ci: use windows-2022 to sign and bundle (#8941)

ollama requires vcruntime140_1.dll which isn't found on 2019. previously
the job used the windows runner (2019) but it explicitly installs
2022 to build the app. since the sign job doesn't actually build
anything, it can use the windows-2022 runner instead.

.gitattributes

@@ -15,6 +15,10 @@ ml/backend/**/*.cu linguist-vendored
 ml/backend/**/*.cuh linguist-vendored
 ml/backend/**/*.m linguist-vendored
 ml/backend/**/*.metal linguist-vendored
+ml/backend/**/CMakeLists.txt linguist-vendored
+
+llama/build-info.cpp linguist-generated
+ml/backend/ggml/ggml/src/ggml-metal/ggml-metal-embed.s linguist-generated
 
 * text=auto
 *.go text eol=lf

.github/ISSUE_TEMPLATE/10_bug_report.yml

@@ -9,6 +9,14 @@ body:
       description: What happened? What did you expect to happen?
     validations:
       required: true
+  - type: textarea
+    id: logs
+    attributes:
+      label: Relevant log output
+      description: Please copy and paste any relevant log output. See [Troubleshooting Guide](https://github.com/ollama/ollama/blob/main/docs/troubleshooting.md#how-to-troubleshoot-issues) for details.
+      render: shell
+    validations:
+      required: false
   - type: dropdown
     id: os
     attributes:

.github/workflows/release.yaml

@@ -242,7 +242,7 @@ jobs:
             dist\${{ matrix.os }}-${{ matrix.arch }}-app.exe
 
   windows-sign:
-    runs-on: windows
+    runs-on: windows-2022
     environment: release
     needs: [windows-depends, windows-build]
     steps:
@@ -303,59 +303,104 @@ jobs:
     steps:
       - uses: actions/checkout@v4
       - uses: docker/setup-buildx-action@v3
+      - uses: docker/build-push-action@v6
+        with:
+          context: .
+          platforms: ${{ matrix.os }}/${{ matrix.arch }}
+          target: ${{ matrix.target }}
+          build-args: |
+            GOFLAGS=${{ env.GOFLAGS }}
+            CGO_CFLAGS=${{ env.CGO_CFLAGS }}
+            CGO_CXXFLAGS=${{ env.CGO_CXXFLAGS }}
+          outputs: type=local,dest=dist/${{ matrix.os }}-${{ matrix.arch }}
+          cache-from: type=registry,ref=ollama/ollama:latest
+          cache-to: type=inline
       - run: |
-          sudo apt-get update && sudo apt-get install pigz
-          docker buildx build --platform $PLATFORM --target ${{ matrix.target }} --build-arg GOFLAGS --build-arg CGO_CFLAGS --build-arg CGO_CXXFLAGS --output type=local,dest=dist/$PLATFORM .
-
-          for COMPONENTS in dist/$PLATFORM/* dist/$PLATFORM/lib/ollama/*; do
-              if [ -d "$COMPONENTS" ]; then
-                  case "$COMPONENTS" in
-                      */bin) echo $COMPONENTS >>dist/ollama-${PLATFORM//\//-}.tar.in ;;
-                      */lib/ollama) echo $COMPONENTS >>dist/ollama-${PLATFORM//\//-}.tar.in;;
-                      */lib/ollama/cuda_v11) echo $COMPONENTS >>dist/ollama-${PLATFORM//\//-}.tar.in;;
-                      */lib/ollama/cuda_v12) echo $COMPONENTS >>dist/ollama-${PLATFORM//\//-}.tar.in;;
-                      */lib/ollama/cuda_jetpack5) echo $COMPONENTS >>dist/ollama-${PLATFORM//\//-}-jetpack5.tar.in ;;
-                      */lib/ollama/cuda_jetpack6) echo $COMPONENTS >>dist/ollama-${PLATFORM//\//-}-jetpack6.tar.in ;;
-                      */lib/ollama/rocm) echo $COMPONENTS >>dist/ollama-${PLATFORM//\//-}-rocm.tar.in ;;
-                  esac
-              fi
+          for COMPONENT in bin/* lib/ollama/*; do
+            case "$COMPONENT" in
+              bin/ollama)               echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
+              lib/ollama/*.so)          echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
+              lib/ollama/cuda_v11)      echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
+              lib/ollama/cuda_v12)      echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
+              lib/ollama/cuda_jetpack5) echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-jetpack5.tar.in ;;
+              lib/ollama/cuda_jetpack6) echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-jetpack6.tar.in ;;
+              lib/ollama/rocm)          echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-rocm.tar.in ;;
+            esac
           done
-
-          for ARCHIVE in dist/*.tar.in; do tar c -T $ARCHIVE --strip-components 3 | pigz -9cv >${ARCHIVE//.*/}.tgz; done
-        env:
-          PLATFORM: ${{ matrix.os }}/${{ matrix.arch }}
+        working-directory: dist/${{ matrix.os }}-${{ matrix.arch }}
+      - run: |
+          for ARCHIVE in dist/${{ matrix.os }}-${{ matrix.arch }}/*.tar.in; do tar c -C dist/${{ matrix.os }}-${{ matrix.arch }} -T $ARCHIVE | pigz -9vc >$(basename ${ARCHIVE//.*/}.tgz); done
       - uses: actions/upload-artifact@v4
         with:
-          name: dist-${{ matrix.os }}-${{ matrix.arch }}
+          name: dist-${{ matrix.os }}-${{ matrix.arch }}-${{ matrix.target }}
           path: |
-            dist/*.tgz
+            *.tgz
 
-  docker-build:
+  # Build each Docker variant (OS, arch, and flavor) separately. Using QEMU is unreliable and slower.
+  docker-build-push:
     strategy:
       matrix:
         include:
-          - flavor: 'latest=false'
-            platforms: linux/amd64,linux/arm64
+          - os: linux
+            arch: arm64
             build-args: |
               CGO_CFLAGS
               CGO_CXXFLAGS
               GOFLAGS
-          - flavor: 'latest=false,suffix=-rocm'
-            platforms: linux/amd64
+          - os: linux
+            arch: amd64
+            build-args: |
+              CGO_CFLAGS
+              CGO_CXXFLAGS
+              GOFLAGS
+          - os: linux
+            arch: amd64
+            suffix: '-rocm'
             build-args: |
               CGO_CFLAGS
               CGO_CXXFLAGS
               GOFLAGS
               FLAVOR=rocm
-    env:
-      GOFLAGS: ${{ needs.setup-environment.outputs.GOFLAGS }}
-    runs-on: linux
+    runs-on: ${{ matrix.arch == 'arm64' && format('{0}-{1}', matrix.os, matrix.arch) || matrix.os }}
     environment: release
     needs: setup-environment
+    env:
+      GOFLAGS: ${{ needs.setup-environment.outputs.GOFLAGS }}
     steps:
       - uses: actions/checkout@v4
-      - uses: docker/setup-qemu-action@v2
-      - uses: docker/setup-buildx-action@v2
+      - uses: docker/setup-buildx-action@v3
+      - uses: docker/login-action@v3
+        with:
+          username: ${{ vars.DOCKER_USER }}
+          password: ${{ secrets.DOCKER_ACCESS_TOKEN }}
+      - id: build-push
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          platforms: ${{ matrix.os }}/${{ matrix.arch }}
+          build-args: ${{ matrix.build-args }}
+          outputs: type=image,name=ollama/ollama,push-by-digest=true,name-canonical=true,push=true
+          cache-from: type=registry,ref=ollama/ollama:latest
+          cache-to: type=inline
+      - run: |
+          mkdir -p ${{ matrix.os }}-${{ matrix.arch }}
+          echo "${{ steps.build-push.outputs.digest }}" >${{ matrix.os }}-${{ matrix.arch }}-${{ matrix.suffix }}.txt
+        working-directory: ${{ runner.temp }}
+      - uses: actions/upload-artifact@v4
+        with:
+          name: digest-${{ matrix.os }}-${{ matrix.arch }}-${{ matrix.suffix }}
+          path: |
+            ${{ runner.temp }}/${{ matrix.os }}-${{ matrix.arch }}-${{ matrix.suffix }}.txt
+
+  # Merge Docker images for the same flavor into a single multi-arch manifest
+  docker-merge-push:
+    strategy:
+      matrix:
+        suffix: ['', '-rocm']
+    runs-on: linux
+    environment: release
+    needs: [docker-build-push]
+    steps:
       - uses: docker/login-action@v3
         with:
           username: ${{ vars.DOCKER_USER }}
@@ -363,22 +408,23 @@ jobs:
       - id: metadata
         uses: docker/metadata-action@v4
         with:
-          flavor: ${{ matrix.flavor }}
+          flavor: |
+            latest=false
+            suffix=${{ matrix.suffix }}
           images: |
             ollama/ollama
           tags: |
+            type=ref,enable=true,priority=600,prefix=pr-,event=pr
             type=semver,pattern={{version}}
-      - uses: docker/build-push-action@v6
+      - uses: actions/download-artifact@v4
         with:
-          context: .
-          push: true
-          platforms: ${{ matrix.platforms }}
-          build-args: ${{ matrix.build-args }}
-          tags: ${{ steps.metadata.outputs.tags }}
-          labels: ${{ steps.metadata.outputs.labels }}
-          cache-from: type=registry,ref=ollama/ollama:latest
-          cache-to: type=inline
-          provenance: false
+          pattern: digest-*
+          path: ${{ runner.temp }}
+          merge-multiple: true
+      - run: |
+          docker buildx imagetools create $(echo '${{ steps.metadata.outputs.json }}' | jq -cr '.tags | map("-t", .) | join(" ")') $(cat *-${{ matrix.suffix }}.txt | xargs printf 'ollama/ollama@%s ')
+          docker buildx imagetools inspect ollama/ollama:${{ steps.metadata.outputs.version }}
+        working-directory: ${{ runner.temp }}
 
