1 Release published by 1 person

• v0.6.5

  published Apr 6, 2025

15 Pull requests merged by 13 people

• types: include the 'items' and '$defs' fields to properly handle "array" types

  #10091 merged Apr 10, 2025

• Fix nondeterministic model unload order

  #10185 merged Apr 9, 2025

• Fix dockerfile.

  #9855 merged Apr 9, 2025

• fix(integration): move waitgroup Add(1) outside goroutine to avoid potential issue

  #10070 merged Apr 8, 2025

• kvcache: stub out test structs

  #10120 merged Apr 8, 2025

• types: add any type and validation for ToolFunction enum

  #10166 merged Apr 8, 2025

• cleanup: remove OLLAMA_TMPDIR and references to temporary executables

  #10182 merged Apr 8, 2025

• ollamarunner: Preallocate worst case graph at startup

  #10171 merged Apr 8, 2025

• Update README.md

  #10173 merged Apr 8, 2025

• Update README.md

  #10156 merged Apr 7, 2025

• Update README.md

  #10168 merged Apr 7, 2025

• types: allow tool function parameters with either a single type or an array of types

  #9434 merged Apr 7, 2025

• CONTRIBUTING: fix code block formatting

  #10169 merged Apr 7, 2025

• digest files in parallel

  #10134 merged Apr 7, 2025

• Create a new file descriptor for each goroutine during GGML model loading to improve loading from network filesystems.

  #10133 merged Apr 5, 2025

23 Pull requests opened by 18 people

• Added ollama4j-ui

  #10129 opened Apr 4, 2025

• server: improve spacing for JSON grammar

  #10131 opened Apr 4, 2025

• create blobs in parallel

  #10135 opened Apr 5, 2025

• wip: llama4 multimodal

  #10141 opened Apr 5, 2025

• chore: add miss error check

  #10144 opened Apr 6, 2025

• CONTRIBUTING: fix rendering the commit messages title format in browsers

  #10145 opened Apr 6, 2025

• Fix OpenAI model retrieval for models with slashes

  #10147 opened Apr 6, 2025

• discover: make unique_id check optional for AMD GPU detection

  #10150 opened Apr 6, 2025

• create: check architecture rather than vision.block_count when importing GGUF

  #10162 opened Apr 7, 2025

• Server: Enhance API/tag with Capability Information

  #10174 opened Apr 8, 2025

• chore: add miss Close() func

  #10179 opened Apr 8, 2025

• feat(installer): add gpu support options

  #10186 opened Apr 9, 2025

• fix: ensure log file is properly closed after logging completes

  #10187 opened Apr 9, 2025

• llama: update to commit 7538246e

  #10192 opened Apr 9, 2025

• scripts/install.sh: make curl progress bar optional

  #10196 opened Apr 9, 2025

• server: do not attempt to parse offset file as gguf

  #10201 opened Apr 9, 2025

• Update README.md

  #10202 opened Apr 9, 2025

• feat: capitalise ollama in ollama help description

  #10203 opened Apr 9, 2025

• Update README.md

  #10220 opened Apr 10, 2025

• ggml: Log filesystem errors

  #10221 opened Apr 10, 2025

• clarify quantization behavior in docs

  #10224 opened Apr 10, 2025

• ggml: fix crash for when head counts are arrays

  #10225 opened Apr 10, 2025

• ggml: Fix memory leak on input tensors

  #10226 opened Apr 10, 2025

41 Issues closed by 21 people

• Quantization Uses System Drive

  #10223 closed Apr 10, 2025

• hf pull has broken

  #10195 closed Apr 10, 2025

• Ollama 'API' at localhost:11434 returns a string of numbers, not a response.

  #10191 closed Apr 10, 2025

• Linux ollama 0.4.0, 0.4.2, 0.4.5, 0.5.0 custom compile for AMD ROCm fails missing ggml_rocm in go compile

  #7565 closed Apr 9, 2025

• Fails to build on macOS with "fatal error: {'string','cstdint'} file not found"

  #7392 closed Apr 9, 2025

• how to force ollama to use different cpu runners / how to compile windows avx512 runner?

