github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #6505] glm4 model function call support #50606

New Issue
Open
opened 2026-04-28 16:29:32 -05:00 by GiteaMirror · 13 comments
GiteaMirror commented 2026-04-28 16:29:32 -05:00
Owner
Copy Link

Originally created by @EntropyYue on GitHub (Aug 25, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/6505

The glm4 model supports function calls, but the templates in the model repository do not.

Originally created by @EntropyYue on GitHub (Aug 25, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/6505 The glm4 model supports function calls, but the templates in the model repository do not.
GiteaMirror added the modelfeature request labels 2026-04-28 16:29:32 -05:00
GiteaMirror commented 2026-04-28 16:29:34 -05:00
Author
Owner
Copy Link

@EntropyYue commented on GitHub (Aug 26, 2024):

FROM glm4:9b

SYSTEM """
你是一个名为 GLM-4 的人工智能助手。你是基于智谱AI训练的语言模型 GLM-4 模型开发的，你的任务是针对用户的问题和要求提供适当的答复和支持。
"""

TEMPLATE """
[gMASK]<sop>
{{ if .Messages }}
{{- if or .System .Tools }}<|system|>
{{- if .System }}

{{ .System }}
{{- end }}
{{- if .Tools }}

你是一个名为 GLM-4 的人工智能助手。你是基于智谱AI训练的语言模型 GLM-4 模型开发的，你的任务是针对用户的问题和要求提供适当的答复和支持。当您收到函数调用请求时，使用输出格式化原始使用问题的答案。

{{- end }}
{{- end }}<|eot_id|>
{{- $lastUserIdx := -1 }}

{{- range $i, $_ := .Messages }}
{{- with eq .Role "user" }}
{{- $lastUserIdx = $i }}{{ end }}
{{- end }}

{{ $lastUserIdx }}

{{- range $i, $_ := .Messages }}
{{- $last := eq (len (slice $.Messages $i)) 1 }}
{{- if eq .Role "user" }}<|user|>
{{- if and $.Tools (eq $i $lastUserIdx) }}

给定以下函数，请为函数调用返回一个JSON，其中包含最能回答给定提示的正确参数。

以以下格式响应 {"name": function name, "parameters": dictionary of argument name and its value}。 不要使用变量。

{{ $.Tools }}
{{- end }}

{{ .Content }}<|eot_id|>{{ if $last }}<|assistant|>

{{ end }}
{{- else if eq .Role "assistant" }}<|assistant|>
{{- if .ToolCalls }}

{{- range .ToolCalls }}{"name": "{{ .Function.Name }}", "parameters": {{ .Function.Arguments }}}{{ end }}
{{- else }}

{{ .Content }}{{ if not $last }}<|eot_id|>{{ end }}
{{- end }}
{{- else if eq .Role "tool" }}Tools:

{{ .Content }}<|eot_id|>{{ if $last }}<|assistant|>

{{ end }}
{{- end }}
{{- end }}
{{ if .System }}<|system|>
{{ .System }}{{ end }}{{ if .Prompt }}<|user|>
{{ .Prompt }}{{ end }}<|assistant|>
{{ .Response }}
{{ end }}
"""

