github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-08 17:49:24 -05:00
Code Issues 14.6k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #9092] Enabling tools breaks stream=True on /v1 endpoint, only returns a single complete response #5916

New Issue
Open
opened 2026-04-12 17:15:15 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-12 17:15:15 -05:00
Owner
Copy Link

Originally created by @huangdihd on GitHub (Feb 14, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9092

What is the issue?

Issue Description

When enabling tools, the Ollama API seems to break streaming (stream=True) on the /v1 endpoint. Instead of returning chunks of data progressively, it waits and sends the entire response as a single block.

Steps to Reproduce

  1. Run the following Python script using the OpenAI-compatible API (via the /v1 endpoint):

    import openai

client = openai.OpenAI(base_url="http://localhost:11434/v1", api_key="ollama")

response = client.chat.completions.create(
    model="qwen2.5:32b",
    messages=[{"role": "user", "content": "Tell me a story"}],
    stream=True,
    tools=[{"type": "function", "function": {"name": "test_tool", "description": "Test tool"}}]
)

for chunk in response:
    print(chunk)

  2. Expected behavior:

    • Without tools, stream=True correctly streams chunks of choices[0].delta.content.
    • With tools enabled, the response is only returned as a single complete block.

Expected Behavior

Even when using tools, Ollama should support proper streaming behavior and return incremental chunks instead of buffering the full response.

Environment

  • Ollama Version: (0.5.10)
  • Operating System: Linux / Mac / Windows
  • API Method: OpenAI-compatible API (/v1 endpoint) / requests / Ollama Python SDK
  • Model Tested: qwen2.5:32b

Additional Information

When testing with curl against the /v1 endpoint, streaming works correctly:

curl -N -X POST http://localhost:11434/v1/chat/completions \
     -H "Content-Type: application/json" \
     -d '{
        "model": "qwen2.5:32b",
        "messages": [{"role": "user", "content": "Tell me a story"}],
        "stream": true
     }'

However, when using Python and enabling tools, streaming does not work as expected.

Question

  • Is this an intentional limitation when using tools with /v1 endpoints?
  • If so, is there a workaround to allow streaming while still using tools?

Thanks for your help!

Relevant log output

curl -N -X POST http://localhost:11434/v1/chat/completions \
     -H "Content-Type: application/json" \
     -d '{
        "model": "qwen2.5:32b",
        "messages": [{"role": "user", "content": "Tell me a story"}],
        "stream": true
     }'
...
data: {"id":"chatcmpl-634","object":"chat.completion.chunk","created":1739506822,"model":"qwen2.5:32b","system_fingerprint":"fp_ollama","choices":[{"index":0,"delta":{"role":"assistant","content":"."},"finish_reason":null}]}

data: {"id":"chatcmpl-634","object":"chat.completion.chunk","created":1739506823,"model":"qwen2.5:32b","system_fingerprint":"fp_ollama","choices":[{"index":0,"delta":{"role":"assistant","content":""},"finish_reason":"stop"}]}

data: [DONE]