   # Aggregate all the assets and ship a release
   release:
@@ -391,9 +437,6 @@ jobs:
       GH_TOKEN: ${{ github.token }}
     steps:
       - uses: actions/checkout@v4
-      - name: Set Version
-        shell: bash
-        run: |
       - uses: actions/download-artifact@v4
         with:
           name: dist-darwin
@@ -406,14 +449,12 @@ jobs:
         with:
           pattern: dist-linux-*
           path: dist
-      - run: |
-          ls -lh dist/
-          (cd dist; find . -type f | xargs sha256sum > ../sha256sum.txt)
-          mv sha256sum.txt dist/
-          cat dist/sha256sum.txt
+          merge-multiple: true
+      - run: find . -type f -not -name 'sha256sum.txt' | xargs sha256sum | tee sha256sum.txt
+        working-directory: dist
       - name: Create or update Release
         run: |
-          RELEASE_VERSION=$(echo ${GITHUB_REF_NAME} | cut -f1 -d-)"
+          RELEASE_VERSION="$(echo ${GITHUB_REF_NAME} | cut -f1 -d-)"
 
           echo "Looking for existing release for ${RELEASE_VERSION}"
           OLD_TAG=$(gh release ls --json name,tagName | jq -r ".[] | select(.name == \"${RELEASE_VERSION}\") | .tagName")

.github/workflows/test.yaml

@@ -163,5 +163,5 @@ jobs:
       - uses: actions/checkout@v4
       - name: Verify patches apply cleanly and do not change files
         run: |
-          make -f Makefile.sync clean checkout sync
+          make -f Makefile.sync clean sync
           git diff --compact-summary --exit-code

CMakeLists.txt

@@ -29,6 +29,11 @@ if((NOT CMAKE_OSX_ARCHITECTURES MATCHES "arm64")
     set(GGML_CPU_ALL_VARIANTS ON)
 endif()
 
+if (CMAKE_OSX_ARCHITECTURES MATCHES "x86_64")
+    set(CMAKE_BUILD_RPATH "@loader_path")
+    set(CMAKE_INSTALL_RPATH "@loader_path")
+endif()
+
 set(OLLAMA_BUILD_DIR ${CMAKE_BINARY_DIR}/lib/ollama)
 set(OLLAMA_INSTALL_DIR ${CMAKE_INSTALL_PREFIX}/lib/ollama)
 
@@ -80,6 +85,11 @@ if(CMAKE_CUDA_COMPILER)
     )
 endif()
 
+set(WINDOWS_AMDGPU_TARGETS_EXCLUDE_REGEX "^gfx(906|908|90a):xnack[+-]$"
+    CACHE STRING
+    "Regular expression describing AMDGPU_TARGETS not supported on Windows. Override to force building these targets. Default \"^gfx(906|908|90a):xnack[+-]$\"."
+)
+
 check_language(HIP)
 if(CMAKE_HIP_COMPILER)
     set(HIP_PLATFORM "amd")
@@ -87,15 +97,18 @@ if(CMAKE_HIP_COMPILER)
     find_package(hip REQUIRED)
     if(NOT AMDGPU_TARGETS)
         list(FILTER AMDGPU_TARGETS INCLUDE REGEX "^gfx(900|94[012]|101[02]|1030|110[012])$")
+    elseif(WIN32 AND WINDOWS_AMDGPU_TARGETS_EXCLUDE_REGEX)
+        list(FILTER AMDGPU_TARGETS EXCLUDE REGEX ${WINDOWS_AMDGPU_TARGETS_EXCLUDE_REGEX})
     endif()
 
     if(AMDGPU_TARGETS)
         add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src/ggml-hip)
+
         set(OLLAMA_HIP_INSTALL_DIR ${OLLAMA_INSTALL_DIR}/rocm)
         install(TARGETS ggml-hip
             RUNTIME_DEPENDENCIES
                 DIRECTORIES ${HIP_BIN_INSTALL_DIR} ${HIP_LIB_INSTALL_DIR}
-                PRE_INCLUDE_REGEXES amdhip64 hipblas rocblas amd_comgr hsa_runtime64 rocprofiler-register drm_amdgpu drm numa
+                PRE_INCLUDE_REGEXES hipblas rocblas amdhip64 rocsolver amd_comgr hsa-runtime64 rocsparse tinfo rocprofiler-register drm drm_amdgpu numa elf
                 PRE_EXCLUDE_REGEXES ".*"
                 POST_EXCLUDE_REGEXES "system32"
             RUNTIME DESTINATION ${OLLAMA_HIP_INSTALL_DIR} COMPONENT HIP

CMakePresets.json

@@ -56,7 +56,7 @@
       "name": "ROCm 6",
       "inherits": [ "ROCm" ],
       "cacheVariables": {
-        "AMDGPU_TARGETS": "gfx900;gfx940;gfx941;gfx942;gfx1010;gfx1012;gfx1030;gfx1100;gfx1101;gfx1102"
+        "AMDGPU_TARGETS": "gfx900;gfx940;gfx941;gfx942;gfx1010;gfx1012;gfx1030;gfx1100;gfx1101;gfx1102;gfx906:xnack-;gfx908:xnack-;gfx90a:xnack+;gfx90a:xnack-"
       }
     }
   ],

Makefile.sync

@@ -15,7 +15,11 @@ help:
 	@echo "    make -f $(lastword $(MAKEFILE_LIST)) clean sync"
 
 .PHONY: sync
-sync: llama/llama.cpp ml/backend/ggml/ggml apply-patches
+sync: llama/build-info.cpp llama/llama.cpp ml/backend/ggml/ggml apply-patches
+
+.PHONY: llama/build-info.cpp
+llama/build-info.cpp: llama/build-info.cpp.in
+	sed -e 's|@FETCH_HEAD@|$(FETCH_HEAD)|' $< > $@
 
 .PHONY: llama/llama.cpp
 llama/llama.cpp: llama/vendor/ apply-patches

README.md

@@ -18,7 +18,7 @@ Get up and running with large language models.
 
 ### Linux
 
-```
+```shell
 curl -fsSL https://ollama.com/install.sh | sh
 ```
 
@@ -42,7 +42,7 @@ The official [Ollama Docker image](https://hub.docker.com/r/ollama/ollama) `olla
 
 To run and chat with [Llama 3.2](https://ollama.com/library/llama3.2):
 
-```
+```shell
 ollama run llama3.2
 ```
 
@@ -54,6 +54,8 @@ Here are some example models that can be downloaded:
 
 | Model              | Parameters | Size  | Download                         |
 | ------------------ | ---------- | ----- | -------------------------------- |
+| DeepSeek-R1        | 7B         | 4.7GB | `ollama run deepseek-r1`         |
+| DeepSeek-R1        | 671B       | 404GB | `ollama run deepseek-r1:671b`    |
 | Llama 3.3          | 70B        | 43GB  | `ollama run llama3.3`            |
 | Llama 3.2          | 3B         | 2.0GB | `ollama run llama3.2`            |
 | Llama 3.2          | 1B         | 1.3GB | `ollama run llama3.2:1b`         |
@@ -92,13 +94,13 @@ Ollama supports importing GGUF models in the Modelfile:
 
 2. Create the model in Ollama
 
-   ```
+   ```shell
    ollama create example -f Modelfile
    ```
 
 3. Run the model
 
-   ```
+   ```shell
    ollama run example
    ```
 
@@ -110,7 +112,7 @@ See the [guide](docs/import.md) on importing models for more information.
 
 Models from the Ollama library can be customized with a prompt. For example, to customize the `llama3.2` model:
 
-```
+```shell
 ollama pull llama3.2
 ```
 
@@ -145,13 +147,13 @@ For more information on working with a Modelfile, see the [Modelfile](docs/model
 
 `ollama create` is used to create a model from a Modelfile.
 
-```
+```shell
 ollama create mymodel -f ./Modelfile
 ```
 
 ### Pull a model
 
-```
+```shell
 ollama pull llama3.2
 ```
 
@@ -159,13 +161,13 @@ ollama pull llama3.2
 
 ### Remove a model
 
-```
+```shell
 ollama rm llama3.2
 ```
 
 ### Copy a model
 
-```
+```shell
 ollama cp llama3.2 my-model
 ```
 
@@ -184,37 +186,39 @@ I'm a basic program that prints the famous "Hello, world!" message to the consol
 
 ```
 ollama run llava "What's in this image? /Users/jmorgan/Desktop/smile.png"
-The image features a yellow smiley face, which is likely the central focus of the picture.
 ```
 
+> **Output**: The image features a yellow smiley face, which is likely the central focus of the picture.
+
 ### Pass the prompt as an argument
 
+```shell
+ollama run llama3.2 "Summarize this file: $(cat README.md)"
 ```
-$ ollama run llama3.2 "Summarize this file: $(cat README.md)"
- Ollama is a lightweight, extensible framework for building and running language models on the local machine. It provides a simple API for creating, running, and managing models, as well as a library of pre-built models that can be easily used in a variety of applications.
-```
+
+> **Output**: Ollama is a lightweight, extensible framework for building and running language models on the local machine. It provides a simple API for creating, running, and managing models, as well as a library of pre-built models that can be easily used in a variety of applications.
 
