[GH-ISSUE #14579] Qwen3.5 Much slower speeds compared to LlamaCPP #35212

New Issue

GiteaMirror · 2026-04-22T19:35:45-05:00

GiteaMirror commented

2026-04-22 19:35:45 -05:00

Originally created by @iChristGit on GitHub (Mar 3, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14579

What is the issue?

I am running the same Q4 quant of Qwen3.5-35B-A3B
On ollama I get around 15 to 20 tokens per second, very slow I’ve tried many different starting arguments and just default ollama startup.

The same PC just running straight up llama CPP I get 100 tokens per second even with default settings no argument, context 160000.

Win11
3090Ti
64Gb ram

I am not sure what to add here, il upload whatever needed.

Relevant log output

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

0.7.15

Originally created by @iChristGit on GitHub (Mar 3, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14579 ### What is the issue? I am running the same Q4 quant of Qwen3.5-35B-A3B On ollama I get around 15 to 20 tokens per second, very slow I’ve tried many different starting arguments and just default ollama startup. The same PC just running straight up llama CPP I get 100 tokens per second even with default settings no argument, context 160000. <img width="590" height="1278" alt="Image" src="https://github.com/user-attachments/assets/23af5b9e-d483-4ecb-874f-efc25d882a35" /> Win11 3090Ti 64Gb ram I am not sure what to add here, il upload whatever needed. ### Relevant log output ```shell ``` ### OS Windows ### GPU Nvidia ### CPU Intel ### Ollama version 0.7.15

GiteaMirror added the bug label 2026-04-22 19:35:45 -05:00

GiteaMirror commented

2026-04-22 19:35:47 -05:00

@Menooker commented on GitHub (Mar 5, 2026):

Possibly related to https://github.com/ollama/ollama/issues/11772
See also https://www.reddit.com/r/LocalLLaMA/comments/1rei65v/qwen3535ba3b_quantization_quality_speed/?tl=en

Someone reported in above link. Quote

I actually love you so much. I'm running this on a 5070ti 12700k 32GB 5400MT system and I had no clue how much difference using the MOE layer option improves performance. Went from 10tps (using gpu offload settings) to 57tps (using your 24 cpu layer config) and then to around 70tps (using 14 cpu layers instead).

Looks like CPU MOE offloading is important for this model.

@Menooker commented on GitHub (Mar 5, 2026): Possibly related to https://github.com/ollama/ollama/issues/11772 See also https://www.reddit.com/r/LocalLLaMA/comments/1rei65v/qwen3535ba3b_quantization_quality_speed/?tl=en Someone reported in above link. Quote >I actually love you so much. I'm running this on a 5070ti 12700k 32GB 5400MT system and I had no clue how much difference using the MOE layer option improves performance. Went from 10tps (using gpu offload settings) to 57tps (using your 24 cpu layer config) and then to around 70tps (using 14 cpu layers instead). Looks like CPU MOE offloading is important for this model.

GiteaMirror commented

2026-04-22 19:35:47 -05:00

@OniricApps commented on GitHub (Mar 5, 2026):

I also realized a very slow performance of qwen3.5:27b

I can confirm only GPU is used (a RTX 3090 with 24GB of VRAM):

$ ollama ps
NAME ID SIZE PROCESSOR CONTEXT UNTIL
qwen3.5:27b 7653528ba5cb 23 GB 100% GPU 16000 4 minutes from now

$ ollama --version
ollama version is 0.17.5

I realized that also CPU is at 100% with "top" command.

qwen3.5 is taking from 3 to 6 minutes per inferences than I execute in less than 20 seconds using other similar sized models like gemma3:27b. Concretely, I'm doing a test with 4 inferences in a row that takes 46 seconds for gemma3 and 3118 seconds for qwen3.5, it is 67x slower. Not normal at all.

@OniricApps commented on GitHub (Mar 5, 2026): I also realized a very slow performance of qwen3.5:27b I can confirm only GPU is used (a RTX 3090 with 24GB of VRAM): $ ollama ps NAME ID SIZE PROCESSOR CONTEXT UNTIL qwen3.5:27b 7653528ba5cb 23 GB 100% GPU 16000 4 minutes from now $ ollama --version ollama version is 0.17.5 I realized that also CPU is at 100% with "top" command. qwen3.5 is taking from 3 to 6 minutes per inferences than I execute in less than 20 seconds using other similar sized models like gemma3:27b. Concretely, I'm doing a test with 4 inferences in a row that takes 46 seconds for gemma3 and 3118 seconds for qwen3.5, it is 67x slower. Not normal at all.