PARAMETER stop "<|system|>"
PARAMETER stop "<|user|>"
PARAMETER stop "<|assistant|>"
PARAMETER stop "<|eot_id|>"
<!-- gh-comment-id:2310577684 --> @EntropyYue commented on GitHub (Aug 26, 2024): ``` FROM glm4:9b SYSTEM """ 你是一个名为 GLM-4 的人工智能助手。你是基于智谱AI训练的语言模型 GLM-4 模型开发的，你的任务是针对用户的问题和要求提供适当的答复和支持。 """ TEMPLATE """ [gMASK]<sop> {{ if .Messages }} {{- if or .System .Tools }}<|system|> {{- if .System }} {{ .System }} {{- end }} {{- if .Tools }} 你是一个名为 GLM-4 的人工智能助手。你是基于智谱AI训练的语言模型 GLM-4 模型开发的，你的任务是针对用户的问题和要求提供适当的答复和支持。当您收到函数调用请求时，使用输出格式化原始使用问题的答案。 {{- end }} {{- end }}<|eot_id|> {{- $lastUserIdx := -1 }} {{- range $i, $_ := .Messages }} {{- with eq .Role "user" }} {{- $lastUserIdx = $i }}{{ end }} {{- end }} {{ $lastUserIdx }} {{- range $i, $_ := .Messages }} {{- $last := eq (len (slice $.Messages $i)) 1 }} {{- if eq .Role "user" }}<|user|> {{- if and $.Tools (eq $i $lastUserIdx) }} 给定以下函数，请为函数调用返回一个JSON，其中包含最能回答给定提示的正确参数。 以以下格式响应 {"name": function name, "parameters": dictionary of argument name and its value}。 不要使用变量。 {{ $.Tools }} {{- end }} {{ .Content }}<|eot_id|>{{ if $last }}<|assistant|> {{ end }} {{- else if eq .Role "assistant" }}<|assistant|> {{- if .ToolCalls }} {{- range .ToolCalls }}{"name": "{{ .Function.Name }}", "parameters": {{ .Function.Arguments }}}{{ end }} {{- else }} {{ .Content }}{{ if not $last }}<|eot_id|>{{ end }} {{- end }} {{- else if eq .Role "tool" }}Tools: {{ .Content }}<|eot_id|>{{ if $last }}<|assistant|> {{ end }} {{- end }} {{- end }} {{ if .System }}<|system|> {{ .System }}{{ end }}{{ if .Prompt }}<|user|> {{ .Prompt }}{{ end }}<|assistant|> {{ .Response }} {{ end }} """ PARAMETER stop "<|system|>" PARAMETER stop "<|user|>" PARAMETER stop "<|assistant|>" PARAMETER stop "<|eot_id|>" ```
GiteaMirror commented 2026-04-28 16:29:37 -05:00
Author
Owner
Copy Link

@EntropyYue commented on GitHub (Aug 26, 2024):

Sorry, this type of code is too complex, can someone provide support?

<!-- gh-comment-id:2310582176 --> @EntropyYue commented on GitHub (Aug 26, 2024): Sorry, this type of code is too complex, can someone provide support?
GiteaMirror commented 2026-04-28 16:29:40 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 27, 2024):

This is a modelfile that adds tool support to glm4. You save this in a file called Modelfile, and then create a new model:

ollama create glm4-tool:9b -f Modelfile

Then call it with tools:

$ curl -s localhost:11434/v1/chat/completions -d '{"model": "glm4-tool:9b","tools":[{"type":"function","function": {}}], "messages": [{"role":"user","content":"weather in zurich"}], "stream": false}' | jq
{
  "id": "chatcmpl-595",
  "object": "chat.completion",
  "created": 1724750444,
  "model": "glm4-tool:9b",
  "system_fingerprint": "fp_ollama",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "",
        "tool_calls": [
          {
            "id": "call_2h4gimgk",
            "type": "function",
            "function": {
              "name": "get_weather",
              "arguments": "{\"location\":\"Zurich\"}"
            }
          }
        ]
      },
      "finish_reason": "tool_calls"
    }
  ],
  "usage": {
    "prompt_tokens": 247,
    "completion_tokens": 18,
    "total_tokens": 265
  }
}
<!-- gh-comment-id:2312004680 --> @rick-github commented on GitHub (Aug 27, 2024): This is a modelfile that adds tool support to glm4. You save this in a file called Modelfile, and then create a new model: ``` ollama create glm4-tool:9b -f Modelfile ``` Then call it with tools: ``` $ curl -s localhost:11434/v1/chat/completions -d '{"model": "glm4-tool:9b","tools":[{"type":"function","function": {}}], "messages": [{"role":"user","content":"weather in zurich"}], "stream": false}' | jq { "id": "chatcmpl-595", "object": "chat.completion", "created": 1724750444, "model": "glm4-tool:9b", "system_fingerprint": "fp_ollama", "choices": [ { "index": 0, "message": { "role": "assistant", "content": "", "tool_calls": [ { "id": "call_2h4gimgk", "type": "function", "function": { "name": "get_weather", "arguments": "{\"location\":\"Zurich\"}" } } ] }, "finish_reason": "tool_calls" } ], "usage": { "prompt_tokens": 247, "completion_tokens": 18, "total_tokens": 265 } } ```
GiteaMirror commented 2026-04-28 16:29:42 -05:00
Author
Owner
Copy Link

@EntropyYue commented on GitHub (Aug 27, 2024):