 ### Show model information
 
-```
+```shell
 ollama show llama3.2
 ```
 
 ### List models on your computer
 
-```
+```shell
 ollama list
 ```
 
 ### List which models are currently loaded
 
-```
+```shell
 ollama ps
 ```
 
 ### Stop a model which is currently running
 
-```
+```shell
 ollama stop llama3.2
 ```
 
@@ -230,13 +234,13 @@ See the [developer guide](https://github.com/ollama/ollama/blob/main/docs/develo
 
 Next, start the server:
 
-```
+```shell
 ./ollama serve
 ```
 
 Finally, in a separate shell, run a model:
 
-```
+```shell
 ./ollama run llama3.2
 ```
 
@@ -246,7 +250,7 @@ Ollama has a REST API for running and managing models.
 
 ### Generate a response
 
-```
+```shell
 curl http://localhost:11434/api/generate -d '{
   "model": "llama3.2",
   "prompt":"Why is the sky blue?"
@@ -255,7 +259,7 @@ curl http://localhost:11434/api/generate -d '{
 
 ### Chat with a model
 
-```
+```shell
 curl http://localhost:11434/api/chat -d '{
   "model": "llama3.2",
   "messages": [
@@ -353,6 +357,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Web management](https://github.com/lemonit-eric-mao/ollama-web-management) (Web management page)
 - [Promptery](https://github.com/promptery/promptery) (desktop client for Ollama.)
 - [Ollama App](https://github.com/JHubi1/ollama-app) (Modern and easy-to-use multi-platform client for Ollama)
+- [chat-ollama](https://github.com/annilq/chat-ollama) (a React Native client for Ollama)
 - [SpaceLlama](https://github.com/tcsenpai/spacellama) (Firefox and Chrome extension to quickly summarize web pages with ollama in a sidebar)
 - [YouLama](https://github.com/tcsenpai/youlama) (Webapp to quickly summarize any YouTube video, supporting Invidious as well)
 - [DualMind](https://github.com/tcsenpai/dualmind) (Experimental app allowing two models to talk to each other in the terminal or in a web interface)
@@ -369,9 +374,12 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Minima](https://github.com/dmayboroda/minima) (RAG with on-premises or fully local workflow)
 - [aidful-ollama-model-delete](https://github.com/AidfulAI/aidful-ollama-model-delete) (User interface for simplified model cleanup)
 - [Perplexica](https://github.com/ItzCrazyKns/Perplexica) (An AI-powered search engine & an open-source alternative to Perplexity AI)
+- [Ollama Chat WebUI for Docker ](https://github.com/oslook/ollama-webui) (Support for local docker deployment, lightweight ollama webui)
 - [AI Toolkit for Visual Studio Code](https://aka.ms/ai-tooklit/ollama-docs) (Microsoft-official VSCode extension to chat, test, evaluate models with Ollama support, and use them in your AI applications.)
 - [MinimalNextOllamaChat](https://github.com/anilkay/MinimalNextOllamaChat) (Minimal Web UI for Chat and Model Control)
 - [Chipper](https://github.com/TilmanGriesel/chipper) AI interface for tinkerers (Ollama, Haystack RAG, Python)
+- [ChibiChat](https://github.com/CosmicEventHorizon/ChibiChat) (Kotlin-based Android app to chat with Ollama and Koboldcpp API endpoints)
+- [LocalLLM](https://github.com/qusaismael/localllm) (Minimal Web-App to run ollama models on it with a GUI)
 
 ### Cloud
 
@@ -485,6 +493,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Ollama for Haskell](https://github.com/tusharad/ollama-haskell)
 - [multi-llm-ts](https://github.com/nbonamy/multi-llm-ts) (A Typescript/JavaScript library allowing access to different LLM in unified API)
 - [LlmTornado](https://github.com/lofcz/llmtornado) (C# library providing a unified interface for major FOSS & Commercial inference APIs)
+- [Ollama for Zig](https://github.com/dravenk/ollama-zig)
 
 ### Mobile
 
@@ -535,6 +544,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [TextCraft](https://github.com/suncloudsmoon/TextCraft) (Copilot in Word alternative using Ollama)
 - [Alfred Ollama](https://github.com/zeitlings/alfred-ollama) (Alfred Workflow)
 - [TextLLaMA](https://github.com/adarshM84/TextLLaMA) A Chrome Extension that helps you write emails, correct grammar, and translate into any language
+- [Simple-Discord-AI](https://github.com/zyphixor/simple-discord-ai)
 
 ### Supported backends
 
@@ -545,3 +555,4 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [OpenLIT](https://github.com/openlit/openlit) is an OpenTelemetry-native tool for monitoring Ollama Applications & GPUs using traces and metrics.
 - [HoneyHive](https://docs.honeyhive.ai/integrations/ollama) is an AI observability and evaluation platform for AI agents. Use HoneyHive to evaluate agent performance, interrogate failures, and monitor quality in production.
 - [Langfuse](https://langfuse.com/docs/integrations/ollama) is an open source LLM observability platform that enables teams to collaboratively monitor, evaluate and debug AI applications.
+- [MLflow Tracing](https://mlflow.org/docs/latest/llms/tracing/index.html#automatic-tracing) is an open source LLM observability tool with a convenient API to log and visualize traces, making it easy to debug and evaluate GenAI applications.

api/examples/README.md

@@ -2,9 +2,10 @@
 
 Run the examples in this directory with:
 
-```
+```shell
 go run example_name/main.go
 ```
+
 ## Chat - Chat with a model
 - [chat/main.go](chat/main.go)
 

app/README.md

@@ -17,6 +17,6 @@ If you want to build the installer, youll need to install
 In the top directory of this repo, run the following powershell script
 to build the ollama CLI, ollama app, and ollama installer.
 
-```
+```powershell
 powershell -ExecutionPolicy Bypass -File .\scripts\build_windows.ps1
 ```

docs/api.md

@@ -31,7 +31,7 @@ Certain endpoints stream responses as JSON objects. Streaming can be disabled by
 
 ## Generate a completion
 
-```shell
+```
 POST /api/generate
 ```
 
@@ -485,7 +485,7 @@ A single JSON object is returned:
 
 ## Generate a chat completion
 
-```shell
+```
 POST /api/chat
 ```
 
@@ -878,6 +878,7 @@ curl http://localhost:11434/api/chat -d '{
 ```
 
 ##### Response
+
 ```json
 {
   "model": "llama3.2",
@@ -924,7 +925,7 @@ A single JSON object is returned:
 
 ## Create a Model
 
-```shell
+```
 POST /api/create
 ```
 
@@ -1020,7 +1021,7 @@ curl http://localhost:11434/api/create -d '{
 
 A stream of JSON objects is returned:
 
-```
+```json
 {"status":"quantizing F16 model to Q4_K_M"}
 {"status":"creating new layer sha256:667b0c1932bc6ffc593ed1d03f895bf2dc8dc6df21db3042284a6f4416b06a29"}
 {"status":"using existing layer sha256:11ce4ee3e170f6adebac9a991c22e22ab3f8530e154ee669954c4bc73061c258"}
@@ -1051,7 +1052,7 @@ curl http://localhost:11434/api/create -d '{
 
 A stream of JSON objects is returned:
 
-```
+```json
 {"status":"parsing GGUF"}
 {"status":"using existing layer sha256:432f310a77f4650a88d0fd59ecdd7cebed8d684bafea53cbff0473542964f0c3"}
 {"status":"writing manifest"}
@@ -1118,7 +1119,7 @@ Return 200 OK if the blob exists, 404 Not Found if it does not.
 
 ## Push a Blob
 
-```shell
+```
 POST /api/blobs/:digest
 ```
 
@@ -1142,7 +1143,7 @@ Return 201 Created if the blob was successfully created, 400 Bad Request if the
 
 ## List Local Models
 
-```shell
+```
 GET /api/tags
 ```
 
@@ -1195,7 +1196,7 @@ A single JSON object will be returned.
 
 ## Show Model Information
 
-```shell
+```
 POST /api/show
 ```
 
@@ -1261,7 +1262,7 @@ curl http://localhost:11434/api/show -d '{
 
 ## Copy a Model
 
-```shell
+```
 POST /api/copy
 ```
 
@@ -1284,7 +1285,7 @@ Returns a 200 OK if successful, or a 404 Not Found if the source model doesn't e
 
 ## Delete a Model
 
-```shell
+```
 DELETE /api/delete
 ```
 