GiteaMirror commented

2026-04-22 19:35:48 -05:00

@MarkMuravev commented on GitHub (Mar 6, 2026):

+1

@MarkMuravev commented on GitHub (Mar 6, 2026): +1

GiteaMirror commented

2026-04-22 19:35:48 -05:00

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

@iChristGit @OniricApps

I also realized a very slow performance of qwen3.5:27b

I can confirm only GPU is used (a RTX 3090 with 24GB of VRAM):

$ ollama ps NAME ID SIZE PROCESSOR CONTEXT UNTIL qwen3.5:27b 7653528ba5cb 23 GB 100% GPU 16000 4 minutes from now

$ ollama --version ollama version is 0.17.5

I realized that also CPU is at 100% with "top" command.

qwen3.5 is taking from 3 to 6 minutes per inferences than I execute in less than 20 seconds using other similar sized models like gemma3:27b. Concretely, I'm doing a test with 4 inferences in a row that takes 46 seconds for gemma3 and 3118 seconds for qwen3.5, it is 67x slower. Not normal at all.

I had the same issue. Its the Context Size. By default qwen3.5-27b runs with 32k Context, which is too big for 24 GB VRAM and overflows. I only got 17 token/s.

Setting context to 16k or 8k increased the inference speed from 17 to 25 toklen/s (16k context) to 40 token/s (8k context). Its still 1/2 the speed of Qwen3 30b with 90 token/s.

"ollama ps" shows 24 (32k context) 23 (16k context) and 22 GB VRAM (8k). So if you have anything taking RAM (like windows), 32k won't fit in anymore.

Also Qwen 3.5 27b reasoning is fucked up. Doing a simple prompt of "tell me a story" gives 5 pages of reasoning for a 2 short paragraphs of story. Just turn off reasoning with "/set nothink"

@TsengSR commented on GitHub (Mar 15, 2026): @iChristGit @OniricApps > I also realized a very slow performance of qwen3.5:27b > > I can confirm only GPU is used (a RTX 3090 with 24GB of VRAM): > > $ ollama ps NAME ID SIZE PROCESSOR CONTEXT UNTIL qwen3.5:27b 7653528ba5cb 23 GB 100% GPU 16000 4 minutes from now > > $ ollama --version ollama version is 0.17.5 > > I realized that also CPU is at 100% with "top" command. > > qwen3.5 is taking from 3 to 6 minutes per inferences than I execute in less than 20 seconds using other similar sized models like gemma3:27b. Concretely, I'm doing a test with 4 inferences in a row that takes 46 seconds for gemma3 and 3118 seconds for qwen3.5, it is 67x slower. Not normal at all. I had the same issue. Its the Context Size. By default qwen3.5-27b runs with 32k Context, which is too big for 24 GB VRAM and overflows. I only got 17 token/s. Setting context to 16k or 8k increased the inference speed from 17 to 25 toklen/s (16k context) to 40 token/s (8k context). Its still 1/2 the speed of Qwen3 30b with 90 token/s. "ollama ps" shows 24 (32k context) 23 (16k context) and 22 GB VRAM (8k). So if you have anything taking RAM (like windows), 32k won't fit in anymore. Also Qwen 3.5 27b reasoning is fucked up. Doing a simple prompt of "tell me a story" gives 5 pages of reasoning for a 2 short paragraphs of story. Just turn off reasoning with "/set nothink"

GiteaMirror commented

2026-04-22 19:35:49 -05:00

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

@iChristGit @OniricApps

I also realized a very slow performance of qwen3.5:27b
I can confirm only GPU is used (a RTX 3090 with 24GB of VRAM):
$ ollama ps NAME ID SIZE PROCESSOR CONTEXT UNTIL qwen3.5:27b 7653528ba5cb 23 GB 100% GPU 16000 4 minutes from now
$ ollama --version ollama version is 0.17.5
I realized that also CPU is at 100% with "top" command.
qwen3.5 is taking from 3 to 6 minutes per inferences than I execute in less than 20 seconds using other similar sized models like gemma3:27b. Concretely, I'm doing a test with 4 inferences in a row that takes 46 seconds for gemma3 and 3118 seconds for qwen3.5, it is 67x slower. Not normal at all.

