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 #14752] ollama launch codex - ⚠ Model metadata for gpt-oss:20b not found. Defaulting to fallback metadata; this can degrade performance and cause issues. #71598

New Issue
Closed
opened 2026-05-05 02:14:04 -05:00 by GiteaMirror · 11 comments
GiteaMirror commented 2026-05-05 02:14:04 -05:00
Owner
Copy Link

Originally created by @vt-alt on GitHub (Mar 10, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14752

What is the issue?

Tested on 0.17.7 and codex-cli 0.104.0.

  1. Frequent message ⚠ Model metadata for gpt-oss:20b not found. Defaulting to fallback metadata; this can degrade performance and cause issues.
  2. The requested task is frequently just stopped.
  3. The patch does not apply with errors like these:
• - Update File: path ..."]}. So we need to escape newlines. In the JSON, we need to encode newline as \n. The example from system used raw string with
    newlines? Actually the assistant earlier wrote apply_patch with plain text. Maybe the tool supports multiline string as is? In the earlier conversation,
    the assistant used apply_patch like:

  {"command":["apply_patch","*** Begin Patch\\n*** Update File: path/...\\n@@ ..."]}

  But not with escapes. The error "unsupported call: apply_patch" may be due to quoting issues. Let's try using triple quotes but escape newline? Let's attempt
  minimal patch.

  We'll patch the section in a simpler way: replace the lines. Provide patch as a single string with newline escapes.

  We'll craft patch string manually: "*** Begin Patch\n*** Update File: bzlib.c\n@@\n-   bzf->strm.bzalloc  = NULL;\n-   bzf->strm.bzfree   = NULL;\n-
  bzf->strm.opaque   = NULL;\n+   bzf->strm.bzalloc  = NULL;\n+   bzf->strm.bzfree   = NULL;\n+   /* Preserve small flag /\n+   bzf->strm.opaque   = (void)
  (intptr_t)small;\n*** End Patch"

  Let's try.

• Ran apply_patch
  └ /bin/bash: line 1: apply_patch: command not found

Then in went into the rabbit hole thinking why apply_patch tool does not work.

So basically ollama launch codex does not work for gpt-oss:20b. This might be codex-cli issue, but ollama launch codex is expected to configure it properly to work isn't it?

Repro

I request a simple change to modify bzip2:

$ git clone https://sourceware.org/git/bzip2.git
$ cd bzip2
$ ollama launch codex --model gpt-oss:20b
> Implement multiple concatenated streams support in `BZ2_bzread` (zlib-compatibility API).

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.17.7

Originally created by @vt-alt on GitHub (Mar 10, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14752 ### What is the issue? Tested on 0.17.7 and codex-cli 0.104.0. 1. Frequent message `⚠ Model metadata for `gpt-oss:20b` not found. Defaulting to fallback metadata; this can degrade performance and cause issues.` 2. The requested task is frequently just stopped. 3. The patch does not apply with errors like these: ``` • - Update File: path ..."]}. So we need to escape newlines. In the JSON, we need to encode newline as \n. The example from system used raw string with newlines? Actually the assistant earlier wrote apply_patch with plain text. Maybe the tool supports multiline string as is? In the earlier conversation, the assistant used apply_patch like: {"command":["apply_patch","*** Begin Patch\\n*** Update File: path/...\\n@@ ..."]} But not with escapes. The error "unsupported call: apply_patch" may be due to quoting issues. Let's try using triple quotes but escape newline? Let's attempt minimal patch. We'll patch the section in a simpler way: replace the lines. Provide patch as a single string with newline escapes. We'll craft patch string manually: "*** Begin Patch\n*** Update File: bzlib.c\n@@\n- bzf->strm.bzalloc = NULL;\n- bzf->strm.bzfree = NULL;\n- bzf->strm.opaque = NULL;\n+ bzf->strm.bzalloc = NULL;\n+ bzf->strm.bzfree = NULL;\n+ /* Preserve small flag /\n+ bzf->strm.opaque = (void) (intptr_t)small;\n*** End Patch" Let's try. ``` ``` • Ran apply_patch └ /bin/bash: line 1: apply_patch: command not found ``` Then in went into the rabbit hole thinking why `apply_patch` tool does not work. So basically `ollama launch codex` does not work for `gpt-oss:20b`. This might be `codex-cli` issue, but `ollama launch codex` is expected to configure it properly to work isn't it? ### Repro I request a simple change to modify bzip2: ```shell $ git clone https://sourceware.org/git/bzip2.git $ cd bzip2 $ ollama launch codex --model gpt-oss:20b > Implement multiple concatenated streams support in `BZ2_bzread` (zlib-compatibility API). ``` ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.17.7
GiteaMirror added the bug label 2026-05-05 02:14:04 -05:00
GiteaMirror commented 2026-05-05 02:14:06 -05:00
Author
Owner
Copy Link

@coderenewables commented on GitHub (Mar 13, 2026):

trying to run same with qwen3:4b model. same error!

