[GH-ISSUE #13391] Model Request GLM-4.6V #8841

Open
opened 2026-04-12 21:38:00 -05:00 by GiteaMirror · 11 comments
Owner

Originally created by @snapo on GitHub (Dec 9, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/13391

z.ai released 2 new models that would run very good on consumer HW.

eval settings where done with :
top_p: 0.6
top_k: 2 <------ strange but it works amazingly good
temperature: 0.8
repetition_penalty: 1.1
max_generate_tokens: 16K

and it has tools and visual capability's ....

Would love to see them added to ollama as llama.cpp already supports them. Especially the 4.6V-Fast with only 9B active params (dense model) would be absolutely amazing....

Originally created by @snapo on GitHub (Dec 9, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/13391 z.ai released 2 new models that would run very good on consumer HW. eval settings where done with : top_p: 0.6 top_k: 2 <------ strange but it works amazingly good temperature: 0.8 repetition_penalty: 1.1 max_generate_tokens: 16K and it has tools and visual capability's .... Would love to see them added to ollama as llama.cpp already supports them. Especially the 4.6V-Fast with only 9B active params (dense model) would be absolutely amazing....
GiteaMirror added the model label 2026-04-12 21:38:00 -05:00
Author
Owner

@yumo666666 commented on GitHub (Dec 9, 2025):

加一

<!-- gh-comment-id:3631504486 --> @yumo666666 commented on GitHub (Dec 9, 2025): 加一
Author
Owner

@rick-github commented on GitHub (Dec 9, 2025):

https://github.com/ggml-org/llama.cpp/pull/16600

<!-- gh-comment-id:3631531347 --> @rick-github commented on GitHub (Dec 9, 2025): https://github.com/ggml-org/llama.cpp/pull/16600
Author
Owner

@kir3d commented on GitHub (Dec 9, 2025):

I tried ollama run hf.co/unsloth/GLM-4.6V-Flash-GGUF:Q8_K_XL, but ollama reports an error:
This model does not support images.

:(

<!-- gh-comment-id:3634424050 --> @kir3d commented on GitHub (Dec 9, 2025): I tried `ollama run hf.co/unsloth/GLM-4.6V-Flash-GGUF:Q8_K_XL`, but ollama reports an error: This model does not support images. :(
Author
Owner

@rick-github commented on GitHub (Dec 9, 2025):

The HuggigFace model only contains the text weights because support for vision is still in progress: https://github.com/ggml-org/llama.cpp/pull/16600

<!-- gh-comment-id:3634510945 --> @rick-github commented on GitHub (Dec 9, 2025): The HuggigFace model only contains the text weights because support for vision is still in progress: https://github.com/ggml-org/llama.cpp/pull/16600
Author
Owner

@Alias4D commented on GitHub (Dec 16, 2025):

Please Support GLM4.6V-Fast 9B by ollama 🙏🙏🙏🙏🙏🙏🙏🙏

<!-- gh-comment-id:3661097501 --> @Alias4D commented on GitHub (Dec 16, 2025): Please Support GLM4.6V-Fast 9B by ollama 🙏🙏🙏🙏🙏🙏🙏🙏
Author
Owner

@SurealCereal commented on GitHub (Dec 20, 2025):

It seems that GLM4V vision encoder functionality has just been added to llama.cpp here: https://github.com/ggml-org/llama.cpp/pull/18042
It is sufficient for GLM-4.6V vision features to become supported by Ollama ?

<!-- gh-comment-id:3677803065 --> @SurealCereal commented on GitHub (Dec 20, 2025): It seems that GLM4V vision encoder functionality has just been added to `llama.cpp` here: https://github.com/ggml-org/llama.cpp/pull/18042 It is sufficient for GLM-4.6V vision features to become supported by Ollama ?
Author
Owner

@rick-github commented on GitHub (Dec 21, 2025):

ollama 0.13.5 loads the vision projector but there is a mis-alignment of some sort as the model generates content not related to the image.

The developers have indicated that supporting split models is not a goal and so models like this and qwen3-vl may only be supported by a fused model. As there is currently no native support for GLM, support for the vision variants of GLM may take some time.

<!-- gh-comment-id:3678317364 --> @rick-github commented on GitHub (Dec 21, 2025): ollama 0.13.5 loads the vision projector but there is a mis-alignment of some sort as the model generates content not related to the image. The developers have indicated that supporting split models is [not a goal](https://github.com/ollama/ollama/pull/13306#issuecomment-3604748281) and so models like this and [qwen3-vl](https://github.com/ollama/ollama/issues/13480) may only be supported by a fused model. As there is currently no [native support](https://github.com/ollama/ollama/tree/main/model/models) for GLM, support for the vision variants of GLM may take some time.
Author
Owner

@snapo commented on GitHub (Dec 21, 2025):

Also very important the Vision part of the model when creating a GGUF should be in f16 (rest can be Q4_0) but the vision part degrades extremely quickly if less than f16 is used....