I had the same issue. Its the Context Size. By default qwen3.5-27b runs with 32k Context, which is too big for 24 GB VRAM and overflows. I only got 17 token/s.

Setting context to 16k or 8k increased the inference speed from 17 to 25 toklen/s (16k context) to 40 token/s (8k context). Its still 1/2 the speed of Qwen3 30b with 90 token/s.

"ollama ps" shows 24 (32k context) 23 (16k context) and 22 GB VRAM (8k). So if you have anything taking RAM (like windows), 32k won't fit in anymore.

Also Qwen 3.5 27b reasoning is fucked up. Doing a simple prompt of "tell me a story" gives 5 pages of reasoning for a 2 short paragraphs of story. Just turn off reasoning with "/set nothink"

I am testing llama cpp at 60-120k tokens and still manage 35-40tk/s
while on ollama even at 16k its very very slow, at a point of unusable..
I am not sure its an issue with context.

@iChristGit commented on GitHub (Mar 15, 2026): > [@iChristGit](https://github.com/iChristGit) [@OniricApps](https://github.com/OniricApps) > > > I also realized a very slow performance of qwen3.5:27b > > I can confirm only GPU is used (a RTX 3090 with 24GB of VRAM): > > $ ollama ps NAME ID SIZE PROCESSOR CONTEXT UNTIL qwen3.5:27b 7653528ba5cb 23 GB 100% GPU 16000 4 minutes from now > > $ ollama --version ollama version is 0.17.5 > > I realized that also CPU is at 100% with "top" command. > > qwen3.5 is taking from 3 to 6 minutes per inferences than I execute in less than 20 seconds using other similar sized models like gemma3:27b. Concretely, I'm doing a test with 4 inferences in a row that takes 46 seconds for gemma3 and 3118 seconds for qwen3.5, it is 67x slower. Not normal at all. > > I had the same issue. Its the Context Size. By default qwen3.5-27b runs with 32k Context, which is too big for 24 GB VRAM and overflows. I only got 17 token/s. > > Setting context to 16k or 8k increased the inference speed from 17 to 25 toklen/s (16k context) to 40 token/s (8k context). Its still 1/2 the speed of Qwen3 30b with 90 token/s. > > "ollama ps" shows 24 (32k context) 23 (16k context) and 22 GB VRAM (8k). So if you have anything taking RAM (like windows), 32k won't fit in anymore. > > Also Qwen 3.5 27b reasoning is fucked up. Doing a simple prompt of "tell me a story" gives 5 pages of reasoning for a 2 short paragraphs of story. Just turn off reasoning with "/set nothink" I am testing llama cpp at 60-120k tokens and still manage 35-40tk/s while on ollama even at 16k its very very slow, at a point of unusable.. I am not sure its an issue with context.

GiteaMirror commented

2026-04-22 19:35:50 -05:00

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

I am testing llama cpp at 60-120k tokens and still manage 35-40tk/s while on ollama even at 16k its very very slow, at a point of unusable.. I am not sure its an issue with context.

In general Qwen3.5 27b is a good bit slower, i.e. Qwen3 30b at Q4 ran at 90 tokens/s for me, while 3.5 27b only manages 40 with the reduced context size.
Check your ollama ps output when the model runs, if its fully utilizing GPU or if it offloads to CPU.

If it offloads, you may need to reduce the context further. Als keep in mind to turn off reasoning.
In my tests the reasoning is about 5 to 10 times the amount of tokens/text than what really gets generated. Its reasoning kinda runs amok overthinking it before starting the actual response. On my first test (with 17 token/s still) it made 10 to 15 times more reasoning than the actual 2-3 paragraph output, which causes an abnormal response times, even at 40 tokens/second

