github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 08:02:14 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #12174] tool_calls missing from qwen2.5-coder v0.11.7 (also v0.11.8) #70156

New Issue
Open
opened 2026-05-04 20:33:18 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-05-04 20:33:18 -05:00
Owner
Copy Link

Originally created by @ldab on GitHub (Sep 3, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/12174

What is the issue?

When requesting ie.

{
    "model": "qwen2.5-coder",
    "stream": false,
    "temperature": 0.2,
    "seed": -1183861426,
    "messages": [
        {
            "role": "system",
            "content": "Act as a PatchSet Reviewer."
        },
        {
            "role": "user",
            "content": "Why not use malloc and not free data?"
        }
    ],
    "tools": [
        {
            "type": "function",
            "function": {
                "name": "format_replies",
                "description": "Provide the PatchSet review replies.",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "replies": {
                            "type": "array",
                            "items": {
                                "type": "object",
                                "properties": {
                                    "id": {
                                        "type": "integer"
                                    },
                                    "reply": {
                                        "type": "string"
                                    },
                                    "score": {
                                        "type": "integer"
                                    },
                                    "relevance": {
                                        "type": "number"
                                    },
                                    "repeated": {
                                        "type": "boolean"
                                    },
                                    "conflicting": {
                                        "type": "boolean"
                                    },
                                    "filename": {
                                        "type": "string"
                                    },
                                    "lineNumber": {
                                        "type": "integer"
                                    },
                                    "codeSnippet": {
                                        "type": "string"
                                    }
                                },
                                "required": [
                                    "reply"
                                ]
                            }
                        },
                        "changeId": {
                            "type": "string"
                        }
                    },
                    "required": [
                        "replies",
                        "changeId"
                    ]
                }
            }
        }
    ],
    "tool_choice": {
        "type": "function",
        "function": {
            "name": "format_replies"
        }
    }
}

The response does not contains "tool_calls"

{
    "id": "chatcmpl-9",
    "object": "chat.completion",
    "created": 1756934212,
    "model": "qwen2.5-coder",
    "system_fingerprint": "fp_ollama",
    "choices": [
        {
            "index": 0,
            "message": {
                "role": "assistant",
                "content": "{\"name\": \"format_replies\", \"arguments\": {\"changeId\": \"I1234567890abcdef1234567890abcdef1234567\", \"replies\": [{\"filename\": \"example.c\", \"lineNumber\": 10, \"reply\": \"Using malloc without freeing the allocated memory can lead to memory leaks. It's important to ensure that every call to malloc is paired with a corresponding call to free.\", \"score\": 2}]}}"
            },
            "finish_reason": "stop"
        }
    ],
    "usage": {
        "prompt_tokens": 252,
        "completion_tokens": 110,
        "total_tokens": 362
    }
}

Whereas when using qwen2.5 it seems fine, did I miss something?

Relevant log output

OS

Debian 12

GPU

No response

CPU

No response

Ollama version

v0.11.7

Originally created by @ldab on GitHub (Sep 3, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/12174 ### What is the issue? When requesting ie. ``` { "model": "qwen2.5-coder", "stream": false, "temperature": 0.2, "seed": -1183861426, "messages": [ { "role": "system", "content": "Act as a PatchSet Reviewer." }, { "role": "user", "content": "Why not use malloc and not free data?" } ], "tools": [ { "type": "function", "function": { "name": "format_replies", "description": "Provide the PatchSet review replies.", "parameters": { "type": "object", "properties": { "replies": { "type": "array", "items": { "type": "object", "properties": { "id": { "type": "integer" }, "reply": { "type": "string" }, "score": { "type": "integer" }, "relevance": { "type": "number" }, "repeated": { "type": "boolean" }, "conflicting": { "type": "boolean" }, "filename": { "type": "string" }, "lineNumber": { "type": "integer" }, "codeSnippet": { "type": "string" } }, "required": [ "reply" ] } }, "changeId": { "type": "string" } }, "required": [ "replies", "changeId" ] } } } ], "tool_choice": { "type": "function", "function": { "name": "format_replies" } } } ``` The response does not contains "tool_calls" ``` { "id": "chatcmpl-9", "object": "chat.completion", "created": 1756934212, "model": "qwen2.5-coder", "system_fingerprint": "fp_ollama", "choices": [ { "index": 0, "message": { "role": "assistant", "content": "{\"name\": \"format_replies\", \"arguments\": {\"changeId\": \"I1234567890abcdef1234567890abcdef1234567\", \"replies\": [{\"filename\": \"example.c\", \"lineNumber\": 10, \"reply\": \"Using malloc without freeing the allocated memory can lead to memory leaks. It's important to ensure that every call to malloc is paired with a corresponding call to free.\", \"score\": 2}]}}" }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": 252, "completion_tokens": 110, "total_tokens": 362 } } ``` Whereas when using `qwen2.5` it seems fine, did I miss something? ### Relevant log output ```shell ``` ### OS Debian 12 ### GPU _No response_ ### CPU _No response_ ### Ollama version v0.11.7
GiteaMirror added the bug label 2026-05-04 20:33:19 -05:00
GiteaMirror commented 2026-05-04 20:33:22 -05:00
Author
Owner
Copy Link

@neurostream commented on GitHub (Sep 4, 2025):

@ldab Did previous releases of qwen2.5-coder support tool use - and this issue ( #12174 ) is calling out that something changed?

