github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #13519] llama3.2:3b outputs tool calls as JSON in content instead of using tool_calls field #55420

New Issue
Closed
opened 2026-04-29 09:09:21 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-29 09:09:21 -05:00
Owner
Copy Link

Originally created by @sacenox on GitHub (Dec 18, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/13519

What is the issue?

llama3.2:3b outputs tool calls as JSON in content instead of using tool_calls field

Description

When using llama3.2:3b with the Ollama npm library and including tools in the request, the model outputs tool calls as JSON text in the content field instead of using the tool_calls field like other models do.

Expected Behavior

The model should return tool calls in the tool_calls field of the response, similar to how other models (e.g., gpt-oss:latest) handle tool calling.

Actual Behavior

The model outputs tool calls as JSON text in the content field, wrapped in code blocks:

{"name": "query_web_search", "parameters": {"query": "install ollama models"}}

The tool_calls field is empty or not present in the response chunks.

Steps to Reproduce

  1. Use Ollama npm library version 0.6.3
  2. Send a chat request to llama3.2:3b model with tools included
  3. Observe the response - tool calls appear as JSON in content instead of tool_calls

Example code:

import ollama from "ollama";

const tools = [
  {
    type: "function",
    function: {
      name: "query_web_search",
      description: "Perform a web search",
      parameters: {
        type: "object",
        properties: {
          query: {
            type: "string",
            description: "The search query"
          }
        },
        required: ["query"]
      }
    }
  }
];

const stream = await ollama.chat({
  model: "llama3.2:3b",
  messages: [
    { role: "user", content: "Search for install ollama models" }
  ],
  tools: tools,
  stream: true
});

for await (const chunk of stream) {
  console.log("Content:", chunk.message?.content);
  console.log("Tool calls:", chunk.message?.tool_calls);
}

Environment

  • Ollama npm library version: 0.6.3
  • Model: llama3.2:3b
  • Node.js version: >=25.0.0

Additional Context

  • Other models (e.g., gpt-oss, llama3.2:8b) correctly return tool calls in the tool_calls field
  • The issue occurs consistently when tools are included in the request
  • The JSON output in content is properly formatted and parseable, but it's not using the expected tool_calls mechanism

Debug Output Example

When tools are included, the response chunks show:

{
  "role": "assistant",
  "content": "```json\n{\"name\": \"query_web_search\", \"parameters\": {\"query\": \"install ollama models\"}}\n```",
  "tool_calls": null
}

Expected format (as seen with other models):

{
  "role": "assistant",
  "content": "",
  "tool_calls": [
    {
      "id": "call_xxx",
      "function": {
        "name": "query_web_search",
        "arguments": {
          "query": "install ollama models"
        }
      }
    }
  ]
}

Relevant log output

OS

Linux, macOS, Docker

GPU

Nvidia, Apple, Intel

CPU

Apple, Intel

Ollama version

0.13.4

Originally created by @sacenox on GitHub (Dec 18, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/13519 ### What is the issue? # llama3.2:3b outputs tool calls as JSON in content instead of using tool_calls field ## Description When using `llama3.2:3b` with the Ollama npm library and including tools in the request, the model outputs tool calls as JSON text in the `content` field instead of using the `tool_calls` field like other models do. ## Expected Behavior The model should return tool calls in the `tool_calls` field of the response, similar to how other models (e.g., `gpt-oss:latest`) handle tool calling. ## Actual Behavior The model outputs tool calls as JSON text in the `content` field, wrapped in code blocks: ```json {"name": "query_web_search", "parameters": {"query": "install ollama models"}} ``` The `tool_calls` field is empty or not present in the response chunks. ## Steps to Reproduce 1. Use Ollama npm library version `0.6.3` 2. Send a chat request to `llama3.2:3b` model with tools included 3. Observe the response - tool calls appear as JSON in `content` instead of `tool_calls` Example code: ```javascript import ollama from "ollama"; const tools = [ { type: "function", function: { name: "query_web_search", description: "Perform a web search", parameters: { type: "object", properties: { query: { type: "string", description: "The search query" } }, required: ["query"] } } } ]; const stream = await ollama.chat({ model: "llama3.2:3b", messages: [ { role: "user", content: "Search for install ollama models" } ], tools: tools, stream: true }); for await (const chunk of stream) { console.log("Content:", chunk.message?.content); console.log("Tool calls:", chunk.message?.tool_calls); } ``` ## Environment - Ollama npm library version: `0.6.3` - Model: `llama3.2:3b` - Node.js version: `>=25.0.0` ## Additional Context - Other models (e.g., `gpt-oss`, `llama3.2:8b`) correctly return tool calls in the `tool_calls` field - The issue occurs consistently when tools are included in the request - The JSON output in content is properly formatted and parseable, but it's not using the expected `tool_calls` mechanism ## Debug Output Example When tools are included, the response chunks show: ```json { "role": "assistant", "content": "```json\n{\"name\": \"query_web_search\", \"parameters\": {\"query\": \"install ollama models\"}}\n```", "tool_calls": null } ``` Expected format (as seen with other models): ```json { "role": "assistant", "content": "", "tool_calls": [ { "id": "call_xxx", "function": { "name": "query_web_search", "arguments": { "query": "install ollama models" } } } ] } ``` ### Relevant log output ```shell ``` ### OS Linux, macOS, Docker ### GPU Nvidia, Apple, Intel ### CPU Apple, Intel ### Ollama version 0.13.4
GiteaMirror added the bug label 2026-04-29 09:09:21 -05:00
GiteaMirror commented 2026-04-29 09:09:24 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Dec 18, 2025):