@TsengSR commented on GitHub (Mar 15, 2026): > I am testing llama cpp at 60-120k tokens and still manage 35-40tk/s while on ollama even at 16k its very very slow, at a point of unusable.. I am not sure its an issue with context. In general Qwen3.5 27b is a good bit slower, i.e. Qwen3 30b at Q4 ran at 90 tokens/s for me, while 3.5 27b only manages 40 with the reduced context size. Check your `ollama ps` output when the model runs, if its fully utilizing GPU or if it offloads to CPU. If it offloads, you may need to reduce the context further. Als keep in mind to turn off reasoning. In my tests the reasoning is about 5 to 10 times the amount of tokens/text than what really gets generated. Its reasoning kinda runs amok overthinking it before starting the actual response. On my first test (with 17 token/s still) it made 10 to 15 times more reasoning than the actual 2-3 paragraph output, which causes an abnormal response times, even at 40 tokens/second

GiteaMirror commented

2026-04-22 19:35:50 -05:00

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

I am testing llama cpp at 60-120k tokens and still manage 35-40tk/s while on ollama even at 16k its very very slow, at a point of unusable.. I am not sure its an issue with context.

In general Qwen3.5 27b is a good bit slower, i.e. Qwen3 30b at Q4 ran at 90 tokens/s for me, while 3.5 27b only manages 40 with the reduced context size. Check your ollama ps output when the model runs, if its fully utilizing GPU or if it offloads to CPU.

If it offloads, you may need to reduce the context further. Als keep in mind to turn off reasoning. In my tests the reasoning is about 5 to 10 times the amount of tokens/text than what really gets generated. Its reasoning kinda runs amok overthinking it before starting the actual response. On my first test (with 17 token/s still) it made 10 to 15 times more reasoning than the actual 2-3 paragraph output, which causes an abnormal response times, even at 40 tokens/second

Its not about which model, as I test same model with both llama cpp and ollama, in 35b its less appearent because its still 80+tk/s, but when its 12tk/s vs 35tk/s its noticable

@iChristGit commented on GitHub (Mar 15, 2026): > > I am testing llama cpp at 60-120k tokens and still manage 35-40tk/s while on ollama even at 16k its very very slow, at a point of unusable.. I am not sure its an issue with context. > > In general Qwen3.5 27b is a good bit slower, i.e. Qwen3 30b at Q4 ran at 90 tokens/s for me, while 3.5 27b only manages 40 with the reduced context size. Check your `ollama ps` output when the model runs, if its fully utilizing GPU or if it offloads to CPU. > > If it offloads, you may need to reduce the context further. Als keep in mind to turn off reasoning. In my tests the reasoning is about 5 to 10 times the amount of tokens/text than what really gets generated. Its reasoning kinda runs amok overthinking it before starting the actual response. On my first test (with 17 token/s still) it made 10 to 15 times more reasoning than the actual 2-3 paragraph output, which causes an abnormal response times, even at 40 tokens/second Its not about which model, as I test same model with both llama cpp and ollama, in 35b its less appearent because its still 80+tk/s, but when its 12tk/s vs 35tk/s its noticable

GiteaMirror commented

2026-04-22 19:35:51 -05:00

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

Its not about which model, as I test same model with both llama cpp and ollama, in 35b its less appearent because its still 80+tk/s, but when its 12tk/s vs 35tk/s its noticable

Yea, same issue. 17 vs 40 Tokens when context lowered.

Thats to be expected, the qwen3.5 35b is something completely different from 27b. qwen3.5 35b is a mixture-of-experts where each expert has 3B parameters and only one of them is active for a given token, so your GPU/CPU only has to scan 3b parameters for each token vs 27b parameters on the dense model.

35b has more speed but less accurancy due to each expert only having 3b params. But its good for AI PCs with unifed memory as 3b paramters give a good token/s and it can pick the best expert model for a given prompt.

The catch is, you can't load the 35b model into 3090 or 4090, its too big even at Q4, so it will inevitable go into system memory which is 15-30 times slower than GPU memory. The 27b weights are 16 GB but need 24 GB of VRAM to load at 32k context

@TsengSR commented on GitHub (Mar 15, 2026): > Its not about which model, as I test same model with both llama cpp and ollama, in 35b its less appearent because its still 80+tk/s, but when its 12tk/s vs 35tk/s its noticable Yea, same issue. 17 vs 40 Tokens when context lowered. Thats to be expected, the qwen3.5 35b is something completely different from 27b. qwen3.5 35b is a mixture-of-experts where each expert has 3B parameters and only one of them is active for a given token, so your GPU/CPU only has to scan 3b parameters for each token vs 27b parameters on the dense model. 35b has more speed but less accurancy due to each expert only having 3b params. But its good for AI PCs with unifed memory as 3b paramters give a good token/s and it can pick the best expert model for a given prompt. The catch is, you can't load the 35b model into 3090 or 4090, its too big even at Q4, so it will inevitable go into system memory which is 15-30 times slower than GPU memory. The 27b weights are 16 GB but need 24 GB of VRAM to load at 32k context

GiteaMirror commented

2026-04-22 19:35:52 -05:00

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

Its not about which model, as I test same model with both llama cpp and ollama, in 35b its less appearent because its still 80+tk/s, but when its 12tk/s vs 35tk/s its noticable

Yea, same issue. 17 vs 40 Tokens when context lowered.