@@ -1310,7 +1311,7 @@ Returns a 200 OK if successful, 404 Not Found if the model to be deleted doesn't
 
 ## Pull a Model
 
-```shell
+```
 POST /api/pull
 ```
 
@@ -1382,7 +1383,7 @@ if `stream` is set to false, then the response is a single JSON object:
 
 ## Push a Model
 
-```shell
+```
 POST /api/push
 ```
 
@@ -1447,7 +1448,7 @@ If `stream` is set to `false`, then the response is a single JSON object:
 
 ## Generate Embeddings
 
-```shell
+```
 POST /api/embed
 ```
 
@@ -1515,7 +1516,7 @@ curl http://localhost:11434/api/embed -d '{
 ```
 
 ## List Running Models
-```shell
+```
 GET /api/ps
 ```
 
@@ -1562,7 +1563,7 @@ A single JSON object will be returned.
 
 > Note: this endpoint has been superseded by `/api/embed`
 
-```shell
+```
 POST /api/embeddings
 ```
 
@@ -1602,7 +1603,7 @@ curl http://localhost:11434/api/embeddings -d '{
 
 ## Version
 
-```shell
+```
 GET /api/version
 ```
 

docs/development.md

@@ -3,11 +3,11 @@
 Install prerequisites:
 
 - [Go](https://go.dev/doc/install)
-- C/C++ Compiler e.g. Clang on macOS, [TDM-GCC](https://jmeubank.github.io/tdm-gcc/download/) (Windows amd64) or [llvm-mingw](https://github.com/mstorsjo/llvm-mingw) (Windows arm64), GCC/Clang on Linux.
+- C/C++ Compiler e.g. Clang on macOS, [TDM-GCC](https://github.com/jmeubank/tdm-gcc/releases/latest) (Windows amd64) or [llvm-mingw](https://github.com/mstorsjo/llvm-mingw) (Windows arm64), GCC/Clang on Linux.
 
 Then build and run Ollama from the root directory of the repository:
 
-```
+```shell
 go run . serve
 ```
 
@@ -23,14 +23,14 @@ Install prerequisites:
 
 Then, configure and build the project:
 
-```
+```shell
 cmake -B build
 cmake --build build
 ```
 
 Lastly, run Ollama:
 
-```
+```shell
 go run . serve
 ```
 
@@ -57,14 +57,14 @@ Install prerequisites:
 
 Then, configure and build the project:
 
-```
+```shell
 cmake -B build
 cmake --build build --config Release
 ```
 
 Lastly, run Ollama:
 
-```
+```shell
 go run . serve
 ```
 
@@ -88,26 +88,26 @@ Install prerequisites:
 
 Then, configure and build the project:
 
-```
+```shell
 cmake -B build
 cmake --build build
 ```
 
 Lastly, run Ollama:
 
-```
+```shell
 go run . serve
 ```
 
 ## Docker
 
-```
+```shell
 docker build .
 ```
 
 ### ROCm
 
-```
+```shell
 docker build --build-arg FLAVOR=rocm .
 ```
 
@@ -115,6 +115,17 @@ docker build --build-arg FLAVOR=rocm .
 
 To run tests, use `go test`:
 
-```
+```shell
 go test ./...
 ```
+
+## Library detection
+
+Ollama looks for acceleration libraries in the following paths relative to the `ollama` executable:
+
+* `./lib/ollama` (Windows)
+* `../lib/ollama` (Linux)
+* `.` (macOS)
+* `build/lib/ollama` (for development)
+
+If the libraries are not found, Ollama will not run with any acceleration libraries.

docs/docker.md

@@ -2,7 +2,7 @@
 
 ### CPU only
 
-```bash
+```shell
 docker run -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama
 ```
 
@@ -11,42 +11,46 @@ Install the [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-
 
 #### Install with Apt
 1.  Configure the repository
-```bash
-curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey \
-    | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
-curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list \
-    | sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' \
-    | sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list
-sudo apt-get update
-```
+
+    ```shell
+    curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey \
+        | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
+    curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list \
+        | sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' \
+        | sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list
+    sudo apt-get update
+    ```
+
 2.  Install the NVIDIA Container Toolkit packages
-```bash
-sudo apt-get install -y nvidia-container-toolkit
-```
+
+    ```shell
+    sudo apt-get install -y nvidia-container-toolkit
+    ```
 
 #### Install with Yum or Dnf
 1.  Configure the repository
 
-```bash
-curl -s -L https://nvidia.github.io/libnvidia-container/stable/rpm/nvidia-container-toolkit.repo \
-    | sudo tee /etc/yum.repos.d/nvidia-container-toolkit.repo
-```
+    ```shell
+    curl -s -L https://nvidia.github.io/libnvidia-container/stable/rpm/nvidia-container-toolkit.repo \
+        | sudo tee /etc/yum.repos.d/nvidia-container-toolkit.repo
+    ```
 
 2. Install the NVIDIA Container Toolkit packages
 
-```bash
-sudo yum install -y nvidia-container-toolkit
-```
+    ```shell
+    sudo yum install -y nvidia-container-toolkit
+    ```
 
 #### Configure Docker to use Nvidia driver
-```
+
+```shell
 sudo nvidia-ctk runtime configure --runtime=docker
 sudo systemctl restart docker
 ```
 
 #### Start the container
 
-```bash
+```shell
 docker run -d --gpus=all -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama
 ```
 
@@ -57,7 +61,7 @@ docker run -d --gpus=all -v ollama:/root/.ollama -p 11434:11434 --name ollama ol
 
 To run Ollama using Docker with AMD GPUs, use the `rocm` tag and the following command:
 
-```
+```shell
 docker run -d --device /dev/kfd --device /dev/dri -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama:rocm
 ```
 
@@ -65,7 +69,7 @@ docker run -d --device /dev/kfd --device /dev/dri -v ollama:/root/.ollama -p 114
 
 Now you can run a model:
 
-```
+```shell
 docker exec -it ollama ollama run llama3.2
 ```
 

docs/faq.md

@@ -24,7 +24,7 @@ By default, Ollama uses a context window size of 2048 tokens.
 
 To change this when using `ollama run`, use `/set parameter`:
 
-```
+```shell
 /set parameter num_ctx 4096
 ```
 
@@ -46,10 +46,15 @@ Use the `ollama ps` command to see what models are currently loaded into memory.
 
 ```shell
 ollama ps
-NAME      	ID          	SIZE 	PROCESSOR	UNTIL
-llama3:70b	bcfb190ca3a7	42 GB	100% GPU 	4 minutes from now
 ```
 
+> **Output**:
+>
+> ```
+> NAME      	ID          	SIZE 	PROCESSOR	UNTIL
+> llama3:70b	bcfb190ca3a7	42 GB	100% GPU 	4 minutes from now
+> ```
+
 The `Processor` column will show which memory the model was loaded in to:
 * `100% GPU` means the model was loaded entirely into the GPU
 * `100% CPU` means the model was loaded entirely in system memory
@@ -66,7 +71,7 @@ If Ollama is run as a macOS application, environment variables should be set usi
 1. For each environment variable, call `launchctl setenv`.
 
     ```bash
-    launchctl setenv OLLAMA_HOST "0.0.0.0"
+    launchctl setenv OLLAMA_HOST "0.0.0.0:11434"
     ```
 
 2. Restart Ollama application.
@@ -81,14 +86,14 @@ If Ollama is run as a systemd service, environment variables should be set using
 
     ```ini
     [Service]
-    Environment="OLLAMA_HOST=0.0.0.0"
+    Environment="OLLAMA_HOST=0.0.0.0:11434"
     ```
 
 3. Save and exit.
 
 4. Reload `systemd` and restart Ollama:
 
-   ```bash
+   ```shell
    systemctl daemon-reload
    systemctl restart ollama
    ```
@@ -221,16 +226,19 @@ properties.
 If you are using the API you can preload a model by sending the Ollama server an empty request. This works with both the `/api/generate` and `/api/chat` API endpoints.
 
 To preload the mistral model using the generate endpoint, use:
+
 ```shell
 curl http://localhost:11434/api/generate -d '{"model": "mistral"}'
 ```
 
 To use the chat completions endpoint, use:
+
 ```shell
 curl http://localhost:11434/api/chat -d '{"model": "mistral"}'
 ```
 
 To preload a model using the CLI, use the command:
+
 ```shell
 ollama run llama3.2 ""
 ```
@@ -250,11 +258,13 @@ If you're using the API, use the `keep_alive` parameter with the `/api/generate`
 * '0' which will unload the model immediately after generating a response
 
 For example, to preload a model and leave it in memory use:
+
 ```shell
 curl http://localhost:11434/api/generate -d '{"model": "llama3.2", "keep_alive": -1}'
 ```
 
 To unload the model and free up memory use:
+
 ```shell
 curl http://localhost:11434/api/generate -d '{"model": "llama3.2", "keep_alive": 0}'
 ```

docs/import.md

@@ -20,13 +20,13 @@ Make sure that you use the same base model in the `FROM` command as you used to
 
 Now run `ollama create` from the directory where the `Modelfile` was created:
 
-```bash
+```shell
 ollama create my-model
 ```
 