这是一个为 glm4 添加工具支持的 modelfile。将其保存在名为 Modelfile 的文件中，然后创建一个新模型：

ollama create glm4-tool:9b -f Modelfile

然后用工具调用它：

$ curl -s localhost:11434/v1/chat/completions -d '{"model": "glm4-tool:9b","tools":[{"type":"function","function": {}}], "messages": [{"role":"user","content":"weather in zurich"}], "stream": false}' | jq
{
  "id": "chatcmpl-595",
  "object": "chat.completion",
  "created": 1724750444,
  "model": "glm4-tool:9b",
  "system_fingerprint": "fp_ollama",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "",
        "tool_calls": [
          {
            "id": "call_2h4gimgk",
            "type": "function",
            "function": {
              "name": "get_weather",
              "arguments": "{\"location\":\"Zurich\"}"
            }
          }
        ]
      },
      "finish_reason": "tool_calls"
    }
  ],
  "usage": {
    "prompt_tokens": 247,
    "completion_tokens": 18,
    "total_tokens": 265
  }
}

I am aware of the functionality of this modelfile, but I think it is not perfect. I need someone to make some improvements to it.

<!-- gh-comment-id:2312397876 --> @EntropyYue commented on GitHub (Aug 27, 2024): > 这是一个为 glm4 添加工具支持的 modelfile。将其保存在名为 Modelfile 的文件中，然后创建一个新模型： > > ``` > ollama create glm4-tool:9b -f Modelfile > ``` > > 然后用工具调用它： > > ``` > $ curl -s localhost:11434/v1/chat/completions -d '{"model": "glm4-tool:9b","tools":[{"type":"function","function": {}}], "messages": [{"role":"user","content":"weather in zurich"}], "stream": false}' | jq > { > "id": "chatcmpl-595", > "object": "chat.completion", > "created": 1724750444, > "model": "glm4-tool:9b", > "system_fingerprint": "fp_ollama", > "choices": [ > { > "index": 0, > "message": { > "role": "assistant", > "content": "", > "tool_calls": [ > { > "id": "call_2h4gimgk", > "type": "function", > "function": { > "name": "get_weather", > "arguments": "{\"location\":\"Zurich\"}" > } > } > ] > }, > "finish_reason": "tool_calls" > } > ], > "usage": { > "prompt_tokens": 247, > "completion_tokens": 18, > "total_tokens": 265 > } > } > ``` I am aware of the functionality of this modelfile, but I think it is not perfect. I need someone to make some improvements to it.
GiteaMirror commented 2026-04-28 16:29:43 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 27, 2024):

What's wrong with it?

<!-- gh-comment-id:2312401208 --> @rick-github commented on GitHub (Aug 27, 2024): What's wrong with it?
GiteaMirror commented 2026-04-28 16:29:44 -05:00
Author
Owner
Copy Link

@EntropyYue commented on GitHub (Aug 28, 2024):

它有什么问题？

When using this prompt template, the model's perplexity increases significantly

<!-- gh-comment-id:2314605524 --> @EntropyYue commented on GitHub (Aug 28, 2024): > 它有什么问题？ When using this prompt template, the model's perplexity increases significantly
GiteaMirror commented 2026-04-28 16:29:45 -05:00
Author
Owner
Copy Link

@EntropyYue commented on GitHub (Sep 8, 2024):

FROM glm4:9b

SYSTEM """
你是一个名为 GLM-4 的人工智能助手。你是基于智谱AI训练的语言模型 GLM-4 模型开发的，你的任务是针对用户的问题和要求提供适当的答复和支持。
"""

TEMPLATE """
[gMASK]<sop>
{{- if or .System .Tools }}<|system|>
{{- if .System }}

{{ .System }}
{{- end }}
{{- if .Tools }}

你是能够调用函数的大语言模型，你的任务是针对用户的问题和要求提供适当的答复和支持。不要对可用的函数做出假设，当您收到函数调用请求时，使用输出格式化原始使用问题的答案。

{{- end }}
{{- end }}
{{- $lastUserIdx := -1 }}

{{- range $i, $_ := .Messages }}
{{- with eq .Role "user" }}
{{- $lastUserIdx = $i }}{{ end }}
{{- end }}

{{- range $i, $_ := .Messages }}
{{- $last := eq (len (slice $.Messages $i)) 1 }}
{{- if eq .Role "user" }}<|user|>
{{- if and $.Tools (eq $i $lastUserIdx) }}

你可以调用以下函数，请为函数调用返回一个JSON，其中包含最能回答给定提示的正确参数。
作为一个能够调用函数的大语言模型，你可以完成文本生成以及调用这些函数的任务