Thats to be expected, the qwen3.5 35b is something completely different from 27b. qwen3.5 35b is a mixture-of-experts where each expert has 3B parameters and only one of them is active for a given token, so your GPU/CPU only has to scan 3b parameters for each token vs 27b parameters on the dense model.

35b has more speed but less accurancy due to each expert only having 3b params. But its good for AI PCs with unifed memory as 3b paramters give a good token/s and it can pick the best expert model for a given prompt

My point is that if you take same exact model (Qwen3.5-27B-Q4) and use the SAME amount of context (aka set to 16k), there is still double the performance when running straight llama cpp as opposed to ollama.

You can also spot that difference with the 35B, but since the tk/s is so high people tend to not notice it, but you will still benefit from running llama cpp

@iChristGit commented on GitHub (Mar 15, 2026): > > Its not about which model, as I test same model with both llama cpp and ollama, in 35b its less appearent because its still 80+tk/s, but when its 12tk/s vs 35tk/s its noticable > > Yea, same issue. 17 vs 40 Tokens when context lowered. > > Thats to be expected, the qwen3.5 35b is something completely different from 27b. qwen3.5 35b is a mixture-of-experts where each expert has 3B parameters and only one of them is active for a given token, so your GPU/CPU only has to scan 3b parameters for each token vs 27b parameters on the dense model. > > 35b has more speed but less accurancy due to each expert only having 3b params. But its good for AI PCs with unifed memory as 3b paramters give a good token/s and it can pick the best expert model for a given prompt My point is that if you take same exact model (Qwen3.5-27B-Q4) and use the SAME amount of context (aka set to 16k), there is still double the performance when running straight llama cpp as opposed to ollama. You can also spot that difference with the 35B, but since the tk/s is so high people tend to not notice it, but you will still benefit from running llama cpp

GiteaMirror commented

2026-04-22 19:35:52 -05:00

@kokroo commented on GitHub (Mar 26, 2026):

+1

@kokroo commented on GitHub (Mar 26, 2026): +1

GiteaMirror commented

2026-04-22 19:35:53 -05:00

@iChristGit commented on GitHub (Mar 26, 2026):

This seem to get worse now, llama cpp optimized more and now the Qwen3.5-35B quant runs at 140tk/s and ollama got regressed and takes ages to load the model and VERY slow tk/s

@iChristGit commented on GitHub (Mar 26, 2026): This seem to get worse now, llama cpp optimized more and now the Qwen3.5-35B quant runs at 140tk/s and ollama got regressed and takes ages to load the model and VERY slow tk/s

GiteaMirror commented

2026-04-22 19:35:53 -05:00

@dgit90 commented on GitHub (Mar 27, 2026):

Hardware: RTX 5080 16GB VRAM, AMD Ryzen 9 9950X 16-core, 96GB DDR5-6000 RAM, Windows 11 + WSL2 (Ubuntu 24.04), Ollama installed natively inside WSL2.

Model: qwen3.5:35b-a3b-q4_K_M (23GB, Q4_K_M quantization)

Observed behavior: Simple single-turn prompt ("hi") takes approximately 43 seconds end-to-end. GPU utilization appears low during inference. Model does not fit fully in VRAM (16GB < 23GB), so Ollama is offloading layers.

Expected behavior: With llama.cpp's --n-cpu-moe flag, MoE expert layers can be offloaded to CPU RAM while attention layers remain on GPU, yielding ~70 tok/s on this hardware. Ollama does not currently expose this flag, resulting in naive layer offloading and dramatically worse performance on MoE models that partially exceed VRAM.