<!-- gh-comment-id:3678331042 --> @snapo commented on GitHub (Dec 21, 2025): Also very important the Vision part of the model when creating a GGUF should be in f16 (rest can be Q4_0) but the vision part degrades extremely quickly if less than f16 is used....
Author
Owner

@snapo commented on GitHub (Dec 26, 2025):

Is there a status update when this will be available in ollama?

<!-- gh-comment-id:3692189433 --> @snapo commented on GitHub (Dec 26, 2025): Is there a status update when this will be available in ollama?
Author
Owner

@yumo666666 commented on GitHub (Dec 26, 2025):

什么时候会在Ollama上提供这个功能,有更新吗?

The neighboring LMstudio has already adopted it and can be used visually.

<!-- gh-comment-id:3692258896 --> @yumo666666 commented on GitHub (Dec 26, 2025): > 什么时候会在Ollama上提供这个功能,有更新吗? The neighboring LMstudio has already adopted it and can be used visually.
Author
Owner

@rick-github commented on GitHub (Jan 1, 2026):

The unsloth weights have been updated and the vision component now works:

$ ollama run hf.co/unsloth/GLM-4.6V-Flash-GGUF:Q4_K_M describe this image ./image1.jpg 
Added image './image1.jpg'
<think>Got it, let's describe this image. The picture shows an adorable 
white puppy sitting on a stone or marble surface. The puppy has soft, fluffy 
fur and is wearing a red collar with a small bell attached to it. The 
background appears to be outdoors, maybe near a building with a dark area 
behind the puppy and a lighter-colored structure above the stone surface. 
The puppy's expression looks calm and curious as it gazes slightly to the 
side. The overall scene is warm and cute, focusing on the innocent and 
fluffy appearance of the young dog. Let me check the details: fur texture is 
very soft-looking, collar is red with a few beads or decorations, bell is 
golden. The setting seems like a patio or step, with weathered stone tiles 
that have some wear marks. The lighting is natural, probably daylight, 
giving a gentle tone to the image. So putting it all together, describe the 
elements: subject (white puppy), appearance (fluffy fur, red collar with 
bell), posture (sitting on stone steps), background (outdoor setting with 
architectural details).</think>
This image features an **adorable white fluffy puppy** sitting on a 
weathered, light - colored stone surface (likely a step or patio tile). The 
pup has soft, plush fur and wears a **red collar** adorned with small 
decorative elements; attached to the collar is a tiny golden bell.  

The puppy’s posture is upright yet relaxed as it gazes slightly to its 
right, exuding an innocent and curious demeanor. Its dark eyes and delicate 
facial features (including a small nose and subtle whiskers) add to its 
charm. The background consists of an outdoor setting: a darker area behind 
the puppy (possibly a wall or door) and a lighter - toned structure above 
the stone surface, suggesting a residential or outdoor architectural 
environment. The overall mood is warm and heartwarming, emphasizing the 
puppy’s cute and fluffy appearance against the natural - looking stonework.

The unsloth template doesn't have thinking control, the model can be combined with a better template for that ability:

$ ollama run glm-4.6v:9b-flash-q4_K_M describe this image ./image1.jpg 
Added image './image1.jpg'
Thinking...
Got it, let's describe this image. The main subject is a small white puppy 
sitting on a stone or marble surface. The puppy has fluffy fur, maybe a 
breed like a Samoyed or a similar fluffy dog. It's wearing a red collar with 
two small gold bells attached. One bell is near the center of the collar, 
and another lower one? Wait, looking at the image: the red collar has beads, 
and there are two tiny golden bells. The puppy's expression is calm, maybe a 
bit curious as it looks to the right.

The background shows an outdoor setting with a blurred brown or 
reddish-brown wall behind the stone steps. The ground is made of large, 
textured marble slabs with some white and gray patterns, giving a natural, 
slightly worn look. The lighting is soft, probably natural light, making the 
puppy's white fur stand out against the neutral background.

So putting it all together: A fluffy white puppy with a red collar featuring 
two small golden bells sits on a weathered stone or marble step. The scene 
appears to be outdoors, with a blurred brownish wall in the background, 
suggesting a porch or similar outdoor space. The puppy's fur is soft and 
well-groomed, and its posture is relaxed yet attentive as it gazes to the 
side. The overall mood is gentle and endearing, highlighting the cute 
appearance of the young dog.
...done thinking.

The image features a **fluffy white puppy** (likely a young Samoyed or 
similar breed) sitting on a weathered stone/marble surface. The pup has 
soft, thick fur and wears a **red beaded collar** with two small **golden 
bells** attached—one near the center of the collar, another lower down.  