如果任务需要调用函数，请正确地按照以下格式在 <tool_call></tool_call> xml标签中做出响应，不要使用变量，如果不需要使用函数，返回一个空的字典：
<tool_call>
{"name": 函数名称, "parameters": 参数键值对字典}
</tool_call>

{{ range $.Tools }}
{{- . }}
{{ end }}
{{- end }}

{{ .Content }}{{ if $last }}<|assistant|>

{{ end }}
{{- else if eq .Role "assistant" }}<|assistant|>
{{- if .ToolCalls }}
<tool_call>
{{- range .ToolCalls }}{"name": "{{ .Function.Name }}", "parameters": {{ .Function.Arguments }}}{{ end }}
</tool_call>
{{- else }}

{{ .Content }}{{ if not $last }}{{ end }}
{{- end }}
{{- else if eq .Role "tool" }}ipython

{{ .Content }}{{ if $last }}<|assistant|>

{{ end }}
{{- end }}
{{- end }}
"""

PARAMETER stop "<|system|>"
PARAMETER stop "<|user|>"
PARAMETER stop "<|assistant|>"
<!-- gh-comment-id:2336774942 --> @EntropyYue commented on GitHub (Sep 8, 2024): ``` FROM glm4:9b SYSTEM """ 你是一个名为 GLM-4 的人工智能助手。你是基于智谱AI训练的语言模型 GLM-4 模型开发的，你的任务是针对用户的问题和要求提供适当的答复和支持。 """ TEMPLATE """ [gMASK]<sop> {{- if or .System .Tools }}<|system|> {{- if .System }} {{ .System }} {{- end }} {{- if .Tools }} 你是能够调用函数的大语言模型，你的任务是针对用户的问题和要求提供适当的答复和支持。不要对可用的函数做出假设，当您收到函数调用请求时，使用输出格式化原始使用问题的答案。 {{- end }} {{- end }} {{- $lastUserIdx := -1 }} {{- range $i, $_ := .Messages }} {{- with eq .Role "user" }} {{- $lastUserIdx = $i }}{{ end }} {{- end }} {{- range $i, $_ := .Messages }} {{- $last := eq (len (slice $.Messages $i)) 1 }} {{- if eq .Role "user" }}<|user|> {{- if and $.Tools (eq $i $lastUserIdx) }} 你可以调用以下函数，请为函数调用返回一个JSON，其中包含最能回答给定提示的正确参数。 作为一个能够调用函数的大语言模型，你可以完成文本生成以及调用这些函数的任务 如果任务需要调用函数，请正确地按照以下格式在 <tool_call></tool_call> xml标签中做出响应，不要使用变量，如果不需要使用函数，返回一个空的字典： <tool_call> {"name": 函数名称, "parameters": 参数键值对字典} </tool_call> {{ range $.Tools }} {{- . }} {{ end }} {{- end }} {{ .Content }}{{ if $last }}<|assistant|> {{ end }} {{- else if eq .Role "assistant" }}<|assistant|> {{- if .ToolCalls }} <tool_call> {{- range .ToolCalls }}{"name": "{{ .Function.Name }}", "parameters": {{ .Function.Arguments }}}{{ end }} </tool_call> {{- else }} {{ .Content }}{{ if not $last }}{{ end }} {{- end }} {{- else if eq .Role "tool" }}ipython {{ .Content }}{{ if $last }}<|assistant|> {{ end }} {{- end }} {{- end }} """ PARAMETER stop "<|system|>" PARAMETER stop "<|user|>" PARAMETER stop "<|assistant|>" ```
GiteaMirror commented 2026-04-28 16:29:46 -05:00
Author
Owner
Copy Link

@EntropyYue commented on GitHub (Sep 8, 2024):

Utilize the above template to enable the glm4 model with function call capability

<!-- gh-comment-id:2336775124 --> @EntropyYue commented on GitHub (Sep 8, 2024): Utilize the above template to enable the glm4 model with function call capability
GiteaMirror commented 2026-04-28 16:29:47 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Sep 8, 2024):

@xiaopa233 Could you edit the post and code markdown tags ( ```) around it so that special characters are not mis-interpreted.