Works fine via direct call:

$ curl -s localhost:11434/api/chat -d '{
  "model":"llama3.2:3b",
  "messages":[
    {"role":"user","content":"Search for install ollama models"}
  ],
  "tools":[{
    "type":"function",
    "function":{
      "name":"query_web_search",
      "description":"Perform a web search",
      "parameters":{
        "type":"object",
        "properties":{
          "query":{
            "type":"string",
            "description":"The search query"
          }
        },
        "required":["query"]
      }
    }
  }],
  "stream":false
}' | jq

{
  "model": "llama3.2:3b",
  "created_at": "2025-12-18T14:28:33.538565768Z",
  "message": {
    "role": "assistant",
    "content": "",
    "tool_calls": [
      {
        "id": "call_5nccbtr4",
        "function": {
          "index": 0,
          "name": "query_web_search",
          "arguments": {
            "query": "install olamma models"
          }
        }
      }
    ]
  },
  "done": true,
  "done_reason": "stop",
  "total_duration": 686997782,
  "load_duration": 271693268,
  "prompt_eval_count": 161,
  "prompt_eval_duration": 20326815,
  "eval_count": 21,
  "eval_duration": 381376036
}

The tool call in content looks well-formed but is not recognized by the tool call parser, which may indicate an issue with the template. What's the output of

ollama show --template llama3.2:3b
<!-- gh-comment-id:3670595036 --> @rick-github commented on GitHub (Dec 18, 2025): Works fine via direct call: ```console $ curl -s localhost:11434/api/chat -d '{ "model":"llama3.2:3b", "messages":[ {"role":"user","content":"Search for install ollama models"} ], "tools":[{ "type":"function", "function":{ "name":"query_web_search", "description":"Perform a web search", "parameters":{ "type":"object", "properties":{ "query":{ "type":"string", "description":"The search query" } }, "required":["query"] } } }], "stream":false }' | jq ``` ```json { "model": "llama3.2:3b", "created_at": "2025-12-18T14:28:33.538565768Z", "message": { "role": "assistant", "content": "", "tool_calls": [ { "id": "call_5nccbtr4", "function": { "index": 0, "name": "query_web_search", "arguments": { "query": "install olamma models" } } } ] }, "done": true, "done_reason": "stop", "total_duration": 686997782, "load_duration": 271693268, "prompt_eval_count": 161, "prompt_eval_duration": 20326815, "eval_count": 21, "eval_duration": 381376036 } ``` The tool call in `content` looks well-formed but is not recognized by the tool call parser, which may indicate an issue with the template. What's the output of ``` ollama show --template llama3.2:3b ```
GiteaMirror commented 2026-04-29 09:09:26 -05:00
Author
Owner
Copy Link

@sacenox commented on GitHub (Dec 18, 2025):

Here's the output of the ollama show command:

❯ ollama show --template llama3.2:3b
<|start_header_id|>system<|end_header_id|>

Cutting Knowledge Date: December 2023

{{ if .System }}{{ .System }}
{{- end }}
{{- if .Tools }}When you receive a tool call response, use the output to format an answer to the orginal user question.

You are a helpful assistant with tool calling capabilities.
{{- end }}<|eot_id|>
{{- range $i, $_ := .Messages }}
{{- $last := eq (len (slice $.Messages $i)) 1 }}
{{- if eq .Role "user" }}<|start_header_id|>user<|end_header_id|>
{{- if and $.Tools $last }}

Given the following functions, please respond with a JSON for a function call with its proper arguments that best answers the given prompt.

Respond in the format {"name": function name, "parameters": dictionary of argument name and its value}. Do not use variables.

{{ range $.Tools }}
{{- . }}
{{ end }}
{{ .Content }}<|eot_id|>
{{- else }}

{{ .Content }}<|eot_id|>
{{- end }}{{ if $last }}<|start_header_id|>assistant<|end_header_id|>

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

{{ .Content }}
{{- end }}{{ if not $last }}<|eot_id|>{{ end }}
{{- else if eq .Role "tool" }}<|start_header_id|>ipython<|end_header_id|>

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

{{ end }}
{{- end }}
{{- end }}%

I've also been able to run the same cURL call:

❯ curl -s localhost:11434/api/chat -d '{
  "model":"llama3.2:3b",
  "messages":[
    {"role":"user","content":"Search for install ollama models"}
  ],
  "tools":[{
    "type":"function",
    "function":{
      "name":"query_web_search",
      "description":"Perform a web search",
      "parameters":{
        "type":"object",
        "properties":{
          "query":{
            "type":"string",
            "description":"The search query"
          }
        },
        "required":["query"]
      }
    }
  }],
  "stream":false
}' | jq
{
  "model": "llama3.2:3b",
  "created_at": "2025-12-18T15:29:10.304027Z",
  "message": {
    "role": "assistant",
    "content": "",
    "tool_calls": [
      {
        "id": "call_h8ikz9yn",
        "function": {
          "index": 0,
          "name": "query_web_search",
          "arguments": {
            "query": "install ollama models"
          }
        }
      }
    ]
  },
  "done": true,
  "done_reason": "stop",
  "total_duration": 3659757167,
  "load_duration": 2690591833,
  "prompt_eval_count": 161,
  "prompt_eval_duration": 407400667,
  "eval_count": 22,
  "eval_duration": 548841582
}
<!-- gh-comment-id:3670864588 --> @sacenox commented on GitHub (Dec 18, 2025): Here's the output of the ollama show command: ``` ❯ ollama show --template llama3.2:3b <|start_header_id|>system<|end_header_id|> Cutting Knowledge Date: December 2023 {{ if .System }}{{ .System }} {{- end }} {{- if .Tools }}When you receive a tool call response, use the output to format an answer to the orginal user question. You are a helpful assistant with tool calling capabilities. {{- end }}<|eot_id|> {{- range $i, $_ := .Messages }} {{- $last := eq (len (slice $.Messages $i)) 1 }} {{- if eq .Role "user" }}<|start_header_id|>user<|end_header_id|> {{- if and $.Tools $last }} Given the following functions, please respond with a JSON for a function call with its proper arguments that best answers the given prompt. Respond in the format {"name": function name, "parameters": dictionary of argument name and its value}. Do not use variables. {{ range $.Tools }} {{- . }} {{ end }} {{ .Content }}<|eot_id|> {{- else }} {{ .Content }}<|eot_id|> {{- end }}{{ if $last }}<|start_header_id|>assistant<|end_header_id|> {{ end }} {{- else if eq .Role "assistant" }}<|start_header_id|>assistant<|end_header_id|> {{- if .ToolCalls }} {{ range .ToolCalls }} {"name": "{{ .Function.Name }}", "parameters": {{ .Function.Arguments }}}{{ end }} {{- else }} {{ .Content }} {{- end }}{{ if not $last }}<|eot_id|>{{ end }} {{- else if eq .Role "tool" }}<|start_header_id|>ipython<|end_header_id|> {{ .Content }}<|eot_id|>{{ if $last }}<|start_header_id|>assistant<|end_header_id|> {{ end }} {{- end }} {{- end }}% ``` I've also been able to run the same cURL call: ``` ❯ curl -s localhost:11434/api/chat -d '{ "model":"llama3.2:3b", "messages":[ {"role":"user","content":"Search for install ollama models"} ], "tools":[{ "type":"function", "function":{ "name":"query_web_search", "description":"Perform a web search", "parameters":{ "type":"object", "properties":{ "query":{ "type":"string", "description":"The search query" } }, "required":["query"] } } }], "stream":false }' | jq { "model": "llama3.2:3b", "created_at": "2025-12-18T15:29:10.304027Z", "message": { "role": "assistant", "content": "", "tool_calls": [ { "id": "call_h8ikz9yn", "function": { "index": 0, "name": "query_web_search", "arguments": { "query": "install ollama models" } } } ] }, "done": true, "done_reason": "stop", "total_duration": 3659757167, "load_duration": 2690591833, "prompt_eval_count": 161, "prompt_eval_duration": 407400667, "eval_count": 22, "eval_duration": 548841582 } ```
GiteaMirror commented 2026-04-29 09:09:27 -05:00
Author
Owner
Copy Link

@sacenox commented on GitHub (Dec 18, 2025):

I'm going to close this issue and apologize for any time wasted. It was a bug with the frontend UI render logic... 🤦

Thank you for your time and help 🙇

<!-- gh-comment-id:3671013939 --> @sacenox commented on GitHub (Dec 18, 2025): I'm going to close this issue and apologize for any time wasted. It was a bug with the frontend UI render logic... 🤦 Thank you for your time and help 🙇
GiteaMirror commented 2026-04-29 09:09:28 -05:00
Author
Owner
Copy Link

@johnnymast commented on GitHub (Jan 9, 2026):

This is still happening to this day with llama3.2. My chatbot randomly spits out toolcalls that are not send to tool_calls but to content. This results in that she will talk json randomly. Now i am aware there might be issues with the model but i am not 100% sure of this because i am not techical enough to know under the hood.

<!-- gh-comment-id:3727552494 --> @johnnymast commented on GitHub (Jan 9, 2026): This is still happening to this day with llama3.2. My chatbot randomly spits out toolcalls that are not send to tool_calls but to content. This results in that she will talk json randomly. Now i am aware there might be issues with the model but i am not 100% sure of this because i am not techical enough to know under the hood.
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 bug
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#55420
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?