<!-- gh-comment-id:4054021186 --> @coderenewables commented on GitHub (Mar 13, 2026): trying to run same with qwen3:4b model. same error!
GiteaMirror commented 2026-05-05 02:14:06 -05:00
Author
Owner
Copy Link

@Bombe commented on GitHub (Mar 15, 2026):

The initial issue with gpt-oss:20b also happens on macOS 15.7.4 with ollama 0.18.0 and codex-cli 0.114.0.

<!-- gh-comment-id:4061876590 --> @Bombe commented on GitHub (Mar 15, 2026): The initial issue with `gpt-oss:20b` also happens on macOS 15.7.4 with ollama 0.18.0 and codex-cli 0.114.0.
GiteaMirror commented 2026-05-05 02:14:06 -05:00
Author
Owner
Copy Link

@loonery commented on GitHub (Mar 15, 2026):

Same as other users, I am experiencing this problem with Ollama 0.17.7, codex cli 0.114.0 on NixOS. Changing model does not work, even if the model name is unhyphenated, which I thought might have been the problem due to the behavior described in issue: https://github.com/openai/codex/issues/14276

Everything was working before upgrade from codex 0.92, so this may be a codex issue.

I opened an issue with codex cli: https://github.com/openai/codex/issues/14757

<!-- gh-comment-id:4063582171 --> @loonery commented on GitHub (Mar 15, 2026): Same as other users, I am experiencing this problem with Ollama 0.17.7, codex cli 0.114.0 on NixOS. Changing model does not work, even if the model name is unhyphenated, which I thought might have been the problem due to the behavior described in issue: https://github.com/openai/codex/issues/14276 Everything was working before upgrade from codex 0.92, so this may be a codex issue. I opened an issue with codex cli: https://github.com/openai/codex/issues/14757
GiteaMirror commented 2026-05-05 02:14:07 -05:00
Author
Owner
Copy Link

@ronaldpetty commented on GitHub (Mar 24, 2026):

I have same issue with reverse. Codex using Ollama. I am guessing tool launching order doesn't mater here. On M5 / Tahoe. Just sharing in off chance it helps.

<!-- gh-comment-id:4114506490 --> @ronaldpetty commented on GitHub (Mar 24, 2026): I have same issue with reverse. Codex using Ollama. I am guessing tool launching order doesn't mater here. On M5 / Tahoe. Just sharing in off chance it helps.
GiteaMirror commented 2026-05-05 02:14:07 -05:00
Author
Owner
Copy Link

@loonery commented on GitHub (Mar 24, 2026):

@ronaldpetty If you look at my issue that I opened on Codex's repo, you can see that I found a workaround. I think this is a regression that was introduced for custom models in Codex and there does not seem to be a huge appetite to fix it.

The key is to make sure that in your config you set a 'model_catalog_json' path to a local 'models.json' file. That file should look like https://github.com/openai/codex/blob/main/codex-rs/core/models.json does, but you should modify the file to include your the model names your looking for. The 'slug' and 'display_name' keys should match your specified model name. For me, that meant replacing all 'gpt-oss-20b' instances with 'gpt-oss:20b'.

This is at least, what I did for using gpt-oss:20b. I don't know if this will work for other custom models.

<!-- gh-comment-id:4114608354 --> @loonery commented on GitHub (Mar 24, 2026): @ronaldpetty If you look at my issue that I opened on Codex's repo, you can see that I found a workaround. I think this is a regression that was introduced for custom models in Codex and there does not seem to be a huge appetite to fix it. The key is to make sure that in your config you set a 'model_catalog_json' path to a local 'models.json' file. That file should look like https://github.com/openai/codex/blob/main/codex-rs/core/models.json does, but you should modify the file to include your the model names your looking for. The 'slug' and 'display_name' keys should match your specified model name. For me, that meant replacing all 'gpt-oss-20b' instances with 'gpt-oss:20b'. This is at least, what I did for using gpt-oss:20b. I don't know if this will work for other custom models.
GiteaMirror commented 2026-05-05 02:14:07 -05:00
Author
Owner
Copy Link

@ronaldpetty commented on GitHub (Mar 24, 2026):

Thanks @loonery , I'll give it a whirl. My initial checks were leading that direction, thanks for clearing a path!