<!-- gh-comment-id:2336783803 --> @rick-github commented on GitHub (Sep 8, 2024): @xiaopa233 Could you edit the post and code markdown tags ( \`\`\`) around it so that special characters are not mis-interpreted.
GiteaMirror commented 2026-04-28 16:29:50 -05:00
Author
Owner
Copy Link

@EntropyYue commented on GitHub (Sep 8, 2024):

The editing is complete, sorry I forgot about it

<!-- gh-comment-id:2336812858 --> @EntropyYue commented on GitHub (Sep 8, 2024): The editing is complete, sorry I forgot about it
GiteaMirror commented 2026-04-28 16:29:55 -05:00
Author
Owner
Copy Link

@LuckLittleBoy commented on GitHub (Oct 12, 2024):

Utilize the above template to enable the glm4 model with function call capability

The above templates do not have a function call response capability

<!-- gh-comment-id:2408462495 --> @LuckLittleBoy commented on GitHub (Oct 12, 2024): > Utilize the above template to enable the glm4 model with function call capability The above templates do not have a function call response capability
GiteaMirror commented 2026-04-28 16:29:58 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Oct 14, 2024):

It seems to work fine:

$ ./tool-test.py --model glm4-tool:9b --prompt 'what is the time?'
calling get_datetime({})
The current time is 12:29.
$ ./tool-test.py --model glm4-tool:9b --prompt 'what is 1.34 ^ 10?'
calling power({'x': 1.34, 'y': 10})
The result of \(1.34^{10}\) is approximately 18.666.
$ ./tool-test.py --model glm4-tool:9b --prompt 'what operating system are you running on?'
calling get_operating_system({})
The operating system I am running on is Linux-5.15.0-94-generic-x86_64-with-glibc2.35.
$ ./tool-test.py --model glm4-tool:9b --prompt "summarize the contents of the file ./1.txt"
calling read_file({'filename': './1.txt'})
The contents of the file './1.txt' include the following lines: 'hi', 'ok', 'hello', and 'hi there'.
$ ./tool-test.py --model glm4-tool:9b --prompt "when is sunrise in Auckland tomorrow?"
calling get_datetime({})
calling search_web({'query': 'Auckland sunrise time October 15, 2024'})
Sunrise in Auckland tomorrow is expected to be around 06:38.

It does fail occasionally:

$ ./tool-test.py --model glm4-tool:9b --prompt "when is sunrise in Auckland tomorrow?"
I'm sorry, but I cannot directly retrieve real-time weather data or calculate sunrise times. However, I can help you find this information if you provide the name of a tool that can be used to get the weather details for a specific location
<!-- gh-comment-id:2410860751 --> @rick-github commented on GitHub (Oct 14, 2024): It seems to work fine: ```console $ ./tool-test.py --model glm4-tool:9b --prompt 'what is the time?' calling get_datetime({}) The current time is 12:29. $ ./tool-test.py --model glm4-tool:9b --prompt 'what is 1.34 ^ 10?' calling power({'x': 1.34, 'y': 10}) The result of \(1.34^{10}\) is approximately 18.666. $ ./tool-test.py --model glm4-tool:9b --prompt 'what operating system are you running on?' calling get_operating_system({}) The operating system I am running on is Linux-5.15.0-94-generic-x86_64-with-glibc2.35. $ ./tool-test.py --model glm4-tool:9b --prompt "summarize the contents of the file ./1.txt" calling read_file({'filename': './1.txt'}) The contents of the file './1.txt' include the following lines: 'hi', 'ok', 'hello', and 'hi there'. $ ./tool-test.py --model glm4-tool:9b --prompt "when is sunrise in Auckland tomorrow?" calling get_datetime({}) calling search_web({'query': 'Auckland sunrise time October 15, 2024'}) Sunrise in Auckland tomorrow is expected to be around 06:38. ``` It does fail occasionally: ```console $ ./tool-test.py --model glm4-tool:9b --prompt "when is sunrise in Auckland tomorrow?" I'm sorry, but I cannot directly retrieve real-time weather data or calculate sunrise times. However, I can help you find this information if you provide the name of a tool that can be used to get the weather details for a specific location ```
GiteaMirror commented 2026-04-28 16:30:02 -05:00
Author
Owner
Copy Link

@EntropyYue commented on GitHub (Oct 14, 2024):

Occasionally, work failures are normal phenomena; you need a more stable model with formatted output like qwen2.5

<!-- gh-comment-id:2410941699 --> @EntropyYue commented on GitHub (Oct 14, 2024): Occasionally, work failures are normal phenomena; you need a more stable model with formatted output like qwen2.5
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label feature request model
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#50606
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?