  #6312 closed Apr 9, 2025

• CUSTOM_CPU_FLAGS="" / non avx2 build

  #8058 closed Apr 9, 2025

• go run . Server error

  #9012 closed Apr 9, 2025

• mistral-small v3.1

  #9827 closed Apr 9, 2025

• Mistral-small3.1 crashes on prompt

  #10175 closed Apr 9, 2025

• llama vs ollama

  #10190 closed Apr 9, 2025

• mistral-small3.1 using too much VRAM

  #10177 closed Apr 8, 2025

• Tool call - Ollama enforces usage of string in enums for JSON Schema

  #10164 closed Apr 8, 2025

• Out of memory errors when running `gemma3`

  #9791 closed Apr 8, 2025

• Very strange RAM behavior with v0.6.4 (memory leak?)

  #10132 closed Apr 8, 2025

• Exceeding GPU memory even I have 2 GPUs

  #10089 closed Apr 8, 2025

• Quantized Mistral small 3.1 doesn't utilize NVIDIA GPUs

  #10167 closed Apr 8, 2025

• NewLlamaServer failed - model requires more system memory for gemma3:12b

  #10181 closed Apr 8, 2025

• Issue with HTTP Status "100 ContinueHTTP/1.1 200 OK" When Sending List Input to 'POST /api/embed' Endpoint

  #10126 closed Apr 8, 2025

• Switched to `nomic-embed-text` model but still get `8192` dimension

  #10176 closed Apr 8, 2025

• options (temp, etc) print values list in python

  #9284 closed Apr 8, 2025

• `OLLAMA_CONTEXT_LENGTH=4096` but `OllamaEmbeddings` still shows `8192`

  #10149 closed Apr 8, 2025

• Tools and properties.type Not Supporting Arrays

  #5990 closed Apr 7, 2025

• 404 not found

  #9675 closed Apr 7, 2025

• llama.cpp server API compatibility

  #8579 closed Apr 7, 2025

• How to cancel a generate task by ollama restful api

  #9372 closed Apr 7, 2025

• Unable to run current version on MacOS

  #9736 closed Apr 7, 2025

• GPU Not Being Used Despite CUDA Installation and GPU Detection (Ollama 0.6.3 on Arch Linux(

  #10075 closed Apr 7, 2025

• Models listed aren't in order

  #10153 closed Apr 7, 2025

• Add full support for omni models

  #10004 closed Apr 7, 2025

• granite3.2-vision:latest or any other variants not running when trying to use image understanding capabilites

  #10161 closed Apr 7, 2025

• dial tcp: lookup dd20bb891979d25aebc8bec07b2b3bbc.r2.cloudflarestorage.com: no such host

  #10151 closed Apr 7, 2025

• Llama4ForConditionalGeneration unsupported issue

  #10158 closed Apr 7, 2025

• Failed to load `mistral-small:24b-3.1-instruct-2503-q4_K_M`

  #10154 closed Apr 6, 2025

• Why doesn't my L20 load the model into the GPU after starting ollama serve, here is the content of my OLLAMA_DEBUG=1 ollama serve.

  #10146 closed Apr 6, 2025

• ollama version 5.11.0 is too slow in the generate process

  #10140 closed Apr 6, 2025

• Performance is terrible

  #10137 closed Apr 5, 2025

• Running Ollama as a k8s STS with external script as entrypoint to load models

  #10122 closed Apr 5, 2025

• Ollama 0.6.0 with gemma3 can't load models from mounted Cloud Storage bucket on Cloud Run

  #9691 closed Apr 5, 2025

• llama3-gradient:1048k stuck at loading model

  #10111 closed Apr 4, 2025

• How to allocate more to the GPU?

  #10124 closed Apr 4, 2025

47 Issues opened by 41 people

• Support Jinja chat templates

  #10222 opened Apr 10, 2025

• System crashes when attempting to load a model that exceeds RAM capacity

  #10219 opened Apr 10, 2025

• Image recognition doesn't work with models downloaded from another site

  #10218 opened Apr 10, 2025

• mistral-small3.1 is not loaded fully to GPU on RX 7900 XTX

  #10217 opened Apr 10, 2025

• 8*H100 server didn't use GPU to run model

  #10216 opened Apr 10, 2025

• load_tensors: tensor 'token_embd.weight' cannot be used with preferred buffer type CUDA_Host, using CPU instead

  #10215 opened Apr 10, 2025

• Concurrency Does Not Scale When Increasing GPUs from 2x to 4x RTX 4090 serving `qwq` model

  #10214 opened Apr 10, 2025

• Using JSON-structured output seems to affect the model's output.

