Compare commits
1 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| 4a7f630cad |
+48
-21
@@ -271,13 +271,23 @@ def parse_tool_calls(reply, tools=None):
|
|||||||
salvaged.append(name)
|
salvaged.append(name)
|
||||||
calls.append({"id": "call_" + uuid.uuid4().hex[:24], "type": "function",
|
calls.append({"id": "call_" + uuid.uuid4().hex[:24], "type": "function",
|
||||||
"function": {"name": name, "arguments": json.dumps(args, ensure_ascii=False)}})
|
"function": {"name": name, "arguments": json.dumps(args, ensure_ascii=False)}})
|
||||||
if tools and not calls and re.search(r"</?tool_call>|</?arg_key>|</?arg_value>", reply):
|
if tools and not calls:
|
||||||
# Diagnosi per la #401: il client ha dichiarato i tools e il modello ha PROVATO la
|
# Diagnosi #401: distingui i due modi di fallire. (a) marker presenti ma il parse
|
||||||
# sintassi, ma il parse rigoroso non ha agganciato nulla (tipico output int4 storpiato).
|
# rigoroso non aggancia -> output storpiato (int4). (b) nessun marker -> il modello
|
||||||
# EN: #401 field diagnosis: tools were declared and the model attempted the syntax,
|
# NON ha nemmeno provato a chiamare i tool (prompt/behavior, non parsing). Sapere QUALE
|
||||||
# EN: but the strict parse matched nothing (typically quantization-mangled output).
|
# dei due e' il nodo del report: senza distinguerli si tira a indovinare.
|
||||||
sys.stderr.write("[api] tools declared and tool-call markers present, but no call "
|
# EN: #401: separate the two failure modes -- markers-but-unparsed (mangled int4 output)
|
||||||
"parsed -- output may be quantization-mangled; try COLI_TOOL_SALVAGE=1\n")
|
# EN: vs no-markers-at-all (the model never attempted a tool call: a prompt/behavior
|
||||||
|
# EN: issue, not a parsing one). COLI_DEBUG dumps the raw reply so we can SEE which.
|
||||||
|
if re.search(r"</?tool_call>|</?arg_key>|</?arg_value>", reply):
|
||||||
|
sys.stderr.write("[api] #401: tools declared, tool-call markers present but nothing "
|
||||||
|
"parsed -- output likely quantization-mangled; try COLI_TOOL_SALVAGE=1\n")
|
||||||
|
else:
|
||||||
|
sys.stderr.write("[api] #401: tools declared but the reply has NO tool-call markers -- "
|
||||||
|
"the model answered in plain text (prompt/behavior, not parsing)\n")
|
||||||
|
if os.environ.get("COLI_DEBUG"):
|
||||||
|
head = reply[:800].replace("\n", "\\n")
|
||||||
|
sys.stderr.write("[api] #401 raw model reply (first 800 chars): %s\n" % head)
|
||||||
sys.stderr.flush()
|
sys.stderr.flush()
|
||||||
text = _BOX_RE.sub("", reply)
|
text = _BOX_RE.sub("", reply)
|
||||||
if THINK_CLOSE in text:
|
if THINK_CLOSE in text:
|
||||||
@@ -310,32 +320,47 @@ def render_chat(messages, enable_thinking=False, reasoning_effort=None, tools=No
|
|||||||
if ((t.get("function", t) if isinstance(t, dict) else {}).get("name") == forced)]
|
if ((t.get("function", t) if isinstance(t, dict) else {}).get("name") == forced)]
|
||||||
elif tool_choice == "none":
|
elif tool_choice == "none":
|
||||||
tools = None # the client forbade tools: do not offer them
|
tools = None # the client forbade tools: do not offer them
|
||||||
|
# AUTHORITATIVE GLM-5.2 tool-declaration block (byte-matches chat_template.jinja): the
|
||||||
|
# `# Tools` + <tools></tools> XML structure is what the model was trained on. A made-up
|
||||||
|
# preamble makes it hallucinate other frameworks' syntax (e.g. `end_action`).
|
||||||
|
tool_block = ""
|
||||||
if tools:
|
if tools:
|
||||||
# AUTHORITATIVE GLM-5.2 tool-declaration block (byte-matches chat_template.jinja): the
|
parts = ["\n# Tools\n\nYou may call one or more functions to assist with the user "
|
||||||
# `# Tools` + <tools></tools> XML structure is what the model was trained on. A made-up
|
"query.\n\nYou are provided with function signatures within <tools></tools> "