 Lastly, test the model:
 
-```bash
+```shell
 ollama run my-model
 ```
 

docs/linux.md

@@ -119,7 +119,7 @@ sudo systemctl status ollama
 
 To customize the installation of Ollama, you can edit the systemd service file or the environment variables by running:
 
-```
+```shell
 sudo systemctl edit ollama
 ```
 
@@ -186,3 +186,9 @@ sudo rm -r /usr/share/ollama
 sudo userdel ollama
 sudo groupdel ollama
 ```
+
+Remove installed libraries:
+
+```shell
+sudo rm -rf /usr/local/lib/ollama
+```

docs/modelfile.md

@@ -28,7 +28,7 @@ A model file is the blueprint to create and share models with Ollama.
 
 The format of the `Modelfile`:
 
-```modelfile
+```
 # comment
 INSTRUCTION arguments
 ```
@@ -49,7 +49,7 @@ INSTRUCTION arguments
 
 An example of a `Modelfile` creating a mario blueprint:
 
-```modelfile
+```
 FROM llama3.2
 # sets the temperature to 1 [higher is more creative, lower is more coherent]
 PARAMETER temperature 1
@@ -69,24 +69,30 @@ To use this:
 
 To view the Modelfile of a given model, use the `ollama show --modelfile` command.
 
-  ```bash
-  > ollama show --modelfile llama3.2
-  # Modelfile generated by "ollama show"
-  # To build a new Modelfile based on this one, replace the FROM line with:
-  # FROM llama3.2:latest
-  FROM /Users/pdevine/.ollama/models/blobs/sha256-00e1317cbf74d901080d7100f57580ba8dd8de57203072dc6f668324ba545f29
-  TEMPLATE """{{ if .System }}<|start_header_id|>system<|end_header_id|>
+```shell
+ollama show --modelfile llama3.2
+```
 
-  {{ .System }}<|eot_id|>{{ end }}{{ if .Prompt }}<|start_header_id|>user<|end_header_id|>
+> **Output**:
+>
+> ```
+> # Modelfile generated by "ollama show"
+> # To build a new Modelfile based on this one, replace the FROM line with:
+> # FROM llama3.2:latest
+> FROM /Users/pdevine/.ollama/models/blobs/sha256-00e1317cbf74d901080d7100f57580ba8dd8de57203072dc6f668324ba545f29
+> TEMPLATE """{{ if .System }}<|start_header_id|>system<|end_header_id|>
+>
+> {{ .System }}<|eot_id|>{{ end }}{{ if .Prompt }}<|start_header_id|>user<|end_header_id|>
+>
+> {{ .Prompt }}<|eot_id|>{{ end }}<|start_header_id|>assistant<|end_header_id|>
+>
+> {{ .Response }}<|eot_id|>"""
+> PARAMETER stop "<|start_header_id|>"
+> PARAMETER stop "<|end_header_id|>"
+> PARAMETER stop "<|eot_id|>"
+> PARAMETER stop "<|reserved_special_token"
+> ```
 
-  {{ .Prompt }}<|eot_id|>{{ end }}<|start_header_id|>assistant<|end_header_id|>
-
-  {{ .Response }}<|eot_id|>"""
-  PARAMETER stop "<|start_header_id|>"
-  PARAMETER stop "<|end_header_id|>"
-  PARAMETER stop "<|eot_id|>"
-  PARAMETER stop "<|reserved_special_token"
-  ```
 
 ## Instructions
 
@@ -94,13 +100,13 @@ To view the Modelfile of a given model, use the `ollama show --modelfile` comman
 
 The `FROM` instruction defines the base model to use when creating a model.
 
-```modelfile
+```
 FROM <model name>:<tag>
 ```
 
 #### Build from existing model
 
-```modelfile
+```
 FROM llama3.2
 ```
 
@@ -111,7 +117,7 @@ Additional models can be found at:
 
 #### Build from a Safetensors model
 
-```modelfile
+```
 FROM <model directory>
 ```
 
@@ -125,7 +131,7 @@ Currently supported model architectures:
 
 #### Build from a GGUF file
 
-```modelfile
+```
 FROM ./ollama-model.gguf
 ```
 
@@ -136,7 +142,7 @@ The GGUF file location should be specified as an absolute path or relative to th
 
 The `PARAMETER` instruction defines a parameter that can be set when the model is run.
 
-```modelfile
+```
 PARAMETER <parameter> <parametervalue>
 ```
 
@@ -183,7 +189,7 @@ TEMPLATE """{{ if .System }}<|im_start|>system
 
 The `SYSTEM` instruction specifies the system message to be used in the template, if applicable.
 
-```modelfile
+```
 SYSTEM """<system message>"""
 ```
 
@@ -193,7 +199,7 @@ The `ADAPTER` instruction specifies a fine tuned LoRA adapter that should apply
 
 #### Safetensor adapter
 
-```modelfile
+```
 ADAPTER <path to safetensor adapter>
 ```
 
@@ -204,7 +210,7 @@ Currently supported Safetensor adapters:
 
 #### GGUF adapter
 
-```modelfile
+```
 ADAPTER ./ollama-lora.gguf
 ```
 
@@ -212,7 +218,7 @@ ADAPTER ./ollama-lora.gguf
 
 The `LICENSE` instruction allows you to specify the legal license under which the model used with this Modelfile is shared or distributed.
 
-```modelfile
+```
 LICENSE """
 <license text>
 """
@@ -222,7 +228,7 @@ LICENSE """
 
 The `MESSAGE` instruction allows you to specify a message history for the model to use when responding. Use multiple iterations of the MESSAGE command to build up a conversation which will guide the model to answer in a similar way.
 
-```modelfile
+```
 MESSAGE <role> <message>
 ```
 
@@ -237,7 +243,7 @@ MESSAGE <role> <message>
 
 #### Example conversation
 
-```modelfile
+```
 MESSAGE user Is Toronto in Canada?
 MESSAGE assistant yes
 MESSAGE user Is Sacramento in Canada?

docs/openai.md

@@ -1,6 +1,7 @@
 # OpenAI compatibility
 
-> **Note:** OpenAI compatibility is experimental and is subject to major adjustments including breaking changes. For fully-featured access to the Ollama API, see the Ollama [Python library](https://github.com/ollama/ollama-python), [JavaScript library](https://github.com/ollama/ollama-js) and [REST API](https://github.com/ollama/ollama/blob/main/docs/api.md).
+> [!NOTE]
+> OpenAI compatibility is experimental and is subject to major adjustments including breaking changes. For fully-featured access to the Ollama API, see the Ollama [Python library](https://github.com/ollama/ollama-python), [JavaScript library](https://github.com/ollama/ollama-js) and [REST API](https://github.com/ollama/ollama/blob/main/docs/api.md).
 
 Ollama provides experimental compatibility with parts of the [OpenAI API](https://platform.openai.com/docs/api-reference) to help connect existing applications to Ollama.
 
@@ -59,8 +60,10 @@ embeddings = client.embeddings.create(
     input=["why is the sky blue?", "why is the grass green?"],
 )
 ```
+
 #### Structured outputs
-```py
+
+```python
 from pydantic import BaseModel
 from openai import OpenAI
 
@@ -144,7 +147,7 @@ const embedding = await openai.embeddings.create({
 
 ### `curl`
 
-``` shell
+```shell
 curl http://localhost:11434/v1/chat/completions \
     -H "Content-Type: application/json" \
     -d '{
@@ -319,7 +322,7 @@ ollama pull llama3.2
 
 For tooling that relies on default OpenAI model names such as `gpt-3.5-turbo`, use `ollama cp` to copy an existing model name to a temporary name:
 
-```
+```shell
 ollama cp llama3.2 gpt-3.5-turbo
 ```
 
@@ -343,7 +346,7 @@ curl http://localhost:11434/v1/chat/completions \
 
 The OpenAI API does not have a way of setting the context size for a model. If you need to change the context size, create a `Modelfile` which looks like:
 
-```modelfile
+```
 FROM <some model>
 PARAMETER num_ctx <context size>
 ```

docs/troubleshooting.md

@@ -17,6 +17,7 @@ When you run Ollama in a **container**, the logs go to stdout/stderr in the cont
 ```shell
 docker logs <container-name>
 ```
+
 (Use `docker ps` to find the container name)
 
 If manually running `ollama serve` in a terminal, the logs will be on that terminal.
@@ -28,6 +29,7 @@ When you run Ollama on **Windows**, there are a few different locations. You can
 - `explorer %TEMP%` where temporary executable files are stored in one or more `ollama*` directories
 
 To enable additional debug logging to help troubleshoot problems, first **Quit the running app from the tray menu** then in a powershell terminal
+
 ```powershell
 $env:OLLAMA_DEBUG="1"
 & "ollama app.exe"
@@ -49,12 +51,13 @@ Dynamic LLM libraries [rocm_v6 cpu cpu_avx cpu_avx2 cuda_v11 rocm_v5]
 
 You can set OLLAMA_LLM_LIBRARY to any of the available LLM libraries to bypass autodetection, so for example, if you have a CUDA card, but want to force the CPU LLM library with AVX2 vector support, use:
 
-```
+```shell
 OLLAMA_LLM_LIBRARY="cpu_avx2" ollama serve
 ```
 