  #10213 opened Apr 10, 2025

• request: run models directly in the browser

  #10212 opened Apr 10, 2025

• Error: could not connect to ollama app, is it running? (MacOS)

  #10211 opened Apr 10, 2025

• ollama.com: profile links section incomplete

  #10210 opened Apr 10, 2025

• CLI: environment variable to disable stream

  #10209 opened Apr 10, 2025

• ollama show --verbose reporting wrong infromation

  #10208 opened Apr 10, 2025

• API Default Model

  #10207 opened Apr 10, 2025

• There are ads on the official website models?

  #10206 opened Apr 10, 2025

• Ollama ignores CUDA after reboot, falls back to CPU only

  #10204 opened Apr 9, 2025

• [Documentation] Automation of addition of latest models in readme through Github Actions

  #10200 opened Apr 9, 2025

• Error: vocabulary is larger than expected '262145' instead of '262144'

  #10199 opened Apr 9, 2025

• windows系统的，点开就默认安装c盘了，可以让修改一下么？

  #10198 opened Apr 9, 2025

• Add Human Feedback

  #10194 opened Apr 9, 2025

• Supporting private Hugging Face repos hosted on Jfrog Artifactory

  #10193 opened Apr 9, 2025

• 双服务器显卡共用

  #10189 opened Apr 9, 2025

• Support Dream 7b

  #10188 opened Apr 9, 2025

• Some kernel names are not shown on Nvidia Nsight System

  #10184 opened Apr 8, 2025

• Understanding context length

  #10183 opened Apr 8, 2025

• ollama run phi4-mini error

  #10180 opened Apr 8, 2025

• Option to disable CPU fallback for SOC with unified memory

  #10178 opened Apr 8, 2025

• tensor-split problem

  #10172 opened Apr 8, 2025

• Multimodal broken in 6.5?

  #10170 opened Apr 7, 2025

• Capitalize Ollama in `ollama` help description

  #10165 opened Apr 7, 2025

• qwen2.5:72b and llama3:70b not using GPU – extremely slow and consume 40GB+ RAM

  #10163 opened Apr 7, 2025

• Will --ctx-size 24576 override the environment variable OLLAMA_CONTEXT_LENGTH?

  #10160 opened Apr 7, 2025

• gemma3:27b gets stuck into generating the same token and producing useless gibberish output

  #10159 opened Apr 7, 2025

• Q6 quant (with vision support) for mistral-small:24b-3.1-instruct-2503?

  #10157 opened Apr 7, 2025

• Feature Request:Always use the GPU

  #10155 opened Apr 7, 2025

• RX580

  #10152 opened Apr 6, 2025

• Llama 4 support

  #10143 opened Apr 5, 2025

• Using Ollama Run -> /clear = Not clearing chat context

  #10142 opened Apr 5, 2025

• OpenAI API: Models with slashes not retrievable

  #10139 opened Apr 5, 2025

• Installer corrupted: cublastLt64_12.dll corrupted + other files.

  #10138 opened Apr 5, 2025

• Where is the log file and how can I configure the location of it?

  #10136 opened Apr 5, 2025

• Ai Vision - Wildy Different Results Between Gemma3 34b OpenAI / Vs Ollama Endpoint

  #10130 opened Apr 4, 2025

• Incorrect VRAM estimation

  #10128 opened Apr 4, 2025

• panic: failed to decode batch: could not find a kv cache slot (length: 6656)

  #10127 opened Apr 4, 2025

• qwen2.5-coder-cline:14b not using NVIDIA GPU

  #10125 opened Apr 4, 2025

• Structure output "allOf":

  #10123 opened Apr 4, 2025

• Can't use official QAT GGUF of Gemma-3-27b-it

  #10121 opened Apr 4, 2025

81 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• llama: remove model loading for grammar

  #10096 commented on Apr 10, 2025 • 19 new comments

• cmd: default to client2 and simplify pull progress display

  #10069 commented on Apr 9, 2025 • 6 new comments

• add ollama stop all

  #10043 commented on Apr 7, 2025 • 2 new comments

• Granite new engine

  #9966 commented on Apr 4, 2025 • 1 new comment

• support no caching when kvCacheType is "nocache" for deterministic completion

  #10064 commented on Apr 6, 2025 • 0 new comments

• NVIDIA GPU drivers not loaded on Jeston Orin Nano

  #9503 commented on Apr 10, 2025 • 0 new comments

• SPAM in your model database

  #9134 commented on Apr 10, 2025 • 0 new comments

• System memory leak. Gemma3

  #10040 commented on Apr 10, 2025 • 0 new comments

• SIGSEGV: segmentation violation

  #9665 commented on Apr 10, 2025 • 0 new comments

• AMD Ryzen NPU support

  #5186 commented on Apr 10, 2025 • 0 new comments

• AMD RX9070/9070XT support

  #9812 commented on Apr 10, 2025 • 0 new comments

• vram usage does not go back down after model unloads - stuck in Stopping...

