Merge branch 'ollama:main' into main

Author: likelovewant

.github/workflows/release.yaml

@@ -103,11 +103,6 @@ jobs:
         arch: [amd64]
         preset: ['CPU']
         include:
-          - os: windows
-            arch: amd64
-            preset: 'CUDA 11'
-            install: https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.89_win10.exe
-            cuda-version: '11.3'
           - os: windows
             arch: amd64
             preset: 'CUDA 12'
@@ -324,7 +319,6 @@ jobs:
             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 ;;

.github/workflows/test.yaml

@@ -46,7 +46,7 @@ jobs:
         include:
           - preset: CPU
           - preset: CUDA
-            container: nvidia/cuda:11.8.0-devel-ubuntu22.04
+            container: nvidia/cuda:12.8.1-devel-ubuntu22.04
             flags: '-DCMAKE_CUDA_ARCHITECTURES=87'
           - preset: ROCm
             container: rocm/dev-ubuntu-22.04:6.1.2
@@ -78,7 +78,7 @@ jobs:
         include:
           - preset: CPU
           - preset: CUDA
-            install: https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.89_win10.exe
+            install: https://developer.download.nvidia.com/compute/cuda/12.8.0/local_installers/cuda_12.8.0_571.96_windows.exe
             flags: '-DCMAKE_CUDA_ARCHITECTURES=80'
           - preset: ROCm
             install: https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-24.Q4-WinSvr2022-For-HIP.exe
@@ -102,7 +102,7 @@ jobs:
           $ErrorActionPreference = "Stop"
           if ("${{ steps.cache-install.outputs.cache-hit }}" -ne 'true') {
             Invoke-WebRequest -Uri "${{ matrix.install }}" -OutFile "install.exe"
-            Start-Process -FilePath .\install.exe -ArgumentList (@("-s", "cudart_11.3", "nvcc_11.3", "cublas_11.3", "cublas_dev_11.3")) -NoNewWindow -Wait
+            Start-Process -FilePath .\install.exe -ArgumentList (@("-s", "cudart_12.8", "nvcc_12.8", "cublas_12.8", "cublas_dev_12.8")) -NoNewWindow -Wait
           }
 
           $cudaPath = (Resolve-Path "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\*").path

CMakePresets.json

@@ -17,14 +17,6 @@
       "name": "CUDA",
       "inherits": [ "Default" ]
     },
-    {
-      "name": "CUDA 11",
-      "inherits": [ "CUDA" ],
-      "cacheVariables": {
-        "CMAKE_CUDA_ARCHITECTURES": "50;52;53;60;61;70;75;80;86",
-        "CMAKE_CUDA_FLAGS": "-Wno-deprecated-gpu-targets"
-      }
-    },
     {
       "name": "CUDA 12",
       "inherits": [ "CUDA" ],
@@ -78,11 +70,6 @@
       "configurePreset": "CUDA",
       "targets": [ "ggml-cuda" ]
     },
-    {
-      "name": "CUDA 11",
-      "inherits": [ "CUDA" ],
-      "configurePreset": "CUDA 11"
-    },
     {
       "name": "CUDA 12",
       "inherits": [ "CUDA" ],

Dockerfile

@@ -7,14 +7,10 @@ ARG JETPACK5VERSION=r35.4.1
 ARG JETPACK6VERSION=r36.4.0
 ARG CMAKEVERSION=3.31.2
 
-# CUDA v11 requires gcc v10.  v10.3 has regressions, so the rockylinux 8.5 AppStream has the latest compatible version
 FROM --platform=linux/amd64 rocm/dev-almalinux-8:${ROCMVERSION}-complete AS base-amd64
 RUN yum install -y yum-utils \
-    && yum-config-manager --add-repo https://dl.rockylinux.org/vault/rocky/8.5/AppStream/\$basearch/os/ \
-    && rpm --import https://dl.rockylinux.org/pub/rocky/RPM-GPG-KEY-Rocky-8 \
-    && dnf install -y yum-utils ccache gcc-toolset-10-gcc-10.2.1-8.2.el8 gcc-toolset-10-gcc-c++-10.2.1-8.2.el8 gcc-toolset-10-binutils-2.35-11.el8 \
+    && dnf install -y ccache \
     && yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-rhel8.repo
-ENV PATH=/opt/rh/gcc-toolset-10/root/usr/bin:$PATH
 