 You can see what features your CPU has with the following.
-```
+
+```shell
 cat /proc/cpuinfo| grep flags | head -1
 ```
 
@@ -62,8 +65,8 @@ cat /proc/cpuinfo| grep flags | head -1
 
 If you run into problems on Linux and want to install an older version, or you'd like to try out a pre-release before it's officially released, you can tell the install script which version to install.
 
-```sh
-curl -fsSL https://ollama.com/install.sh | OLLAMA_VERSION="0.1.29" sh
+```shell
+curl -fsSL https://ollama.com/install.sh | OLLAMA_VERSION=0.5.7 sh
 ```
 
 ## Linux tmp noexec 

docs/windows.md

@@ -47,6 +47,7 @@ If Ollama is already running, Quit the tray application and relaunch it from the
 ## API Access
 
 Here's a quick example showing API access from `powershell`
+
 ```powershell
 (Invoke-WebRequest -method POST -Body '{"model":"llama3.2", "prompt":"Why is the sky blue?", "stream": false}' -uri http://localhost:11434/api/generate ).Content | ConvertFrom-json
 ```

format/bytes.go

@@ -40,8 +40,6 @@ func HumanBytes(b int64) string {
 	}
 
 	switch {
-	case value >= 100:
-		return fmt.Sprintf("%d %s", int(value), unit)
 	case value >= 10:
 		return fmt.Sprintf("%d %s", int(value), unit)
 	case value != math.Trunc(value):

format/bytes_test.go

@@ -0,0 +1,91 @@
+package format
+
+import (
+	"testing"
+)
+
+func TestHumanBytes(t *testing.T) {
+	type testCase struct {
+		input    int64
+		expected string
+	}
+
+	tests := []testCase{
+		// Test bytes (B)
+		{0, "0 B"},
+		{1, "1 B"},
+		{999, "999 B"},
+
+		// Test kilobytes (KB)
+		{1000, "1 KB"},
+		{1500, "1.5 KB"},
+		{999999, "999 KB"},
+
+		// Test megabytes (MB)
+		{1000000, "1 MB"},
+		{1500000, "1.5 MB"},
+		{999999999, "999 MB"},
+
+		// Test gigabytes (GB)
+		{1000000000, "1 GB"},
+		{1500000000, "1.5 GB"},
+		{999999999999, "999 GB"},
+
+		// Test terabytes (TB)
+		{1000000000000, "1 TB"},
+		{1500000000000, "1.5 TB"},
+		{1999999999999, "2.0 TB"},
+
+		// Test fractional values
+		{1234, "1.2 KB"},
+		{1234567, "1.2 MB"},
+		{1234567890, "1.2 GB"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.expected, func(t *testing.T) {
+			result := HumanBytes(tc.input)
+			if result != tc.expected {
+				t.Errorf("Expected %s, got %s", tc.expected, result)
+			}
+		})
+	}
+}
+
+func TestHumanBytes2(t *testing.T) {
+	type testCase struct {
+		input    uint64
+		expected string
+	}
+
+	tests := []testCase{
+		// Test bytes (B)
+		{0, "0 B"},
+		{1, "1 B"},
+		{1023, "1023 B"},
+
+		// Test kibibytes (KiB)
+		{1024, "1.0 KiB"},
+		{1536, "1.5 KiB"},
+		{1048575, "1024.0 KiB"},
+
+		// Test mebibytes (MiB)
+		{1048576, "1.0 MiB"},
+		{1572864, "1.5 MiB"},
+		{1073741823, "1024.0 MiB"},
+
+		// Test gibibytes (GiB)
+		{1073741824, "1.0 GiB"},
+		{1610612736, "1.5 GiB"},
+		{2147483648, "2.0 GiB"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.expected, func(t *testing.T) {
+			result := HumanBytes2(tc.input)
+			if result != tc.expected {
+				t.Errorf("Expected %s, got %s", tc.expected, result)
+			}
+		})
+	}
+}

llama/README.md

@@ -8,7 +8,7 @@ Ollama vendors [llama.cpp](https://github.com/ggerganov/llama.cpp/) and [ggml](h
 
 If you update the vendoring code, start by running the following command to establish the tracking llama.cpp repo in the `./vendor/` directory.
 
-```
+```shell
 make -f Makefile.sync apply-patches
 ```
 
@@ -22,7 +22,7 @@ When updating to a newer base commit, the existing patches may not apply cleanly
 
 Start by applying the patches. If any of the patches have conflicts, the `git am` will stop at the first failure.
 
-```
+```shell
 make -f Makefile.sync apply-patches
 ```
 
@@ -30,7 +30,7 @@ If there are conflicts, you will see an error message. Resolve the conflicts in
 
 Once all patches are applied, commit the changes to the tracking repository.
 
-```
+```shell
 make -f Makefile.sync format-patches sync
 ```
 
@@ -38,13 +38,13 @@ make -f Makefile.sync format-patches sync
 
 When working on new fixes or features that impact vendored code, use the following model. First get a clean tracking repo with all current patches applied:
 
-```
+```shell
 make -f Makefile.sync clean apply-patches
 ```
 
 Iterate until you're ready to submit PRs. Once your code is ready, commit a change in the `./vendor/` directory, then generate the patches for ollama with
 