  #7606 commented on Apr 9, 2025 • 0 new comments

• [ ROCm error: out of memory ] Runner Terminated: num_ctx within model / hardware limits reliably crashes

  #9957 commented on Apr 9, 2025 • 0 new comments

• Models can't be stopped correctly when using Webui combine with Ollama.

  #8969 commented on Apr 9, 2025 • 0 new comments

• qwen 2.5 coder stuck "Stopping"

  #8178 commented on Apr 9, 2025 • 0 new comments

• Stopping misbehaving model after some amount of time

  #9617 commented on Apr 9, 2025 • 0 new comments

• Provide logits or logprobs in the API

  #2415 commented on Apr 9, 2025 • 0 new comments

• Ollama always choses iGPU for computations in hybrind discrete+iGPU rocm setups

  #9588 commented on Apr 9, 2025 • 0 new comments

• Speed ten times slower than llamafile

  #8305 commented on Apr 9, 2025 • 0 new comments

• Batch embeddings get progressively worse with larger batches

  #6262 commented on Apr 9, 2025 • 0 new comments

• `pulling manifest Error: EOF` when pulling after disk is full

  #1731 commented on Apr 9, 2025 • 0 new comments

• Ollama errors on older versions of Linux/GLIBC on 0.5.13

  #9506 commented on Apr 9, 2025 • 0 new comments

• Ollama REFUSES to use GFX803 EVEN when detected

  #9807 commented on Apr 9, 2025 • 0 new comments

• Adding a search command

  #10046 commented on Apr 6, 2025 • 0 new comments

• server: enable content streaming with tools

  #10028 commented on Apr 4, 2025 • 0 new comments

• server: prevent model thrashing from unset API fields

  #10003 commented on Apr 7, 2025 • 0 new comments

• server: support streaming near tool usage

  #9973 commented on Apr 7, 2025 • 0 new comments

• Integration test improvements

  #9654 commented on Apr 9, 2025 • 0 new comments

• feat: add debug logging in chat/generate functions

  #8957 commented on Apr 6, 2025 • 0 new comments

• feat: Support Moore Threads GPU

  #7554 commented on Apr 10, 2025 • 0 new comments

• fix: consider any status code as redirect

  #7231 commented on Apr 10, 2025 • 0 new comments

• FEAT: add rerank support

  #7219 commented on Apr 8, 2025 • 0 new comments

• AMD integrated graphic on linux kernel 6.9.9+, GTT memory, loading freeze fix

  #6282 commented on Apr 5, 2025 • 0 new comments

• Enable AMD iGPU 780M in Linux, Create amd-igpu-780m.md

  #5426 commented on Apr 8, 2025 • 0 new comments

• llm/server.go: Fix ollama ps show 100%GPU even use CPU as runner

  #4906 commented on Apr 9, 2025 • 0 new comments

• cobra shell completions

  #4690 commented on Apr 9, 2025 • 0 new comments

• Ollama not freeing and eventually running out of memory [all models]

  #10114 commented on Apr 10, 2025 • 0 new comments

• Ollama hangs while generating a response

  #10119 commented on Apr 10, 2025 • 0 new comments

• Inference with OpenVINO on Intel

  #2169 commented on Apr 10, 2025 • 0 new comments

• CUDA error: an illegal memory access was encountered

  #9018 commented on Apr 10, 2025 • 0 new comments

• Add support for older AMD GPU gfx803, gfx802, gfx805 (e.g. Radeon RX 580, FirePro W7100)

  #2453 commented on Apr 10, 2025 • 0 new comments

• Add an easy way to list all models and their capabilities

  #10097 commented on Apr 6, 2025 • 0 new comments

• Unable to push: max retries exceeded on slower connections

  #2155 commented on Apr 5, 2025 • 0 new comments

• Possibility to remove max retries exceeded when downloading models from a slow connection

  #3162 commented on Apr 5, 2025 • 0 new comments

• Llama.cpp now supports distributed inference across multiple machines.