 FROM --platform=linux/arm64 almalinux:8 AS base-arm64
 # install epel-release for ccache
@@ -38,15 +34,6 @@ RUN --mount=type=cache,target=/root/.ccache \
         && cmake --build --parallel --preset 'CPU' \
         && cmake --install build --component CPU --strip --parallel 8
 
-FROM base AS cuda-11
-ARG CUDA11VERSION=11.3
-RUN dnf install -y cuda-toolkit-${CUDA11VERSION//./-}
-ENV PATH=/usr/local/cuda-11/bin:$PATH
-RUN --mount=type=cache,target=/root/.ccache \
-    cmake --preset 'CUDA 11' \
-        && cmake --build --parallel --preset 'CUDA 11' \
-        && cmake --install build --component CUDA --strip --parallel 8
-
 FROM base AS cuda-12
 ARG CUDA12VERSION=12.8
 RUN dnf install -y cuda-toolkit-${CUDA12VERSION//./-}
@@ -98,11 +85,9 @@ RUN --mount=type=cache,target=/root/.cache/go-build \
     go build -trimpath -buildmode=pie -o /bin/ollama .
 
 FROM --platform=linux/amd64 scratch AS amd64
-COPY --from=cuda-11 dist/lib/ollama/cuda_v11 /lib/ollama/cuda_v11
 COPY --from=cuda-12 dist/lib/ollama/cuda_v12 /lib/ollama/cuda_v12
 
 FROM --platform=linux/arm64 scratch AS arm64
-COPY --from=cuda-11 dist/lib/ollama/cuda_v11 /lib/ollama/cuda_v11
 COPY --from=cuda-12 dist/lib/ollama/cuda_v12 /lib/ollama/cuda_v12
 COPY --from=jetpack-5 dist/lib/ollama/cuda_v11 /lib/ollama/cuda_jetpack5
 COPY --from=jetpack-6 dist/lib/ollama/cuda_v12 /lib/ollama/cuda_jetpack6

Makefile.sync

@@ -1,6 +1,6 @@
 UPSTREAM=https://github.com/ggerganov/llama.cpp.git
 WORKDIR=llama/vendor
-FETCH_HEAD=2016f07bd106c73699ecbaace80f55db5ed95dac
+FETCH_HEAD=e1e8e0991ffd9e99a445c6812bb519d5bac9f4b5
 
 .PHONY: help
 help:

README.md

@@ -83,6 +83,8 @@ Here are some example models that can be downloaded:
 | QwQ                | 32B        | 20GB  | `ollama run qwq`                 |
 | DeepSeek-R1        | 7B         | 4.7GB | `ollama run deepseek-r1`         |
 | DeepSeek-R1        | 671B       | 404GB | `ollama run deepseek-r1:671b`    |
+| Llama 4            | 109B       | 67GB  | `ollama run llama4:scout`        |
+| Llama 4            | 400B       | 245GB | `ollama run llama4:maverick`     |
 | 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`         |
@@ -99,7 +101,7 @@ Here are some example models that can be downloaded:
 | Code Llama         | 7B         | 3.8GB | `ollama run codellama`           |
 | Llama 2 Uncensored | 7B         | 3.8GB | `ollama run llama2-uncensored`   |
 | LLaVA              | 7B         | 4.5GB | `ollama run llava`               |
-| Granite-3.2         | 8B         | 4.9GB | `ollama run granite3.2`          |
+| Granite-3.3         | 8B         | 4.9GB | `ollama run granite3.3`          |
 
