github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-16 05:05:37 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #15635] /v1/responses endpoint ignores reasoning_effort: "none" — thinking not disabled for gemma4:e2b #56486

New Issue
Open
opened 2026-04-29 10:53:47 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-29 10:53:47 -05:00
Owner
Copy Link

Originally created by @waywardgeek on GitHub (Apr 16, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15635

What is the issue?

Description
The /v1/chat/completions endpoint correctly disables thinking when reasoning_effort: "none" is set, but the /v1/responses endpoint ignores it entirely. The model still produces full thinking output, adding significant latency.

Environment
Ollama: HEAD-2bb7ea0 (also reproduced on 0.20.7 stable)
Model: gemma4:e2b
OS: macOS (Apple Silicon)
Reproduction
Chat Completions — works correctly (0.5s, no thinking):

curl -s http://localhost:11434/v1/chat/completions
-H "Content-Type: application/json"
-d '{
"model": "gemma4:e2b",
"messages": [{"role": "user", "content": "Reply with ONLY: [1,3]"}],
"reasoning_effort": "none",
"stream": false
}'
Result: 6 tokens, no reasoning field, ~0.5s

Responses API — thinking NOT disabled (4.7s, full reasoning):

curl -s http://localhost:11434/v1/responses
-H "Content-Type: application/json"
-d '{
"model": "gemma4:e2b",
"input": "Reply with ONLY: [1,3]",
"reasoning_effort": "none",
"stream": false
}'
Result: 200+ tokens, full reasoning output block with thinking content, ~4.7s

Also tried "reasoning": {"effort": "none"} on the Responses endpoint — same result.

Expected Behavior
Both endpoints should honor reasoning_effort: "none" and suppress thinking output. The Ollama docs list "Reasoning/thinking control (for thinking models)" as a supported feature for Chat Completions, and list reasoning_effort in the Responses API supported fields.

Impact
This is a significant latency issue for applications using thinking models as lightweight classifiers via the Responses API. In our case, we use gemma4:e2b as a relevance judge for memory recall — it should return a JSON array in ~0.5s but takes ~5s due to unnecessary thinking.

Relevant log output

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

Ollama: HEAD-2bb7ea0 (also reproduced on 0.20.7 stable)

Originally created by @waywardgeek on GitHub (Apr 16, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15635 ### What is the issue? Description The /v1/chat/completions endpoint correctly disables thinking when reasoning_effort: "none" is set, but the /v1/responses endpoint ignores it entirely. The model still produces full thinking output, adding significant latency. Environment Ollama: HEAD-2bb7ea0 (also reproduced on 0.20.7 stable) Model: gemma4:e2b OS: macOS (Apple Silicon) Reproduction Chat Completions — works correctly (0.5s, no thinking): curl -s http://localhost:11434/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{ "model": "gemma4:e2b", "messages": [{"role": "user", "content": "Reply with ONLY: [1,3]"}], "reasoning_effort": "none", "stream": false }' Result: 6 tokens, no reasoning field, ~0.5s Responses API — thinking NOT disabled (4.7s, full reasoning): curl -s http://localhost:11434/v1/responses \ -H "Content-Type: application/json" \ -d '{ "model": "gemma4:e2b", "input": "Reply with ONLY: [1,3]", "reasoning_effort": "none", "stream": false }' Result: 200+ tokens, full reasoning output block with thinking content, ~4.7s Also tried "reasoning": {"effort": "none"} on the Responses endpoint — same result. Expected Behavior Both endpoints should honor reasoning_effort: "none" and suppress thinking output. The [Ollama docs](https://docs.ollama.com/api/openai-compatibility) list "Reasoning/thinking control (for thinking models)" as a supported feature for Chat Completions, and list reasoning_effort in the Responses API supported fields. Impact This is a significant latency issue for applications using thinking models as lightweight classifiers via the Responses API. In our case, we use gemma4:e2b as a relevance judge for memory recall — it should return a JSON array in ~0.5s but takes ~5s due to unnecessary thinking. ### Relevant log output ```shell ``` ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version Ollama: HEAD-2bb7ea0 (also reproduced on 0.20.7 stable)
GiteaMirror added the bug label 2026-04-29 10:53:47 -05:00
GiteaMirror commented 2026-04-29 10:53:51 -05:00
Author
Owner
Copy Link

@balgaly commented on GitHub (Apr 18, 2026):

I've opened a PR for this -- https://github.com/ollama/ollama/pull/15664

<!-- gh-comment-id:4273353501 --> @balgaly commented on GitHub (Apr 18, 2026): I've opened a PR for this -- https://github.com/ollama/ollama/pull/15664
GiteaMirror commented 2026-04-29 10:53:52 -05:00
Author
Owner
Copy Link

@PureBlissAK commented on GitHub (Apr 18, 2026):

🤖 Automated Triage & Analysis Report

Issue: #15635
Analyzed: 2026-04-18T18:19:19.289770

Analysis

  • Type: unknown
  • Severity: medium
  • Components: unknown

Implementation Plan

  • Effort: medium
  • Steps:

This issue has been triaged and marked for implementation.

<!-- gh-comment-id:4274304676 --> @PureBlissAK commented on GitHub (Apr 18, 2026): <!-- ollama-issue-orchestrator:v1 issue:15635 --> ## 🤖 Automated Triage & Analysis Report **Issue**: #15635 **Analyzed**: 2026-04-18T18:19:19.289770 ### Analysis - **Type**: unknown - **Severity**: medium - **Components**: unknown ### Implementation Plan - **Effort**: medium - **Steps**: *This issue has been triaged and marked for implementation.*
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#56486
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?