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.5.10

Originally created by @huangdihd on GitHub (Feb 14, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9092 ### What is the issue? ### **Issue Description** When enabling `tools`, the Ollama API seems to break streaming (`stream=True`) on the `/v1` endpoint. Instead of returning chunks of data progressively, it waits and sends the entire response as a single block. ### **Steps to Reproduce** 1. Run the following Python script using the OpenAI-compatible API (via the `/v1` endpoint): ```python import openai client = openai.OpenAI(base_url="http://localhost:11434/v1", api_key="ollama") response = client.chat.completions.create( model="qwen2.5:32b", messages=[{"role": "user", "content": "Tell me a story"}], stream=True, tools=[{"type": "function", "function": {"name": "test_tool", "description": "Test tool"}}] ) for chunk in response: print(chunk) ``` 2. Expected behavior: * **Without `tools`**, `stream=True` correctly streams chunks of `choices[0].delta.content`. * **With `tools` enabled**, the response is only returned as a single complete block. ### **Expected Behavior** Even when using `tools`, Ollama should support proper streaming behavior and return incremental chunks instead of buffering the full response. ### **Environment** * **Ollama Version:** (`0.5.10`) * **Operating System:** Linux / Mac / Windows * **API Method:** OpenAI-compatible API (`/v1` endpoint) / `requests` / Ollama Python SDK * **Model Tested:** qwen2.5:32b ### **Additional Information** When testing with `curl` against the `/v1` endpoint, streaming works correctly: ```sh curl -N -X POST http://localhost:11434/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{ "model": "qwen2.5:32b", "messages": [{"role": "user", "content": "Tell me a story"}], "stream": true }' ``` However, when using Python and enabling `tools`, streaming does not work as expected. ### **Question** * Is this an intentional limitation when using `tools` with `/v1` endpoints? * If so, is there a workaround to allow streaming while still using `tools`? Thanks for your help! ### Relevant log output ```shell curl -N -X POST http://localhost:11434/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{ "model": "qwen2.5:32b", "messages": [{"role": "user", "content": "Tell me a story"}], "stream": true }' ... data: {"id":"chatcmpl-634","object":"chat.completion.chunk","created":1739506822,"model":"qwen2.5:32b","system_fingerprint":"fp_ollama","choices":[{"index":0,"delta":{"role":"assistant","content":"."},"finish_reason":null}]} data: {"id":"chatcmpl-634","object":"chat.completion.chunk","created":1739506823,"model":"qwen2.5:32b","system_fingerprint":"fp_ollama","choices":[{"index":0,"delta":{"role":"assistant","content":""},"finish_reason":"stop"}]} data: [DONE] ``` ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.5.10
GiteaMirror added the bug label 2026-04-12 17:15:15 -05:00
GiteaMirror commented 2026-04-12 17:15:17 -05:00
Author
Owner
Copy Link

@LeisureLinux commented on GitHub (Feb 14, 2025):

https://github.com/ollama/ollama/issues/8517

<!-- gh-comment-id:2659453944 --> @LeisureLinux commented on GitHub (Feb 14, 2025): https://github.com/ollama/ollama/issues/8517
GiteaMirror commented 2026-04-12 17:15:18 -05:00
Author
Owner
Copy Link

@nonsleepr commented on GitHub (Feb 28, 2025):

Have similar issue hitting /v1/chat/completions with that payload:

{
  "model": "llama3.2:latest",
  "stream": true,
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant."
    },
    {
      "role": "tool",
      "content": "Write a short poem."
    }
  ],
  "tools": [{}]
}

It doesn't stream if the role is "tool".

<!-- gh-comment-id:2691589724 --> @nonsleepr commented on GitHub (Feb 28, 2025): Have similar issue hitting `/v1/chat/completions` with that payload: ```json { "model": "llama3.2:latest", "stream": true, "messages": [ { "role": "system", "content": "You are a helpful assistant." }, { "role": "tool", "content": "Write a short poem." } ], "tools": [{}] } ``` It doesn't stream if the role is "tool".
GiteaMirror commented 2026-04-12 17:15:19 -05:00
Author
Owner
Copy Link

@huangdihd commented on GitHub (Mar 8, 2025):

Have similar issue hitting /v1/chat/completions with that payload:

{
  "model": "llama3.2:latest",
  "stream": true,
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant."
    },
    {
      "role": "tool",
      "content": "Write a short poem."
    }
  ],
  "tools": [{}]
}

It doesn't stream if the role is "tool".

Role "tool" means this message is a response to a tool_call.
It requires parameters named "tool_call_id" and "name".

<!-- gh-comment-id:2708007327 --> @huangdihd commented on GitHub (Mar 8, 2025): > Have similar issue hitting `/v1/chat/completions` with that payload: > ```json > { > "model": "llama3.2:latest", > "stream": true, > "messages": [ > { > "role": "system", > "content": "You are a helpful assistant." > }, > { > "role": "tool", > "content": "Write a short poem." > } > ], > "tools": [{}] > } > ``` > > It doesn't stream if the role is "tool". Role "tool" means this message is a response to a tool_call. It requires parameters named "tool_call_id" and "name".
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#5916
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?