj:
|
| 165 |
+
assert (
|
| 166 |
+
"arguments" not in obj
|
| 167 |
+
), "model generated both parameters and arguments"
|
| 168 |
+
obj["arguments"] = obj["parameters"]
|
| 169 |
+
tool_call_arr.append(obj)
|
| 170 |
+
except partial_json_parser.core.exceptions.MalformedJSON:
|
| 171 |
+
logger.debug("not enough tokens to parse into JSON yet")
|
| 172 |
+
return None
|
| 173 |
+
|
| 174 |
+
# select as the current tool call the one we're on the state at
|
| 175 |
+
current_tool_call: dict = (
|
| 176 |
+
tool_call_arr[self.current_tool_id] if len(tool_call_arr) > 0 else {}
|
| 177 |
+
)
|
| 178 |
+
|
| 179 |
+
# case -- if no tokens have been streamed for the tool, e.g.
|
| 180 |
+
# only the array brackets, stream nothing
|
| 181 |
+
if len(tool_call_arr) == 0:
|
| 182 |
+
return None
|
| 183 |
+
|
| 184 |
+
# case: we are starting a new tool in the array
|
| 185 |
+
# -> array has > 0 length AND length has moved past cursor
|
| 186 |
+
elif (
|
| 187 |
+
len(tool_call_arr) > 0 and len(tool_call_arr) > self.current_tool_id + 1
|
| 188 |
+
):
|
| 189 |
+
if self.current_tool_id >= 0:
|
| 190 |
+
cur_arguments = current_tool_call.get("arguments")
|
| 191 |
+
if cur_arguments:
|
| 192 |
+
cur_args_json = json.dumps(cur_arguments)
|
| 193 |
+
sent = len(self.streamed_args_for_tool[self.current_tool_id])
|
| 194 |
+
argument_diff = cur_args_json[sent:]
|
| 195 |
+
|
| 196 |
+
logger.debug("got arguments diff: %s", argument_diff)
|
| 197 |
+
delta = DeltaMessage(
|
| 198 |
+
tool_calls=[
|
| 199 |
+
DeltaToolCall(
|
| 200 |
+
index=self.current_tool_id,
|
| 201 |
+
function=DeltaFunctionCall(
|
| 202 |
+
arguments=argument_diff
|
| 203 |
+
).model_dump(exclude_none=True),
|
| 204 |
+
)
|
| 205 |
+
]
|
| 206 |
+
)
|
| 207 |
+
self.streamed_args_for_tool[
|
| 208 |
+
self.current_tool_id
|
| 209 |
+
] += argument_diff
|
| 210 |
+
else:
|
| 211 |
+
delta = None
|
| 212 |
+
else:
|
| 213 |
+
delta = None
|
| 214 |
+
# re-set stuff pertaining to progress in the current tool
|
| 215 |
+
self.current_tool_id = len(tool_call_arr) - 1
|
| 216 |
+
self.current_tool_name_sent = False
|
| 217 |
+
self.streamed_args_for_tool.append("")
|
| 218 |
+
logger.debug("starting on new tool %d", self.current_tool_id)
|
| 219 |
+
return delta
|
| 220 |
+
|
| 221 |
+
# if the current tool name hasn't been sent, send if available
|
| 222 |
+
# - otherwise send nothing
|
| 223 |
+
elif not self.current_tool_name_sent:
|
| 224 |
+
function_name = current_tool_call.get("name")
|
| 225 |
+
if function_name:
|
| 226 |
+
delta = DeltaMessage(
|
| 227 |
+
tool_calls=[
|
| 228 |
+
DeltaToolCall(
|
| 229 |
+
index=self.current_tool_id,
|
| 230 |
+
type="function",
|
| 231 |
+
id=f"chatcmpl-tool-{random_uuid()}",
|
| 232 |
+
function=DeltaFunctionCall(
|
| 233 |
+
name=function_name
|
| 234 |
+
).model_dump(exclude_none=True),
|
| 235 |
+
)
|
| 236 |
+
]
|
| 237 |
+
)
|
| 238 |
+
self.current_tool_name_sent = True
|
| 239 |
+
else:
|
| 240 |
+
delta = None
|
| 241 |
+
|
| 242 |
+
else:
|
| 243 |
+
cur_arguments = current_tool_call.get("arguments")
|
| 244 |
+
delta = None
|
| 245 |
+
|
| 246 |
+
if cur_arguments:
|
| 247 |
+
sent = len(self.streamed_args_for_tool[self.current_tool_id])
|
| 248 |
+
cur_args_json = json.dumps(cur_arguments)
|
| 249 |
+
prev_arguments = self.prev_tool_call_arr[self.current_tool_id].get(
|
| 250 |
+
"arguments"
|
| 251 |
+
)
|
| 252 |
+
|
| 253 |
+
argument_diff = None
|
| 254 |
+
if is_complete[self.current_tool_id]:
|
| 255 |
+
argument_diff = cur_args_json[sent:]
|
| 256 |
+
elif prev_arguments:
|
| 257 |
+
prev_args_json = json.dumps(prev_arguments)
|
| 258 |
+
if cur_args_json != prev_args_json:
|
| 259 |
+
prefix = find_common_prefix(prev_args_json, cur_args_json)
|
| 260 |
+
argument_diff = prefix[sent:]
|
| 261 |
+
|
| 262 |
+
if argument_diff is not None:
|
| 263 |
+
delta = DeltaMessage(
|
| 264 |
+
tool_calls=[
|
| 265 |
+
DeltaToolCall(
|
| 266 |
+
index=self.current_tool_id,
|
| 267 |
+
function=DeltaFunctionCall(
|
| 268 |
+
arguments=argument_diff
|
| 269 |
+
).model_dump(exclude_none=True),
|
| 270 |
+
)
|
| 271 |
+
]
|
| 272 |
+
)
|
| 273 |
+
self.streamed_args_for_tool[
|
| 274 |
+
self.current_tool_id
|
| 275 |
+
] += argument_diff
|
| 276 |
+
|
| 277 |
+
self.prev_tool_call_arr = tool_call_arr
|
| 278 |
+
return delta
|
| 279 |
+
|
| 280 |
+
except Exception:
|
| 281 |
+
logger.exception("Error trying to handle streaming tool call.")
|
| 282 |
+
logger.debug(
|
| 283 |
+
"Skipping chunk as a result of tool streaming extraction error"
|
| 284 |
+
)
|
| 285 |
+
return None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|