Interesting that the coder variation of the qwen3 models was split off from the rest on ollama.com without tool use as well ( #11621 ).

<!-- gh-comment-id:3251983289 --> @neurostream commented on GitHub (Sep 4, 2025): @ldab Did previous releases of qwen2.5-coder support tool use - and this issue ( #12174 ) is calling out that something changed? Interesting that the coder variation of the qwen3 models was split off from the rest on ollama.com without tool use as well ( #11621 ).
GiteaMirror commented 2026-05-04 20:33:23 -05:00
Author
Owner
Copy Link

@ldab commented on GitHub (Sep 4, 2025):

This is my first usage of this model, I don’t know if previous releases included the field, I can see on the template tho :/

I think I also saw some differences between qwen2.5 7b and 14b… if I remember right the 7b also missed tool_calls

<!-- gh-comment-id:3252019322 --> @ldab commented on GitHub (Sep 4, 2025): This is my first usage of this model, I don’t know if previous releases included the field, I can see on the template tho :/ I think I also saw some differences between qwen2.5 7b and 14b… if I remember right the 7b also missed tool_calls
GiteaMirror commented 2026-05-04 20:33:25 -05:00
Author
Owner
Copy Link

@ninthhousestudios commented on GitHub (Apr 15, 2026):

Reproducing this on ollama 0.20.7 with qwen2.5-coder:14b — same failure: model emits bare JSON in content, tool_calls is null.

Additional data points that narrow the cause:

  1. The same pipeline works for qwen3:8b — confirms the OAI shim and tool routing are fine. Only the coder variant fails.

    # qwen3:8b  →  OK
tool_calls: [{'function': {'name': 'calculator', 'arguments': '{"expr":"17 * 23"}'}}]

# qwen2.5-coder:14b  →  broken
tool_calls: None
content: {"name": "calculator", "arguments": {"expr": "17 * 23"}}

  2. Native /api/chat endpoint fails identically — not OAI-shim-specific.

    curl http://127.0.0.1:11434/api/chat -d '{"model":"qwen2.5-coder:14b","tools":[...],"messages":[...]}'
# tool_calls: null, content: bare JSON

  3. The template explicitly instructs the model to use <tool_call> tags:

    For each function call, return a json object with function name and arguments within <tool_call></tool_call> with NO other text.

    But the model consistently outputs bare JSON without the tags, so the parser can't recognize it as a tool call.

Likely root cause: the template shipped with qwen2.5-coder assumes the same tool-call format as base qwen2.5, but the coder variant appears to have been trained on a different (or weaker) tool-call distribution and doesn't reliably produce the expected <tool_call> wrappers. Either the template needs adjustment to match the model's actual output, or the tools capability flag should be demoted / removed for this model so downstream frameworks (Goose, opencode, etc.) don't assume tool support that won't work reliably.

Related: #13968 (closed) reports the same symptom on qwen2.5:14b base.

Happy to test any proposed fix on the 14B coder variant.

<!-- gh-comment-id:4254002185 --> @ninthhousestudios commented on GitHub (Apr 15, 2026): Reproducing this on **ollama 0.20.7** with `qwen2.5-coder:14b` — same failure: model emits bare JSON in `content`, `tool_calls` is `null`. **Additional data points that narrow the cause:** 1. **The same pipeline works for `qwen3:8b`** — confirms the OAI shim and tool routing are fine. Only the coder variant fails. ``` # qwen3:8b → OK tool_calls: [{'function': {'name': 'calculator', 'arguments': '{"expr":"17 * 23"}'}}] # qwen2.5-coder:14b → broken tool_calls: None content: {"name": "calculator", "arguments": {"expr": "17 * 23"}} ``` 2. **Native `/api/chat` endpoint fails identically** — not OAI-shim-specific. ```bash curl http://127.0.0.1:11434/api/chat -d '{"model":"qwen2.5-coder:14b","tools":[...],"messages":[...]}' # tool_calls: null, content: bare JSON ``` 3. **The template explicitly instructs the model to use `<tool_call>` tags:** > For each function call, return a json object with function name and arguments within `<tool_call></tool_call>` with NO other text. But the model consistently outputs bare JSON without the tags, so the parser can't recognize it as a tool call. **Likely root cause:** the template shipped with `qwen2.5-coder` assumes the same tool-call format as base `qwen2.5`, but the coder variant appears to have been trained on a different (or weaker) tool-call distribution and doesn't reliably produce the expected `<tool_call>` wrappers. Either the template needs adjustment to match the model's actual output, or the `tools` capability flag should be demoted / removed for this model so downstream frameworks (Goose, opencode, etc.) don't assume tool support that won't work reliably. Related: #13968 (closed) reports the same symptom on `qwen2.5:14b` base. Happy to test any proposed fix on the 14B coder variant.
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#70156
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?