-```
+```shell
 make -f Makefile.sync format-patches
 ```
 

llama/build-info.cpp

@@ -1,4 +1,4 @@
 int LLAMA_BUILD_NUMBER = 0;
-char const *LLAMA_COMMIT = "ba1cb19cdd0d92e012e0f6e009e0620f854b6afd";
+char const *LLAMA_COMMIT = "46e3556e01b824e52395fb050b29804b6cff2a7c";
 char const *LLAMA_COMPILER = "";
 char const *LLAMA_BUILD_TARGET = "";

llama/build-info.cpp.in

@@ -0,0 +1,4 @@
+int LLAMA_BUILD_NUMBER = 0;
+char const *LLAMA_COMMIT = "@FETCH_HEAD@";
+char const *LLAMA_COMPILER = "";
+char const *LLAMA_BUILD_TARGET = "";

llama/llama.cpp/examples/llava/clip.cpp

@@ -1235,35 +1235,15 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
         }
     }
 
-#ifdef GGML_USE_CUDA
-   new_clip->backend = ggml_backend_cuda_init(0);
-   LOG_INF("%s: CLIP using CUDA backend\n", __func__);
-#endif
-
-#ifdef GGML_USE_METAL
-   new_clip->backend = ggml_backend_metal_init();
-   LOG_INF("%s: CLIP using Metal backend\n", __func__);
-#endif
-
-#ifdef GGML_USE_CANN
-   new_clip->backend = ggml_backend_cann_init(0);
-   LOG_INF("%s: CLIP using CANN backend\n", __func__);
-#endif
-
-#ifdef GGML_USE_VULKAN
-   new_clip->backend = ggml_backend_vk_init(0);
-   LOG_INF("%s: CLIP using Vulkan backend\n", __func__);
-#endif
-
-#ifdef GGML_USE_SYCL
-   new_clip->backend = ggml_backend_sycl_init(0);
-   LOG_INF("%s: CLIP using SYCL backend\n", __func__);
-#endif
-
-    if (!new_clip->backend) {
-        new_clip->backend = ggml_backend_cpu_init();
-        LOG_INF("%s: CLIP using CPU backend\n", __func__);
+    ggml_backend_t backend = ggml_backend_init_best();
+    if (backend == nullptr) {
+        LOG_ERR("%s: failed to initialize backend\n", __func__);
+        clip_free(new_clip);
+        gguf_free(ctx);
+        return nullptr;
     }
+    LOG_INF("%s: using %s backend\n", __func__, ggml_backend_name(backend));
+    new_clip->backend = backend;
 
     // model size and capabilities
     {

llama/llama.go

@@ -199,21 +199,25 @@ func (c *Context) KvCacheDefrag() {
 
 // Get the embeddings for a sequence id
 func (c *Context) GetEmbeddingsSeq(seqId int) []float32 {
-	embeddings := unsafe.Pointer(C.llama_get_embeddings_seq(c.c, C.int(seqId)))
-	if embeddings == nil {
+	e := unsafe.Pointer(C.llama_get_embeddings_seq(c.c, C.int(seqId)))
+	if e == nil {
 		return nil
 	}
 
-	return unsafe.Slice((*float32)(embeddings), c.Model().NEmbd())
+	embeddings := make([]float32, c.Model().NEmbd())
+	_ = copy(embeddings, unsafe.Slice((*float32)(e), c.Model().NEmbd()))
+	return embeddings
 }
 
 func (c *Context) GetEmbeddingsIth(i int) []float32 {
-	embeddings := unsafe.Pointer(C.llama_get_embeddings_ith(c.c, C.int32_t(i)))
-	if embeddings == nil {
+	e := unsafe.Pointer(C.llama_get_embeddings_ith(c.c, C.int32_t(i)))
+	if e == nil {
 		return nil
 	}
 
-	return unsafe.Slice((*float32)(embeddings), c.Model().NEmbd())
+	embeddings := make([]float32, c.Model().NEmbd())
+	_ = copy(embeddings, unsafe.Slice((*float32)(e), c.Model().NEmbd()))
+	return embeddings
 }
 
 type ModelParams struct {

llama/mllama.cpp

@@ -558,30 +558,15 @@ struct mllama_ctx *mllama_model_load(const char *fname, const int verbosity = 1)
 
     mllama_ctx *new_mllama = new mllama_ctx{};
 
-#ifdef GGML_USE_CUDA
-    new_mllama->backend = ggml_backend_cuda_init(0);
-    LOG("vision using CUDA backend");
-#endif
-
-#ifdef GGML_USE_METAL
-    new_mllama->backend = ggml_backend_metal_init();
-    LOG("vision using Metal backend");
-#endif
-
-#ifdef GGML_USE_CANN
-    new_mllama->backend = ggml_backend_cann_init(0);
-    LOG("vision using CANN backend");
-#endif
-
-#ifdef GGML_USE_VULKAN
-    new_mllama->backend = ggml_backend_vk_init(0);
-    LOG("vision using Vulkan backend");
-#endif
-
-    if (!new_mllama->backend) {
-        new_mllama->backend = ggml_backend_cpu_init();
-        LOG("vision using CPU backend");
+    ggml_backend_t backend = ggml_backend_init_best();
+    if (backend == nullptr) {
+        LOG("%s: failed to initialize backend\n", __func__);
+        mllama_free(new_mllama);
+        gguf_free(ctx);
+        return nullptr;
     }
+    LOG("%s: using %s backend\n", __func__, ggml_backend_name(backend));
+    new_mllama->backend = backend;
 
     // load tensors
     {

llama/patches/0013-use-dynamic-backend-loading-for-clip.patch

@@ -1,14 +1,14 @@
 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
 From: jmorganca <jmorganca@gmail.com>
 Date: Sat, 4 Jan 2025 22:52:48 -0800
-Subject: [PATCH] re-enable gpu for clip
+Subject: [PATCH] use dynamic backend loading for clip
 
 ---
- examples/llava/clip.cpp | 86 ++++++++++++++++++++---------------------
- 1 file changed, 43 insertions(+), 43 deletions(-)
+ examples/llava/clip.cpp | 74 +++++++++++++++--------------------------
+ 1 file changed, 27 insertions(+), 47 deletions(-)
 
 diff --git a/examples/llava/clip.cpp b/examples/llava/clip.cpp
-index b3c1829f..718052e1 100644
+index b3c1829f..86b91d5c 100644
 --- a/examples/llava/clip.cpp
 +++ b/examples/llava/clip.cpp
 @@ -8,25 +8,25 @@
@@ -56,7 +56,7 @@ index b3c1829f..718052e1 100644
  
  #define STB_IMAGE_IMPLEMENTATION
  #include "stb_image.h"
-@@ -1235,30 +1235,30 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
+@@ -1235,35 +1235,15 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
          }
      }
  
@@ -84,30 +84,19 @@ index b3c1829f..718052e1 100644
 -//    new_clip->backend = ggml_backend_sycl_init(0);
 -//    LOG_INF("%s: CLIP using SYCL backend\n", __func__);
 -//#endif
-+#ifdef GGML_USE_CUDA
-+   new_clip->backend = ggml_backend_cuda_init(0);
-+   LOG_INF("%s: CLIP using CUDA backend\n", __func__);
-+#endif
-+
-+#ifdef GGML_USE_METAL
-+   new_clip->backend = ggml_backend_metal_init();
-+   LOG_INF("%s: CLIP using Metal backend\n", __func__);
-+#endif
-+
-+#ifdef GGML_USE_CANN
-+   new_clip->backend = ggml_backend_cann_init(0);
-+   LOG_INF("%s: CLIP using CANN backend\n", __func__);
-+#endif
-+
-+#ifdef GGML_USE_VULKAN
-+   new_clip->backend = ggml_backend_vk_init(0);
-+   LOG_INF("%s: CLIP using Vulkan backend\n", __func__);
-+#endif
-+
-+#ifdef GGML_USE_SYCL
-+   new_clip->backend = ggml_backend_sycl_init(0);
-+   LOG_INF("%s: CLIP using SYCL backend\n", __func__);
-+#endif
+-
+-    if (!new_clip->backend) {
+-        new_clip->backend = ggml_backend_cpu_init();
+-        LOG_INF("%s: CLIP using CPU backend\n", __func__);
++    ggml_backend_t backend = ggml_backend_init_best();
++    if (backend == nullptr) {
++        LOG_ERR("%s: failed to initialize backend\n", __func__);
++        clip_free(new_clip);
++        gguf_free(ctx);
++        return nullptr;
+     }
++    LOG_INF("%s: using %s backend\n", __func__, ggml_backend_name(backend));
++    new_clip->backend = backend;
  
-     if (!new_clip->backend) {
-         new_clip->backend = ggml_backend_cpu_init();
+     // model size and capabilities
+     {

llama/runner/README.md

@@ -4,18 +4,18 @@
 
 A minimial runner for loading a model and running inference via a http web server.
 
-```
+```shell
 ./runner -model <model binary>
 ```
 
 ### Completion
 
-```
+```shell
 curl -X POST -H "Content-Type: application/json" -d '{"prompt": "hi"}' http://localhost:8080/completion
 ```
 
 ### Embeddings
 
-```
+```shell
 curl -X POST -H "Content-Type: application/json" -d '{"prompt": "turn me into an embedding"}' http://localhost:8080/embedding
 ```

llm/server.go

@@ -281,9 +281,14 @@ func NewLlamaServer(gpus discover.GpuInfoList, model string, ggml *GGML, adapter
 		finalParams = append(finalParams, params...)
 		finalParams = append(finalParams, "--port", strconv.Itoa(port))
 
-		pathEnv := "LD_LIBRARY_PATH"
-		if runtime.GOOS == "windows" {
+		var pathEnv string
+		switch runtime.GOOS {
+		case "windows":
 			pathEnv = "PATH"
+		case "darwin":
+			pathEnv = "DYLD_LIBRARY_PATH"
+		default:
+			pathEnv = "LD_LIBRARY_PATH"
 		}
 
 		var libraryPaths []string
@@ -385,7 +390,8 @@ func NewLlamaServer(gpus discover.GpuInfoList, model string, ggml *GGML, adapter
 					strings.HasPrefix(ev, "HSA_") ||
 					strings.HasPrefix(ev, "GGML_") ||
 					strings.HasPrefix(ev, "PATH=") ||
-					strings.HasPrefix(ev, "LD_LIBRARY_PATH=") {
+					strings.HasPrefix(ev, "LD_LIBRARY_PATH=") ||
+					strings.HasPrefix(ev, "DYLD_LIBRARY_PATH=") {
 					filteredEnv = append(filteredEnv, ev)
 				}
 			}

macapp/README.md

@@ -6,14 +6,14 @@ This app builds upon Ollama to provide a desktop experience for running models.
 
 First, build the `ollama` binary:
 