 > [!NOTE]
 > You should have at least 8 GB of RAM available to run the 7B models, 16 GB to run the 13B models, and 32 GB to run the 33B models.
@@ -334,6 +336,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Ollama Basic Chat: Uses HyperDiv Reactive UI](https://github.com/rapidarchitect/ollama_basic_chat)
 - [Ollama-chats RPG](https://github.com/drazdra/ollama-chats)
 - [IntelliBar](https://intellibar.app/) (AI-powered assistant for macOS)
+- [Jirapt](https://github.com/AliAhmedNada/jirapt) (Jira Integration to generate issues, tasks, epics)
 - [QA-Pilot](https://github.com/reid41/QA-Pilot) (Interactive chat tool that can leverage Ollama models for rapid understanding and navigation of GitHub code repositories)
 - [ChatOllama](https://github.com/sugarforever/chat-ollama) (Open Source Chatbot based on Ollama with Knowledge Bases)
 - [CRAG Ollama Chat](https://github.com/Nagi-ovo/CRAG-Ollama-Chat) (Simple Web Search with Corrective RAG)
@@ -416,6 +419,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [1Panel](https://github.com/1Panel-dev/1Panel/) (Web-based Linux Server Management Tool)
 - [AstrBot](https://github.com/Soulter/AstrBot/) (User-friendly LLM-based multi-platform chatbot with a WebUI, supporting RAG, LLM agents, and plugins integration)
 - [Reins](https://github.com/ibrahimcetin/reins) (Easily tweak parameters, customize system prompts per chat, and enhance your AI experiments with reasoning model support.)
+- [Flufy](https://github.com/Aharon-Bensadoun/Flufy) (A beautiful chat interface for interacting with Ollama's API. Built with React, TypeScript, and Material-UI.)
 - [Ellama](https://github.com/zeozeozeo/ellama) (Friendly native app to chat with an Ollama instance)
 - [screenpipe](https://github.com/mediar-ai/screenpipe) Build agents powered by your screen history
 - [Ollamb](https://github.com/hengkysteen/ollamb) (Simple yet rich in features, cross-platform built with Flutter and designed for Ollama. Try the [web demo](https://hengkysteen.github.io/demo/ollamb/).)
@@ -491,7 +495,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 
 ### Libraries
 
-- [LangChain](https://python.langchain.com/docs/integrations/llms/ollama) and [LangChain.js](https://js.langchain.com/docs/integrations/chat/ollama/) with [example](https://js.langchain.com/docs/tutorials/local_rag/)
+- [LangChain](https://python.langchain.com/docs/integrations/chat/ollama/) and [LangChain.js](https://js.langchain.com/docs/integrations/chat/ollama/) with [example](https://js.langchain.com/docs/tutorials/local_rag/)
 - [Firebase Genkit](https://firebase.google.com/docs/genkit/plugins/ollama)
 - [crewAI](https://github.com/crewAIInc/crewAI)
 - [Yacana](https://remembersoftwares.github.io/yacana/) (User-friendly multi-agent framework for brainstorming and executing predetermined flows with built-in tool integration)

api/types.go

@@ -283,12 +283,7 @@ type Runner struct {
 	NumBatch  int   `json:"num_batch,omitempty"`
 	NumGPU    int   `json:"num_gpu,omitempty"`
 	MainGPU   int   `json:"main_gpu,omitempty"`
-	LowVRAM   bool  `json:"low_vram,omitempty"`
-	F16KV     bool  `json:"f16_kv,omitempty"` // Deprecated: This option is ignored
-	LogitsAll bool  `json:"logits_all,omitempty"`
-	VocabOnly bool  `json:"vocab_only,omitempty"`
 	UseMMap   *bool `json:"use_mmap,omitempty"`
-	UseMLock  bool  `json:"use_mlock,omitempty"`
 	NumThread int   `json:"num_thread,omitempty"`
 }
 