The background shows an outdoor setting: blurred brown/reddish-brown wall 
behind the steps, suggesting a porch or exterior space. The stone surface 
has natural gray-and-white veining, giving it a worn, textured appearance. 
Lighting is soft (likely natural), emphasizing the puppy’s bright white fur 
against the neutral backdrop.  

The puppy sits attentively, gazing slightly to its right with an endearing, 
calm expression. Its posture is relaxed yet alert, capturing a gentle, 
adorable moment in what appears to be an outdoor environment.
<!-- gh-comment-id:3703334248 --> @rick-github commented on GitHub (Jan 1, 2026): The unsloth weights have been [updated](https://huggingface.co/unsloth/GLM-4.6V-Flash-GGUF/discussions/4) and the vision component now works: ```console $ ollama run hf.co/unsloth/GLM-4.6V-Flash-GGUF:Q4_K_M describe this image ./image1.jpg Added image './image1.jpg' <think>Got it, let's describe this image. The picture shows an adorable white puppy sitting on a stone or marble surface. The puppy has soft, fluffy fur and is wearing a red collar with a small bell attached to it. The background appears to be outdoors, maybe near a building with a dark area behind the puppy and a lighter-colored structure above the stone surface. The puppy's expression looks calm and curious as it gazes slightly to the side. The overall scene is warm and cute, focusing on the innocent and fluffy appearance of the young dog. Let me check the details: fur texture is very soft-looking, collar is red with a few beads or decorations, bell is golden. The setting seems like a patio or step, with weathered stone tiles that have some wear marks. The lighting is natural, probably daylight, giving a gentle tone to the image. So putting it all together, describe the elements: subject (white puppy), appearance (fluffy fur, red collar with bell), posture (sitting on stone steps), background (outdoor setting with architectural details).</think> This image features an **adorable white fluffy puppy** sitting on a weathered, light - colored stone surface (likely a step or patio tile). The pup has soft, plush fur and wears a **red collar** adorned with small decorative elements; attached to the collar is a tiny golden bell. The puppy’s posture is upright yet relaxed as it gazes slightly to its right, exuding an innocent and curious demeanor. Its dark eyes and delicate facial features (including a small nose and subtle whiskers) add to its charm. The background consists of an outdoor setting: a darker area behind the puppy (possibly a wall or door) and a lighter - toned structure above the stone surface, suggesting a residential or outdoor architectural environment. The overall mood is warm and heartwarming, emphasizing the puppy’s cute and fluffy appearance against the natural - looking stonework. ``` The unsloth template doesn't have thinking control, the model can be combined with a [better template](https://github.com/ollama/ollama/issues/11563#issuecomment-3218169593) for that ability: ```console $ ollama run glm-4.6v:9b-flash-q4_K_M describe this image ./image1.jpg Added image './image1.jpg' Thinking... Got it, let's describe this image. The main subject is a small white puppy sitting on a stone or marble surface. The puppy has fluffy fur, maybe a breed like a Samoyed or a similar fluffy dog. It's wearing a red collar with two small gold bells attached. One bell is near the center of the collar, and another lower one? Wait, looking at the image: the red collar has beads, and there are two tiny golden bells. The puppy's expression is calm, maybe a bit curious as it looks to the right. The background shows an outdoor setting with a blurred brown or reddish-brown wall behind the stone steps. The ground is made of large, textured marble slabs with some white and gray patterns, giving a natural, slightly worn look. The lighting is soft, probably natural light, making the puppy's white fur stand out against the neutral background. So putting it all together: A fluffy white puppy with a red collar featuring two small golden bells sits on a weathered stone or marble step. The scene appears to be outdoors, with a blurred brownish wall in the background, suggesting a porch or similar outdoor space. The puppy's fur is soft and well-groomed, and its posture is relaxed yet attentive as it gazes to the side. The overall mood is gentle and endearing, highlighting the cute appearance of the young dog. ...done thinking. The image features a **fluffy white puppy** (likely a young Samoyed or similar breed) sitting on a weathered stone/marble surface. The pup has soft, thick fur and wears a **red beaded collar** with two small **golden bells** attached—one near the center of the collar, another lower down. The background shows an outdoor setting: blurred brown/reddish-brown wall behind the steps, suggesting a porch or exterior space. The stone surface has natural gray-and-white veining, giving it a worn, textured appearance. Lighting is soft (likely natural), emphasizing the puppy’s bright white fur against the neutral backdrop. The puppy sits attentively, gazing slightly to its right with an endearing, calm expression. Its posture is relaxed yet alert, capturing a gentle, adorable moment in what appears to be an outdoor environment. ```
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#8841