-```
+```shell
 cd ..
 go build .
 ```
 
 Then run the desktop app with `npm start`:
 
-```
+```shell
 cd macapp
 npm install
 npm start

ml/backend/ggml/ggml/src/ggml-cuda/template-instances/generate_cu_files.py

@@ -1,77 +0,0 @@
-#!/usr/bin/env python3
-
-from glob import glob
-import os
-
-TYPES_KV = ["GGML_TYPE_Q4_0", "GGML_TYPE_Q4_1", "GGML_TYPE_Q5_0", "GGML_TYPE_Q5_1", "GGML_TYPE_Q8_0", "GGML_TYPE_F16"]
-
-SOURCE_FATTN_VEC = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.
-
-#include "../fattn-vec-f{vkq_size}.cuh"
-
-DECL_FATTN_VEC_F{vkq_size}_CASE({head_size}, {type_k}, {type_v});
-"""
-
-SOURCE_FATTN_WMMA_START = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.
-
-#include "../fattn-wmma-f16.cuh"
-
-"""
-
-SOURCE_FATTN_WMMA_CASE = "DECL_FATTN_WMMA_F16_CASE({head_size}, {cols_per_block}, {kq_acc_t});\n"
-
-TYPES_MMQ = [
-    "GGML_TYPE_Q4_0", "GGML_TYPE_Q4_1", "GGML_TYPE_Q5_0", "GGML_TYPE_Q5_1", "GGML_TYPE_Q8_0",
-    "GGML_TYPE_Q2_K", "GGML_TYPE_Q3_K", "GGML_TYPE_Q4_K", "GGML_TYPE_Q5_K", "GGML_TYPE_Q6_K",
-    "GGML_TYPE_IQ2_XXS", "GGML_TYPE_IQ2_XS", "GGML_TYPE_IQ2_S", "GGML_TYPE_IQ3_XXS", "GGML_TYPE_IQ3_S",
-    "GGML_TYPE_IQ1_S", "GGML_TYPE_IQ4_NL", "GGML_TYPE_IQ4_XS"
-]
-
-SOURCE_MMQ = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.
-
-#include "../mmq.cuh"
-
-DECL_MMQ_CASE({type});
-"""
-
-
-def get_short_name(long_quant_name):
-    return long_quant_name.replace("GGML_TYPE_", "").lower()
-
-
-def get_head_sizes(type_k, type_v):
-    if type_k == "GGML_TYPE_F16" and type_v == "GGML_TYPE_F16":
-        return [64, 128, 256]
-    if type_k == "GGML_TYPE_F16":
-        return [64, 128]
-    return [128]
-
-
-for filename in glob("*.cu"):
-    os.remove(filename)
-
-for vkq_size in [16, 32]:
-    for type_k in TYPES_KV:
-        for type_v in TYPES_KV:
-            for head_size in get_head_sizes(type_k, type_v):
-                with open(f"fattn-vec-f{vkq_size}-instance-hs{head_size}-{get_short_name(type_k)}-{get_short_name(type_v)}.cu", "w") as f:
-                    f.write(SOURCE_FATTN_VEC.format(vkq_size=vkq_size, head_size=head_size, type_k=type_k, type_v=type_v))
-
-for kq_acc_t in ["half", "float"]:
-    for cols_per_block in [8, 16, 32]:
-        if kq_acc_t == "float" and cols_per_block == 8:
-            continue
-
-        with open(f"fattn-wmma-f16-instance-kq{kq_acc_t}-cpb{cols_per_block}.cu", "w") as f:
-            f.write(SOURCE_FATTN_WMMA_START)
-
-            for head_size in [64, 80, 96, 112, 128, 256]:
-                if cols_per_block == 8 and head_size % 32 != 0: # wmma fragment is 8x32
-                    continue
-                if kq_acc_t == "float" and cols_per_block == 32 and head_size == 256: # register spilling, bad performance
-                    continue
-                f.write(SOURCE_FATTN_WMMA_CASE.format(kq_acc_t=kq_acc_t, cols_per_block=cols_per_block, head_size=head_size))
-
-for type in TYPES_MMQ:
-    with open(f"mmq-instance-{get_short_name(type)}.cu", "w") as f:
-        f.write(SOURCE_MMQ.format(type=type))

ml/backend/ggml/ggml/src/ggml.go

@@ -41,36 +41,48 @@ func sink(level C.int, text *C.char, _ unsafe.Pointer) {
 }
 
 var OnceLoad = sync.OnceFunc(func() {
-	var lib struct{ name, defaultValue string }
+	exe, err := os.Executable()
+	if err != nil {
+		slog.Warn("failed to get executable path", "error", err)
+		exe = "."
+	}
+
+	// PATH, LD_LIBRARY_PATH, and DYLD_LIBRARY_PATH are often
+	// set by the parent process, however, use a default value
+	// if the environment variable is not set.
+	var name, value string
 	switch runtime.GOOS {
-	case "darwin", "linux":
-		lib.name = "LD_LIBRARY_PATH"
-		lib.defaultValue = "/usr/local/lib:/usr/lib"
+	case "darwin":
+		// On macOS, DYLD_LIBRARY_PATH is often not set, so
+		// we use the directory of the executable as the default.
+		name = "DYLD_LIBRARY_PATH"
+		value = filepath.Dir(exe)
 	case "windows":
-		lib.name = "PATH"
-		lib.defaultValue = "."
+		name = "PATH"
+		value = filepath.Join(filepath.Dir(exe), "lib", "ollama")
 	default:
-		return
+		name = "LD_LIBRARY_PATH"
+		value = filepath.Join(filepath.Dir(exe), "..", "lib", "ollama")
 	}
 
-	paths, ok := os.LookupEnv(lib.name)
+	paths, ok := os.LookupEnv(name)
 	if !ok {
-		paths = lib.defaultValue
-	}
-
-	if runtime.GOOS == "darwin" {
-		if _, ok := os.LookupEnv("DYLD_LIBRARY_PATH"); !ok {
-			os.Setenv("DYLD_LIBRARY_PATH", paths)
-		}
+		paths = value
 	}
 
 	split := filepath.SplitList(paths)
 	visited := make(map[string]struct{}, len(split))
 	for _, path := range split {
-		abspath, _ := filepath.Abs(path)
+		abspath, err := filepath.Abs(path)
+		if err != nil {
+			slog.Error("failed to get absolute path", "error", err)
+			continue
+		}
+
 		if _, ok := visited[abspath]; !ok {
 			func() {
-				cpath := C.CString(path)
+				slog.Debug("ggml backend load all from path", "path", abspath)
+				cpath := C.CString(abspath)
 				defer C.free(unsafe.Pointer(cpath))
 				C.ggml_backend_load_all_from_path(cpath)
 			}()

scripts/build_darwin.sh

@@ -32,9 +32,10 @@ _build_darwin() {
             status "Building darwin $ARCH dynamic backends"
             cmake -B build/darwin-$ARCH \
                 -DCMAKE_OSX_ARCHITECTURES=x86_64 \
-                -DCMAKE_OSX_DEPLOYMENT_TARGET=11.3
+                -DCMAKE_OSX_DEPLOYMENT_TARGET=11.3 \
+                -DCMAKE_INSTALL_PREFIX=$INSTALL_PREFIX
             cmake --build build/darwin-$ARCH --target ggml-cpu -j
-            install build/darwin-$ARCH/lib/ollama/*.{dylib,so} $INSTALL_PREFIX
+            cmake --install build/darwin-$ARCH --component CPU
         fi
     done
 }
@@ -43,6 +44,7 @@ _sign_darwin() {
     status "Creating universal binary..."
     mkdir -p dist/darwin
     lipo -create -output dist/darwin/ollama dist/darwin-*/ollama
+    chmod +x dist/darwin/ollama
 
     if [ -n "$APPLE_IDENTITY" ]; then
         for F in dist/darwin/ollama dist/darwin-amd64/lib/ollama/*; do

scripts/build_windows.ps1

@@ -162,8 +162,11 @@ function gatherDependencies() {
         $depArch=$script:TARGET_ARCH
     }
     if ($depArch -eq "x64") {
+        write-host "cp ${env:VCToolsRedistDir}\${depArch}\Microsoft.VC*.CRT\msvcp140*.dll ${script:DIST_DIR}\lib\ollama\"
         cp "${env:VCToolsRedistDir}\${depArch}\Microsoft.VC*.CRT\msvcp140*.dll" "${script:DIST_DIR}\lib\ollama\"
+        write-host "cp ${env:VCToolsRedistDir}\${depArch}\Microsoft.VC*.CRT\vcruntime140.dll ${script:DIST_DIR}\lib\ollama\"
         cp "${env:VCToolsRedistDir}\${depArch}\Microsoft.VC*.CRT\vcruntime140.dll" "${script:DIST_DIR}\lib\ollama\"
+        write-host "cp ${env:VCToolsRedistDir}\${depArch}\Microsoft.VC*.CRT\vcruntime140_1.dll ${script:DIST_DIR}\lib\ollama\"
         cp "${env:VCToolsRedistDir}\${depArch}\Microsoft.VC*.CRT\vcruntime140_1.dll" "${script:DIST_DIR}\lib\ollama\"
         $llvmCrtDir="$env:VCToolsRedistDir\..\..\..\Tools\Llvm\${depArch}\bin"
         foreach ($part in $("runtime", "stdio", "filesystem", "math", "convert", "heap", "string", "time", "locale", "environment")) {

server/download.go

@@ -172,7 +172,10 @@ func (b *blobDownload) Prepare(ctx context.Context, requestURL *url.URL, opts *r
 		}
 	}
 
-	slog.Info(fmt.Sprintf("downloading %s in %d %s part(s)", b.Digest[7:19], len(b.Parts), format.HumanBytes(b.Parts[0].Size)))
+	if len(b.Parts) > 0 {
+		slog.Info(fmt.Sprintf("downloading %s in %d %s part(s)", b.Digest[7:19], len(b.Parts), format.HumanBytes(b.Parts[0].Size)))
+	}
+
 	return nil
 }
 
@@ -365,7 +368,7 @@ func (b *blobDownload) downloadChunk(ctx context.Context, requestURL *url.URL, w
 				lastUpdated := part.lastUpdated
 				part.lastUpdatedMu.Unlock()
 
-				if !lastUpdated.IsZero() && time.Since(lastUpdated) > 5*time.Second {
+				if !lastUpdated.IsZero() && time.Since(lastUpdated) > 30*time.Second {
 					const msg = "%s part %d stalled; retrying. If this persists, press ctrl-c to exit, then 'ollama pull' to find a faster connection."
 					slog.Info(fmt.Sprintf(msg, b.Digest[7:19], part.N))
 					// reset last updated

server/upload.go

@@ -108,7 +108,9 @@ func (b *blobUpload) Prepare(ctx context.Context, requestURL *url.URL, opts *reg
 		offset += size
 	}
 
-	slog.Info(fmt.Sprintf("uploading %s in %d %s part(s)", b.Digest[7:19], len(b.Parts), format.HumanBytes(b.Parts[0].Size)))
+	if len(b.Parts) > 0 {
+		slog.Info(fmt.Sprintf("uploading %s in %d %s part(s)", b.Digest[7:19], len(b.Parts), format.HumanBytes(b.Parts[0].Size)))
+	}
 
 	requestURL, err = url.Parse(location)
 	if err != nil {