Request: Expose --n-cpu-moe (or equivalent MoE-aware offloading) as a configurable parameter in Ollama for MoE architectures. This would significantly improve usability of Qwen3.5 and similar MoE models on consumer hardware with 16GB VRAM.

@dgit90 commented on GitHub (Mar 27, 2026): Hardware: RTX 5080 16GB VRAM, AMD Ryzen 9 9950X 16-core, 96GB DDR5-6000 RAM, Windows 11 + WSL2 (Ubuntu 24.04), Ollama installed natively inside WSL2. Model: qwen3.5:35b-a3b-q4_K_M (23GB, Q4_K_M quantization) Observed behavior: Simple single-turn prompt ("hi") takes approximately 43 seconds end-to-end. GPU utilization appears low during inference. Model does not fit fully in VRAM (16GB < 23GB), so Ollama is offloading layers. Expected behavior: With llama.cpp's --n-cpu-moe flag, MoE expert layers can be offloaded to CPU RAM while attention layers remain on GPU, yielding ~70 tok/s on this hardware. Ollama does not currently expose this flag, resulting in naive layer offloading and dramatically worse performance on MoE models that partially exceed VRAM. Request: Expose --n-cpu-moe (or equivalent MoE-aware offloading) as a configurable parameter in Ollama for MoE architectures. This would significantly improve usability of Qwen3.5 and similar MoE models on consumer hardware with 16GB VRAM.

Sign in to join this conversation.