@@ -671,8 +666,6 @@ func DefaultOptions() Options {
 			NumBatch:  512,
 			NumGPU:    -1, // -1 here indicates that NumGPU should be set dynamically
 			NumThread: 0,  // let the runtime decide
-			LowVRAM:   false,
-			UseMLock:  false,
 			UseMMap:   nil,
 		},
 	}

cmd/cmd.go

@@ -31,6 +31,7 @@ import (
 	"github.com/olekukonko/tablewriter"
 	"github.com/spf13/cobra"
 	"golang.org/x/crypto/ssh"
+	"golang.org/x/sync/errgroup"
 	"golang.org/x/term"
 
 	"github.com/ollama/ollama/api"
@@ -41,6 +42,7 @@ import (
 	"github.com/ollama/ollama/runner"
 	"github.com/ollama/ollama/server"
 	"github.com/ollama/ollama/types/model"
+	"github.com/ollama/ollama/types/syncmap"
 	"github.com/ollama/ollama/version"
 )
 
@@ -106,7 +108,7 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 	}
 	spinner.Stop()
 
-	req.Name = args[0]
+	req.Model = args[0]
 	quantize, _ := cmd.Flags().GetString("quantize")
 	if quantize != "" {
 		req.Quantize = quantize
@@ -117,34 +119,54 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 		return err
 	}
 
-	if len(req.Files) > 0 {
-		fileMap := map[string]string{}
-		for f, digest := range req.Files {
+	var g errgroup.Group
+	g.SetLimit(max(runtime.GOMAXPROCS(0)-1, 1))
+
+	files := syncmap.NewSyncMap[string, string]()
+	for f, digest := range req.Files {
+		g.Go(func() error {
 			if _, err := createBlob(cmd, client, f, digest, p); err != nil {
 				return err
 			}
-			fileMap[filepath.Base(f)] = digest
-		}
-		req.Files = fileMap
+
+			// TODO: this is incorrect since the file might be in a subdirectory
+			//       instead this should take the path relative to the model directory
+			//       but the current implementation does not allow this
+			files.Store(filepath.Base(f), digest)
+			return nil
+		})
 	}
 
-	if len(req.Adapters) > 0 {
-		fileMap := map[string]string{}
-		for f, digest := range req.Adapters {
+	adapters := syncmap.NewSyncMap[string, string]()
+	for f, digest := range req.Adapters {
+		g.Go(func() error {
 			if _, err := createBlob(cmd, client, f, digest, p); err != nil {
 				return err
 			}
-			fileMap[filepath.Base(f)] = digest
-		}
-		req.Adapters = fileMap
+
+			// TODO: same here
+			adapters.Store(filepath.Base(f), digest)
+			return nil
+		})
 	}
 
+	if err := g.Wait(); err != nil {
+		return err
+	}
+
+	req.Files = files.Items()
+	req.Adapters = adapters.Items()
+
 	bars := make(map[string]*progress.Bar)
 	fn := func(resp api.ProgressResponse) error {
 		if resp.Digest != "" {
 			bar, ok := bars[resp.Digest]
 			if !ok {
-				bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
+				msg := resp.Status
+				if msg == "" {
+					msg = fmt.Sprintf("pulling %s...", resp.Digest[7:19])
+				}
+				bar = progress.NewBar(msg, resp.Total, resp.Completed)
 				bars[resp.Digest] = bar
 				p.Add(resp.Digest, bar)
 			}
@@ -213,7 +235,7 @@ func createBlob(cmd *cobra.Command, client *api.Client, path string, digest stri
 		}
 	}()
 
-	if err = client.CreateBlob(cmd.Context(), digest, io.TeeReader(bin, &pw)); err != nil {
+	if err := client.CreateBlob(cmd.Context(), digest, io.TeeReader(bin, &pw)); err != nil {
 		return "", err
 	}
 	return digest, nil

cmd/cmd_test.go

@@ -690,7 +690,7 @@ func TestCreateHandler(t *testing.T) {
 						return
 					}
 
-					if req.Name != "test-model" {
+					if req.Model != "test-model" {
 						t.Errorf("expected model name 'test-model', got %s", req.Name)
 					}
 

convert/convert.go

@@ -4,9 +4,9 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
-	"io"
 	"io/fs"
 	"log/slog"
+	"os"
 	"slices"
 	"strings"
 
