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 #14249] x/z-image-turbo FP8 model fails to load on v0.16.0/v0.16.1 - MLX runner regression #9276

New Issue
Closed
opened 2026-04-12 22:08:45 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-12 22:08:45 -05:00
Owner
Copy Link

Originally created by @NerdSnipe on GitHub (Feb 14, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14249

Description

The x/z-image-turbo (FP8) model fails to load on Ollama v0.16.0 and v0.16.1 with an MLX runner error. The same model works correctly
on v0.15.6 with no changes to the model or system.

Error

Error: failed to load model: 500 Internal Server Error: mlx runner failed: model.norm.weight (exit: exit status 1)

Server logs show the text encoder fails to map weight tensors - every layer (0-35) reports tensor
"model.layers.X.self_attn.q_proj.weight" not found, despite all 2,211 blob files being present and intact on disk.

Full error from server log:
Error: failed to create server: failed to load image model: failed to load zimage model: text encoder: load module: LoadModule:
missing weights:
model.embed_tokens.weight
model.layers.0.self_attn.q_proj: failed to load quantized weight model.layers.0.self_attn.q_proj: tensor
"model.layers.0.self_attn.q_proj.weight" not found
... (all layers 0-35, all projections)
model.norm.weight

Steps to Reproduce

  1. Install Ollama v0.16.1 on macOS (Apple Silicon)
  2. ollama pull x/z-image-turbo
  3. ollama run x/z-image-turbo "a red cat"
  4. Observe 500 error

Workaround

Downgrading to Ollama v0.15.6 resolves the issue. The model loads and generates images without any changes.

Environment

  • Ollama version (broken): v0.16.0 and v0.16.1
  • Ollama version (working): v0.15.6
  • OS: macOS (Darwin 25.2.0)
  • Hardware: Apple M3 Max, 36 GB RAM
  • Model: x/z-image-turbo:latest (FP8, 10.3B params, ~12GB)

Analysis

The v0.16.0 release notes mention "Improvements to Ollama's MLX runner to support GLM-4.7-Flash." This MLX runner change appears to
have broken the weight tensor name mapping for Z-Image pipeline models. The tokenizer loads successfully, but the text encoder
module cannot locate any weight tensors despite the blobs existing on disk.

Originally created by @NerdSnipe on GitHub (Feb 14, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14249 Description The x/z-image-turbo (FP8) model fails to load on Ollama v0.16.0 and v0.16.1 with an MLX runner error. The same model works correctly on v0.15.6 with no changes to the model or system. Error Error: failed to load model: 500 Internal Server Error: mlx runner failed: model.norm.weight (exit: exit status 1) Server logs show the text encoder fails to map weight tensors - every layer (0-35) reports tensor "model.layers.X.self_attn.q_proj.weight" not found, despite all 2,211 blob files being present and intact on disk. Full error from server log: Error: failed to create server: failed to load image model: failed to load zimage model: text encoder: load module: LoadModule: missing weights: model.embed_tokens.weight model.layers.0.self_attn.q_proj: failed to load quantized weight model.layers.0.self_attn.q_proj: tensor "model.layers.0.self_attn.q_proj.weight" not found ... (all layers 0-35, all projections) model.norm.weight Steps to Reproduce 1. Install Ollama v0.16.1 on macOS (Apple Silicon) 2. ollama pull x/z-image-turbo 3. ollama run x/z-image-turbo "a red cat" 4. Observe 500 error Workaround Downgrading to Ollama v0.15.6 resolves the issue. The model loads and generates images without any changes. Environment - Ollama version (broken): v0.16.0 and v0.16.1 - Ollama version (working): v0.15.6 - OS: macOS (Darwin 25.2.0) - Hardware: Apple M3 Max, 36 GB RAM - Model: x/z-image-turbo:latest (FP8, 10.3B params, ~12GB) Analysis The v0.16.0 release notes mention "Improvements to Ollama's MLX runner to support GLM-4.7-Flash." This MLX runner change appears to have broken the weight tensor name mapping for Z-Image pipeline models. The tokenizer loads successfully, but the text encoder module cannot locate any weight tensors despite the blobs existing on disk.
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
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#9276
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?