|
||||||
# preamble makes it hallucinate other frameworks' syntax (e.g. `end_action`).
|
"XML tags:\n<tools>\n"]
|
||||||
prompt.append("<|system|>\n# Tools\n\nYou may call one or more functions to assist with the "
|
|
||||||
"user query.\n\nYou are provided with function signatures within <tools></tools> "
|
|
||||||
"XML tags:\n<tools>\n")
|
|
||||||
for tool in tools:
|
for tool in tools:
|
||||||
fn = tool.get("function", tool) if isinstance(tool, dict) else {}
|
fn = tool.get("function", tool) if isinstance(tool, dict) else {}
|
||||||
clean = {k: v for k, v in fn.items() if k not in ("defer_loading", "strict")}
|
clean = {k: v for k, v in fn.items() if k not in ("defer_loading", "strict")}
|
||||||
prompt.append(json.dumps(clean, ensure_ascii=False) + "\n")
|
parts.append(json.dumps(clean, ensure_ascii=False) + "\n")
|
||||||
prompt.append("</tools>\n\nFor each function call, output the function name and arguments "
|
parts.append("</tools>\n\nFor each function call, output the function name and arguments "
|
||||||
"within the following XML format:\n<tool_call>{function-name}"
|
"within the following XML format:\n<tool_call>{function-name}"
|
||||||
"<arg_key>{arg-key-1}</arg_key><arg_value>{arg-value-1}</arg_value>"
|
"<arg_key>{arg-key-1}</arg_key><arg_value>{arg-value-1}</arg_value>"
|
||||||
"<arg_key>{arg-key-2}</arg_key><arg_value>{arg-value-2}</arg_value>...</tool_call>")
|
"<arg_key>{arg-key-2}</arg_key><arg_value>{arg-value-2}</arg_value>...</tool_call>")
|
||||||
if forced:
|
if forced:
|
||||||
prompt.append(f"\n\nYou must call the function `{forced}`. Do not answer directly.")
|
parts.append(f"\n\nYou must call the function `{forced}`. Do not answer directly.")
|
||||||
elif tool_choice == "required":
|
elif tool_choice == "required":
|
||||||
prompt.append("\n\nYou must call one of the functions above. Do not answer directly.")
|
parts.append("\n\nYou must call one of the functions above. Do not answer directly.")
|
||||||
|
tool_block = "".join(parts)
|
||||||
|
# #401 experiment: a coding CLI sends its own big system prompt, so the default emits TWO
|
||||||
|
# consecutive <|system|> turns (tools, then the client's) -- GLM may prioritize the later
|
||||||
|
# one and never see the tool instructions. COLI_TOOL_SYS_MERGE=1 folds the tool block into
|
||||||
|
# the client's system turn instead (one <|system|>, tools last), to A/B which one actually
|
||||||
|
# makes the model emit tool calls. Default keeps the historical two-block behavior.
|
||||||
|
merge_tools = bool(tool_block) and os.environ.get("COLI_TOOL_SYS_MERGE") == "1"
|
||||||
|
if tool_block and not merge_tools:
|
||||||
|
prompt.append("<|system|>" + tool_block)
|
||||||
|
tools_pending = tool_block if merge_tools else ""
|
||||||
prev_tool = False
|
prev_tool = False
|
||||||
for index, message in enumerate(messages):
|
for index, message in enumerate(messages):
|
||||||
if not isinstance(message, dict):
|
if not isinstance(message, dict):
|
||||||
raise APIError(400, "Each message must be an object.", f"messages.{index}")
|
raise APIError(400, "Each message must be an object.", f"messages.{index}")
|
||||||
role = message.get("role")
|
role = message.get("role")
|
||||||
if role in ("system", "developer"):
|
if role in ("system", "developer"):
|
||||||
prompt.append(f"<|system|>{content_text(message.get('content'), f'messages.{index}.content')}")
|
body = content_text(message.get('content'), f'messages.{index}.content')
|
||||||
|
if tools_pending: # merge mode: fold tools into this system turn
|
||||||
|
body = body + "\n" + tools_pending
|
||||||
|
tools_pending = ""
|
||||||
|
prompt.append(f"<|system|>{body}")
|
||||||
elif role == "user":
|
elif role == "user":
|
||||||
prompt.append(f"<|user|>{content_text(message.get('content'), f'messages.{index}.content')}")
|
prompt.append(f"<|user|>{content_text(message.get('content'), f'messages.{index}.content')}")
|
||||||
elif role == "assistant":
|
elif role == "assistant":
|
||||||
@@ -365,6 +390,8 @@ def render_chat(messages, enable_thinking=False, reasoning_effort=None, tools=No
|
|||||||
raise APIError(400, f"Unsupported message role: {role!r}.",
|
raise APIError(400, f"Unsupported message role: {role!r}.",
|
||||||
f"messages.{index}.role", "unsupported_role")
|
f"messages.{index}.role", "unsupported_role")
|
||||||
prev_tool = (role == "tool")
|
prev_tool = (role == "tool")
|
||||||
|
if tools_pending: # merge mode but no client system turn existed
|
||||||
|
prompt.append("<|system|>" + tools_pending)
|
||||||
prompt.append("<|assistant|><think>" if enable_thinking else
|
prompt.append("<|assistant|><think>" if enable_thinking else
|
||||||
"<|assistant|><think></think>")
|
"<|assistant|><think></think>")
|
||||||
return "".join(prompt)
|
return "".join(prompt)
|
||||||
|
|||||||
Reference in New Issue
Block a user