Merge branch 'main' into install-instructions-archlinux

.dockerignore

@@ -6,3 +6,4 @@ scripts
 llm/llama.cpp/ggml
 llm/llama.cpp/gguf
 .env
+.cache

.gitignore

@@ -6,3 +6,4 @@
 dist
 ollama
 ggml-metal.metal
+.cache

README.md

@@ -216,6 +216,10 @@ See the [API documentation](./docs/api.md) for all endpoints.
 
 ## Community Integrations
 
+### Mobile
+
+- [Mobile Artificial Intelligence Distribution](https://github.com/MaidFoundation/Maid) (Maid)
+
 ### Web & Desktop
 
 - [HTML UI](https://github.com/rtcfirefly/ollama-ui)
@@ -234,12 +238,15 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Emacs client](https://github.com/zweifisch/ollama)
 - [gen.nvim](https://github.com/David-Kunz/gen.nvim)
 - [ollama.nvim](https://github.com/nomnivore/ollama.nvim)
+- [ogpt.nvim](https://github.com/huynle/ogpt.nvim)
 - [gptel Emacs client](https://github.com/karthink/gptel)
 - [ollama package for archlinux](https://archlinux.org/packages/extra/x86_64/ollama/)
+- [Oatmeal](https://github.com/dustinblackman/oatmeal)
 
 ### Libraries
 
 - [LangChain](https://python.langchain.com/docs/integrations/llms/ollama) and [LangChain.js](https://js.langchain.com/docs/modules/model_io/models/llms/integrations/ollama) with [example](https://js.langchain.com/docs/use_cases/question_answering/local_retrieval_qa)
+- [LangChainGo](https://github.com/tmc/langchaingo/) with [example](https://github.com/tmc/langchaingo/tree/main/examples/ollama-completion-example)
 - [LlamaIndex](https://gpt-index.readthedocs.io/en/stable/examples/llm/ollama.html)
 - [LiteLLM](https://github.com/BerriAI/litellm)
 - [OllamaSharp for .NET](https://github.com/awaescher/OllamaSharp)
@@ -248,6 +255,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [ModelFusion Typescript Library](https://modelfusion.dev/integration/model-provider/ollama)
 - [OllamaKit for Swift](https://github.com/kevinhermawan/OllamaKit)
 - [Ollama for Dart](https://github.com/breitburg/dart-ollama)
+- [Ollama for Laravel](https://github.com/cloudstudio/ollama-laravel)
 
 ### Mobile
 
@@ -263,3 +271,4 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Dagger Chatbot](https://github.com/samalba/dagger-chatbot)
 - [Discord AI Bot](https://github.com/mekb-turtle/discord-ai-bot)
 - [Hass Ollama Conversation](https://github.com/ej52/hass-ollama-conversation)
+- [Rivet plugin](https://github.com/abrenneke/rivet-plugin-ollama)

cmd/cmd.go

@@ -30,7 +30,7 @@ import (
 	"github.com/jmorganca/ollama/api"
 	"github.com/jmorganca/ollama/format"
 	"github.com/jmorganca/ollama/parser"
-	"github.com/jmorganca/ollama/progressbar"
+	"github.com/jmorganca/ollama/progress"
 	"github.com/jmorganca/ollama/readline"
 	"github.com/jmorganca/ollama/server"
 	"github.com/jmorganca/ollama/version"
@@ -48,14 +48,16 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 		return err
 	}
 
+	p := progress.NewProgress(os.Stderr)
+	defer p.Stop()
+
+	bars := make(map[string]*progress.Bar)
+
 	modelfile, err := os.ReadFile(filename)
 	if err != nil {
 		return err
 	}
 
-	spinner := NewSpinner("transferring context")
-	go spinner.Spin(100 * time.Millisecond)
-
 	commands, err := parser.Parse(bytes.NewReader(modelfile))
 	if err != nil {
 		return err
@@ -66,6 +68,10 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 		return err
 	}
 