Branches Tags

main

hoyyeva/anthropic-local-image-path

dhiltgen/ci

dhiltgen/llama-runner

parth-remove-claude-desktop-launch

hoyyeva/anthropic-reference-images-path

parth-anthropic-reference-images-path

brucemacd/download-before-remove

hoyyeva/editor-config-repair

parth-mlx-decode-checkpoints

parth-launch-codex-app

hoyyeva/fix-codex-model-metadata-warning

hoyyeva/qwen

parth/hide-claude-desktop-till-release

hoyyeva/opencode-image-modality

parth-add-claude-code-autoinstall

release_v0.22.0

pdevine/manifest-list

codex/fix-codex-model-metadata-warning

pdevine/addressable-manifest

brucemacd/launch-fetch-reccomended

jmorganca/llama-compat

launch-copilot-cli

hoyyeva/opencode-thinking

release_v0.20.7

parth-auto-save-backup

parth-test

jmorganca/gemma4-audio-replacements

fix-manifest-digest-on-pull

hoyyeva/vscode-improve

brucemacd/install-server-wait

parth/update-claude-docs

brucemac/start-ap-install

pdevine/mlx-update

pdevine/qwen35_vision

drifkin/api-show-fallback

mintlify/image-generation-1773352582

hoyyeva/server-context-length-local-config

jmorganca/faster-reptition-penalties

jmorganca/convert-nemotron

parth-pi-thinking

pdevine/sampling-penalties

jmorganca/fix-create-quantization-memory

dongchen/resumable_transfer_fix

pdevine/sampling-cache-error

jessegross/mlx-usage

hoyyeva/openclaw-config

hoyyeva/app-html

pdevine/qwen3next

brucemacd/sign-sh-install

brucemacd/tui-update

brucemacd/usage-api

jmorganca/launch-empty

fix-app-dist-embed

mxyng/mlx-compile

mxyng/mlx-quant

mxyng/mlx-glm4.7

mxyng/mlx

brucemacd/simplify-model-picker

jmorganca/qwen3-concurrent

fix-glm-4.7-flash-mla-config

drifkin/qwen3-coder-opening-tag

brucemacd/usage-cli

fix-cuda12-fattn-shmem

ollama-imagegen-docs

parth/fix-multiline-inputs

brucemacd/config-docs

mxyng/model-files

mxyng/simple-execute

fix-imagegen-ollama-models

mxyng/async-upload

jmorganca/lazy-no-dtype-changes

imagegen-auto-detect-create

parth/decrease-concurrent-download-hf

fix-mlx-quantize-init

jmorganca/x-cleanup

usage

imagegen-readme

jmorganca/glm-image

mlx-gpu-cd

jmorganca/imagegen-modelfile

parth/agent-skills

parth/agent-allowlist

parth/signed-in-offline

parth/agents

parth/fix-context-chopping

improve-cloud-flow

parth/add-models-websearch

parth/prompt-renderer-mcp

jmorganca/native-settings

jmorganca/download-stream-hash

jmorganca/client2-rebased

brucemacd/oai-chat-req-multipart

jessegross/multi_chunk_reserve

grace/additional-omit-empty

grace/mistral-3-large

mxyng/tokenizer2

mxyng/tokenizer

jessegross/flash

hoyyeva/windows-nacked-app

mxyng/cleanup-attention

grace/deepseek-parser

hoyyeva/remember-unsent-prompt

parth/add-lfs-pointer-error-conversion

parth/olmo2-test2

hoyyeva/ollama-launchagent-plist

nicole/olmo-model

parth/olmo-test

mxyng/remove-embedded

parth/render-template

jmorganca/intellect-3

parth/remove-prealloc-linter

jmorganca/cmd-eval

nicole/nomic-embed-text-fix

mxyng/lint-2

hoyyeva/add-gemini-3-pro-preview

hoyyeva/load-model-list

mxyng/expand-path

mxyng/environ-2

hoyyeva/deeplink-json-encoding

parth/improve-tool-calling-tests

hoyyeva/conversation

hoyyeva/assistant-edit-response

hoyyeva/thinking

origin/brucemacd/invalid-char-i-err

parth/improve-tool-calling

jmorganca/required-omitempty

grace/qwen3-vl-tests

mxyng/iter-client

parth/docs-readme

nicole/embed-test

pdevine/integration-benchstat

parth/remove-generate-cmd

parth/add-toolcall-id

mxyng/server-tests

jmorganca/glm-4.6

jmorganca/gin-h-compat

drifkin/stable-tool-args

pdevine/qwen3-more-thinking

parth/add-websearch-client

nicole/websearch_local

jmorganca/qwen3-coder-updates

grace/deepseek-v3-migration-tests

mxyng/fix-create

jmorganca/cloud-errors

pdevine/parser-tidy

revert-12233-parth/simplify-entrypoints-runner

parth/enable-so-gpt-oss

brucemacd/qwen3vl

jmorganca/readme-simplify

parth/gpt-oss-structured-outputs

revert-12039-jmorganca/tools-braces

mxyng/embeddings

mxyng/gguf

mxyng/benchmark

mxyng/types-null

parth/move-parsing

mxyng/gemma2

jmorganca/docs

mxyng/16-bit

mxyng/create-stdin

pdevine/authorizedkeys

mxyng/quant

parth/opt-in-error-context-window

brucemacd/cache-models

brucemacd/runner-completion

jmorganca/llama-update-6

brucemacd/benchmark-list

brucemacd/partial-read-caps

parth/deepseek-r1-tools

mxyng/omit-array

parth/tool-prefix-temp

brucemacd/runner-test

jmorganca/qwen25vl

brucemacd/model-forward-test-ext

parth/python-function-parsing

jmorganca/cuda-compression-none

drifkin/num-parallel

drifkin/chat-truncation-fix

jmorganca/sync