  #4643 commented on Apr 5, 2025 • 0 new comments

• Available memory calculation on AMD APU no longer takes GTT into account

  #5471 commented on Apr 5, 2025 • 0 new comments

• Allow importing multi-file GGUF models

  #5245 commented on Apr 5, 2025 • 0 new comments

• Deepseek R1, 671b is faster than 70b

  #10030 commented on Apr 5, 2025 • 0 new comments

• support deepseek 671b fp4

  #9419 commented on Apr 4, 2025 • 0 new comments

• gemma EOF error on image input due to improper memory management

  #10041 commented on Apr 4, 2025 • 0 new comments

• ollama does not utilize HBM3 memory on MI300A

  #8735 commented on Apr 4, 2025 • 0 new comments

• EOF with Gemma3:27b | POST predict: Post "http://127.0.0.1:35737/completion": EOF (status code: 500)

  #9699 commented on Apr 4, 2025 • 0 new comments

• Add Generate Embedding for Sparse vector

  #6230 commented on Apr 4, 2025 • 0 new comments

• add /metrics endpoint

  #3144 commented on Apr 4, 2025 • 0 new comments

• Llama3: Generated outputs inconsistent despite seed and temperature

  #5321 commented on Apr 4, 2025 • 0 new comments

• Update DeepSeek V3 to improved version

  #9980 commented on Apr 4, 2025 • 0 new comments

• Using Qwen as agent in VS Code

  #10038 commented on Apr 4, 2025 • 0 new comments

• Pull a model on start or without requiring serve

  #3369 commented on Apr 4, 2025 • 0 new comments

• Unsupported Value NaN in Ollama log

  #9639 commented on Apr 4, 2025 • 0 new comments

• Provide a single command for "serve + pull model", to be used in CI/CD

  #5385 commented on Apr 4, 2025 • 0 new comments

• Support for jinaai/jina-embeddings-v3 embedding model

  #6922 commented on Apr 4, 2025 • 0 new comments

• Ollama Not detecting adapter_config.json file

  #9505 commented on Apr 9, 2025 • 0 new comments

• Tool call support in Qwen 2.5 hallucinates with Maybe pattern

  #7051 commented on Apr 9, 2025 • 0 new comments

• Support for AMD 9000 GPUs

  #9633 commented on Apr 8, 2025 • 0 new comments

• Using split memory (RAM+VRAM) should never happen

  #10092 commented on Apr 8, 2025 • 0 new comments

• Provide an updated OpenAPI Specification file (a/k/a "swagger file") with each release

  #3383 commented on Apr 8, 2025 • 0 new comments

• 为什么mac中 gemma3工作不正常

  #9939 commented on Apr 8, 2025 • 0 new comments

• Isn't it time to move onto Omni models?

  #6786 commented on Apr 8, 2025 • 0 new comments

• Compute Capability 3.7 still needed

  #9620 commented on Apr 8, 2025 • 0 new comments

• Support logit_bias

  #3795 commented on Apr 7, 2025 • 0 new comments

• Add support for array for head count GGUF KV

  #9984 commented on Apr 7, 2025 • 0 new comments

• Ollama ps says 22 GB, but nvidia-smi says 16GB with flash attention enabled

  #6160 commented on Apr 7, 2025 • 0 new comments

• Ollama Bug Report: Application Launch Issue

  #9832 commented on Apr 7, 2025 • 0 new comments

• Error: POST predict: Post "http://127.0.0.1:62622/completion": read tcp 127.0.0.1:62627->127.0.0.1:62622: wsarecv: The remote host has closed a connection.

  #9674 commented on Apr 7, 2025 • 0 new comments

• add Qwen2-VL/Qwen2.5-VL

  #6564 commented on Apr 7, 2025 • 0 new comments

• llama_model_load_from_file_impl: failed to load model

  #9541 commented on Apr 7, 2025 • 0 new comments

• Feature request: support for OpenCL

  #4373 commented on Apr 7, 2025 • 0 new comments

• Update broken on Linux

  #10101 commented on Apr 7, 2025 • 0 new comments

• MacOS Ollama not binding to 0.0.0.0

  #3581 commented on Apr 6, 2025 • 0 new comments

• Support AMD GPUs on Intel Macs

  #1016 commented on Apr 6, 2025 • 0 new comments

• [ENHANCE] Add Ubuntu Support for AMD Ryzen AI 9 HX 370 w/ Radeon 890M (gfx1150)

  #9999 commented on Apr 6, 2025 • 0 new comments