<!-- gh-comment-id:4114659505 --> @ronaldpetty commented on GitHub (Mar 24, 2026): Thanks @loonery , I'll give it a whirl. My initial checks were leading that direction, thanks for clearing a path!
GiteaMirror commented 2026-05-05 02:14:08 -05:00
Author
Owner
Copy Link

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

I believe this should be gone with some of our recent releases.

<!-- gh-comment-id:4255535257 --> @ParthSareen commented on GitHub (Apr 15, 2026): I believe this should be gone with some of our recent releases.
GiteaMirror commented 2026-05-05 02:14:08 -05:00
Author
Owner
Copy Link

@savanthongvanh commented on GitHub (Apr 16, 2026):

i was still getting this with 121 btw. i'm expecting it to figure out how to use curl or something and go get the weather

Image
<!-- gh-comment-id:4257349294 --> @savanthongvanh commented on GitHub (Apr 16, 2026): i was still getting this with 121 btw. i'm expecting it to figure out how to use curl or something and go get the weather <img width="2142" height="990" alt="Image" src="https://github.com/user-attachments/assets/f1598997-222f-4a8f-9581-ca86617cabb9" />
GiteaMirror commented 2026-05-05 02:14:08 -05:00
Author
Owner
Copy Link

@kazaff commented on GitHub (Apr 21, 2026):

Image

v122 is still having this issue.

<!-- gh-comment-id:4285437133 --> @kazaff commented on GitHub (Apr 21, 2026): <img width="714" height="371" alt="Image" src="https://github.com/user-attachments/assets/f8e6a062-6572-41da-867e-4019d9d457d9" /> v122 is still having this issue.
GiteaMirror commented 2026-05-05 02:14:08 -05:00
Author
Owner
Copy Link

@namitha393 commented on GitHub (Apr 26, 2026):

v125 has this issue

Image
<!-- gh-comment-id:4322992712 --> @namitha393 commented on GitHub (Apr 26, 2026): v125 has this issue <img width="626" height="289" alt="Image" src="https://github.com/user-attachments/assets/8efd4275-ea20-470d-87c8-01ef0c56d710" />
GiteaMirror commented 2026-05-05 02:14:09 -05:00
Author
Owner
Copy Link

@vt-alt commented on GitHub (Apr 27, 2026):

I believe this should be gone with some of our recent releases.

@ParthSareen This is not resolved. Latest ollama 0.21.2 + latest OpenAI Codex (v0.125.0):

$ ollama launch codex
╭──────────────────────────────────────────────────╮
│ >_ OpenAI Codex (v0.125.0)                       │
│                                                  │
│ model:     gpt-oss:20b medium   /model to change │
│ directory: ~/src/ollama                          │
╰──────────────────────────────────────────────────╯

  Tip: New Build faster with Codex.

› hi

⚠ Model metadata for `gpt-oss:20b` not found. Defaulting to fallback metadata; this can degrade performance and cause issues.

• Hello! How can I help you today?
Token usage: total=7,744 input=7,710 output=34

Can you reopen the issue, please?

<!-- gh-comment-id:4325065101 --> @vt-alt commented on GitHub (Apr 27, 2026): > I believe this should be gone with some of our recent releases. @ParthSareen This is **not** resolved. Latest ollama 0.21.2 + latest OpenAI Codex (v0.125.0): ``` $ ollama launch codex ╭──────────────────────────────────────────────────╮ │ >_ OpenAI Codex (v0.125.0) │ │ │ │ model: gpt-oss:20b medium /model to change │ │ directory: ~/src/ollama │ ╰──────────────────────────────────────────────────╯ Tip: New Build faster with Codex. › hi ⚠ Model metadata for `gpt-oss:20b` not found. Defaulting to fallback metadata; this can degrade performance and cause issues. • Hello! How can I help you today? Token usage: total=7,744 input=7,710 output=34 ``` Can you reopen the issue, please?
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#71598
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?