parth/python-tools-calling

drifkin/array-head-count

brucemacd/create-no-loop

parth/server-enable-content-stream-with-tools

qwen25omni

mxyng/v3

brucemacd/ropeconfig

jmorganca/silence-tokenizer

parth/sample-so-test

parth/sampling-structured-outputs

brucemacd/doc-go-engine

parth/constrained-sampling-json

jmorganca/mistral-wip

brucemacd/mistral-small-convert

parth/sample-unmarshal-json-for-params

brucemacd/jomorganca/mistral

pdevine/bfloat16

jmorganca/mistral

brucemacd/mistral

pdevine/logging

parth/sample-correctness-fix

parth/sample-fix-sorting

jmorgan/sample-fix-sorting-extras

jmorganca/temp-0-images

brucemacd/parallel-embed-models

brucemacd/shim-grammar

jmorganca/fix-gguf-error

bmizerany/nameswork

jmorganca/faster-releases

bmizerany/validatenames

brucemacd/err-no-vocab

brucemacd/rope-config

brucemacd/err-hint

brucemacd/qwen2_5

brucemacd/logprobs

brucemacd/new_runner_graph_bench

progress-flicker

brucemacd/forward-test

brucemacd/go_qwen2

pdevine/gemma2

jmorganca/add-missing-symlink-eval

mxyng/next-debug

parth/set-context-size-openai

brucemacd/next-bpe-bench

brucemacd/next-bpe-test

brucemacd/new_runner_e2e

brucemacd/new_runner_qwen2

pdevine/convert-cohere2

brucemacd/convert-cli

parth/log-probs

mxyng/next-mlx

mxyng/cmd-history

parth/templating

parth/tokenize-detokenize

brucemacd/check-key-register

bmizerany/grammar

jmorganca/vendor-081b29bd

mxyng/func-checks

jmorganca/fix-null-format

parth/fix-default-to-warn-json

jmorganca/qwen2vl

jmorganca/no-concat

parth/cmd-cleanup-SO

brucemacd/check-key-register-structured-err

parth/openai-stream-usage

parth/fix-referencing-so

stream-tools-stop

jmorganca/degin-1

brucemacd/install-path-clean

brucemacd/push-name-validation

brucemacd/browser-key-register

jmorganca/openai-fix-first-message

jmorganca/fix-proxy

jessegross/sample

parth/disallow-streaming-tools

dhiltgen/remove_submodule

jmorganca/ga

jmorganca/mllama

pdevine/newlines

pdevine/geems-2b

jmorganca/llama-bump

mxyng/modelname-7

mxyng/gin-slog

mxyng/modelname-6

jyan/convert-prog

jyan/quant5

paligemma-support

pdevine/import-docs

jmorganca/openai-context

jyan/paligemma

jyan/p2

jyan/palitest

bmizerany/embedspeedup

jmorganca/llama-vit

brucemacd/allow-ollama

royh/ep-methods

royh/whisper

mxyng/api-models

mxyng/fix-memory

jyan/q4_4/8

jyan/ollama-v

royh/stream-tools

roy-embed-parallel

bmizerany/hrm

revert-5963-revert-5924-mxyng/llama3.1-rope

royh/embed-viz

jyan/local2

jyan/auth

jyan/local

jyan/parse-temp

jmorganca/template-mistral

jyan/reord-g

royh-openai-suffixdocs

royh-imgembed

royh-embed-parallel

jyan/quant4

royh-precision

jyan/progress

pdevine/fix-template

jyan/quant3

pdevine/ggla

mxyng/update-registry-domain

jmorganca/ggml-static

mxyng/create-context

jyan/v0.146

mxyng/layers-from-files

build_dist

bmizerany/noseek

royh-ls

royh-name

timeout

mxyng/server-timestamp

bmizerany/nosillyggufslurps

royh-params

jmorganca/llama-cpp-7c26775

royh-openai-delete

royh-show-rigid

jmorganca/enable-fa

jmorganca/no-error-template

jyan/format

royh-testdelete

bmizerany/fastverify

language_support

pdevine/ps-glitches

brucemacd/tokenize

bruce/iq-quants

bmizerany/filepathwithcoloninhost

mxyng/split-bin

bmizerany/client-registry

jmorganca/if-none-match

native

jmorganca/native

jmorganca/batch-embeddings

jmorganca/initcmake

jmorganca/mm

pdevine/showggmlinfo

modenameenforcealphanum

bmizerany/modenameenforcealphanum

jmorganca/done-reason

jmorganca/llama-cpp-8960fe8

ollama.com

bmizerany/filepathnobuild

bmizerany/types/model/defaultfix

rmdisplaylong

nogogen

bmizerany/x

modelfile-readme

bmizerany/replacecolon

jmorganca/limit

jmorganca/execstack

jmorganca/replace-assets

mxyng/tune-concurrency

jmorganca/testing

whitespace-detection

jmorganca/options

upgrade-all

scratch

cuda-search

mattw/airenamer

mattw/allmodelsonhuggingface

mattw/quantcontext

mattw/whatneedstorun

brucemacd/llama-mem-calc

mattw/faq-context

mattw/communitylinks

mattw/noprune

mattw/python-functioncalling

rename

mxyng/install

pulse

remove-first

editor

mattw/selfqueryingretrieval

cgo

mattw/howtoquant

api

matt/streamingapi

format-config

mxyng/extra-args

shell

update-nous-hermes

cp-model

upload-progress

fix-unknown-model

fix-model-names

delete-fix

insecure-registry

ls

deletemodels

progressbar

readme-updates

license-layers

skip-list

list-models

modelpath

matt/examplemodelfiles

distribution

go-opts

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: github-starred/ollama#35212