@@ -89,7 +89,7 @@ type ModelConverter interface {
 	// KV maps parameters to LLM key-values
 	KV(*Tokenizer) ggml.KV
 	// Tensors maps input tensors to LLM tensors. Model specific modifications can be done here.
-	Tensors([]Tensor) []ggml.Tensor
+	Tensors([]Tensor) []*ggml.Tensor
 	// Replacements returns a list of string pairs to replace in tensor names.
 	// See [strings.Replacer](https://pkg.go.dev/strings#Replacer) for details
 	Replacements() []string
@@ -106,13 +106,13 @@ type AdapterConverter interface {
 	// KV maps parameters to LLM key-values
 	KV(ggml.KV) ggml.KV
 	// Tensors maps input tensors to LLM tensors. Adapter specific modifications can be done here.
-	Tensors([]Tensor) []ggml.Tensor
+	Tensors([]Tensor) []*ggml.Tensor
 	// Replacements returns a list of string pairs to replace in tensor names.
 	// See [strings.Replacer](https://pkg.go.dev/strings#Replacer) for details
 	Replacements() []string
 }
 
-func ConvertAdapter(fsys fs.FS, ws io.WriteSeeker, baseKV ggml.KV) error {
+func ConvertAdapter(fsys fs.FS, f *os.File, baseKV ggml.KV) error {
 	bts, err := fs.ReadFile(fsys, "adapter_config.json")
 	if err != nil {
 		return err
@@ -147,14 +147,14 @@ func ConvertAdapter(fsys fs.FS, ws io.WriteSeeker, baseKV ggml.KV) error {
 		return err
 	}
 
-	return writeFile(ws, conv.KV(baseKV), conv.Tensors(ts))
+	return writeFile(f, conv.KV(baseKV), conv.Tensors(ts))
 }
 
 // Convert writes an Ollama compatible model to the provided io.WriteSeeker based on configurations
 // and files it finds in the input path.
 // Supported input model formats include safetensors.
 // Supported input tokenizers files include tokenizer.json (preferred) and tokenizer.model.
-func ConvertModel(fsys fs.FS, ws io.WriteSeeker) error {
+func ConvertModel(fsys fs.FS, f *os.File) error {
 	bts, err := fs.ReadFile(fsys, "config.json")
 	if err != nil {
 		return err
@@ -239,13 +239,13 @@ func ConvertModel(fsys fs.FS, ws io.WriteSeeker) error {
 		return err
 	}
 
-	return writeFile(ws, conv.KV(t), conv.Tensors(ts))
+	return writeFile(f, conv.KV(t), conv.Tensors(ts))
 }
 
-func writeFile(ws io.WriteSeeker, kv ggml.KV, ts []ggml.Tensor) error {
+func writeFile(f *os.File, kv ggml.KV, ts []*ggml.Tensor) error {
 	for i := range ts {
 		ts[i].Shape = slices.Clone(ts[i].Shape)
 		slices.Reverse(ts[i].Shape)
 	}
-	return ggml.WriteGGUF(ws, kv, ts)
+	return ggml.WriteGGUF(f, kv, ts)
 }

convert/convert_bert.go

@@ -132,8 +132,8 @@ func (p *bertModel) KV(t *Tokenizer) ggml.KV {
 	return kv
 }
 
-func (p *bertModel) Tensors(ts []Tensor) []ggml.Tensor {
-	var out []ggml.Tensor
+func (p *bertModel) Tensors(ts []Tensor) []*ggml.Tensor {
+	var out []*ggml.Tensor
 	for _, t := range ts {
 		if slices.Contains([]string{
 			"embeddings.position_ids",
@@ -143,7 +143,7 @@ func (p *bertModel) Tensors(ts []Tensor) []ggml.Tensor {
 			continue
 		}
 
-		out = append(out, ggml.Tensor{
+		out = append(out, &ggml.Tensor{
 			Name:     t.Name(),
 			Kind:     t.Kind(),
 			Shape:    t.Shape(),