+	status := "transferring model data"
+	spinner := progress.NewSpinner(status)
+	p.Add(status, spinner)
+
 	for _, c := range commands {
 		switch c.Name {
 		case "model", "adapter":
@@ -76,6 +82,10 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 				path = filepath.Join(home, path[2:])
 			}
 
+			if !filepath.IsAbs(path) {
+				path = filepath.Join(filepath.Dir(filename), path)
+			}
+
 			bin, err := os.Open(path)
 			if errors.Is(err, os.ErrNotExist) && c.Name == "model" {
 				continue
@@ -99,41 +109,34 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 		}
 	}
 
-	var currentDigest string
-	var bar *progressbar.ProgressBar
-
-	request := api.CreateRequest{Name: args[0], Path: filename, Modelfile: string(modelfile)}
 	fn := func(resp api.ProgressResponse) error {
-		if resp.Digest != currentDigest && resp.Digest != "" {
+		if resp.Digest != "" {
 			spinner.Stop()
-			currentDigest = resp.Digest
-			// pulling
-			bar = progressbar.DefaultBytes(
-				resp.Total,
-				resp.Status,
-			)
-			bar.Set64(resp.Completed)
-		} else if resp.Digest == currentDigest && resp.Digest != "" {
-			bar.Set64(resp.Completed)
-		} else {
-			currentDigest = ""
+
+			bar, ok := bars[resp.Digest]
+			if !ok {
+				bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
+				bars[resp.Digest] = bar
+				p.Add(resp.Digest, bar)
+			}
+
+			bar.Set(resp.Completed)
+		} else if status != resp.Status {
 			spinner.Stop()
-			spinner = NewSpinner(resp.Status)
-			go spinner.Spin(100 * time.Millisecond)
+
+			status = resp.Status
+			spinner = progress.NewSpinner(status)
+			p.Add(status, spinner)
 		}
 
 		return nil
 	}
 
+	request := api.CreateRequest{Name: args[0], Modelfile: string(modelfile)}
 	if err := client.Create(context.Background(), &request, fn); err != nil {
 		return err
 	}
 
-	spinner.Stop()
-	if spinner.description != "success" {
-		return errors.New("unexpected end to create model")
-	}
-
 	return nil
 }
 
@@ -170,36 +173,46 @@ func PushHandler(cmd *cobra.Command, args []string) error {
 		return err
 	}
 
-	var currentDigest string
-	var bar *progressbar.ProgressBar
+	p := progress.NewProgress(os.Stderr)
+	defer p.Stop()
+
+	bars := make(map[string]*progress.Bar)
+	var status string
+	var spinner *progress.Spinner
 
-	request := api.PushRequest{Name: args[0], Insecure: insecure}
 	fn := func(resp api.ProgressResponse) error {
-		if resp.Digest != currentDigest && resp.Digest != "" {
-			currentDigest = resp.Digest
-			bar = progressbar.DefaultBytes(
-				resp.Total,
-				fmt.Sprintf("pushing %s...", resp.Digest[7:19]),
-			)
-
-			bar.Set64(resp.Completed)
-		} else if resp.Digest == currentDigest && resp.Digest != "" {
-			bar.Set64(resp.Completed)
-		} else {
-			currentDigest = ""
-			fmt.Println(resp.Status)
+		if resp.Digest != "" {
+			if spinner != nil {
+				spinner.Stop()
+			}
+
+			bar, ok := bars[resp.Digest]
+			if !ok {
+				bar = progress.NewBar(fmt.Sprintf("pushing %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
+				bars[resp.Digest] = bar
+				p.Add(resp.Digest, bar)
+			}
+
+			bar.Set(resp.Completed)
+		} else if status != resp.Status {
+			if spinner != nil {
+				spinner.Stop()
+			}
+
+			status = resp.Status
+			spinner = progress.NewSpinner(status)
+			p.Add(status, spinner)
 		}
+
 		return nil
 	}
 
+	request := api.PushRequest{Name: args[0], Insecure: insecure}
 	if err := client.Push(context.Background(), &request, fn); err != nil {
 		return err
 	}
 
-	if bar != nil && !bar.IsFinished() {
-		return errors.New("unexpected end to push model")
-	}
-
+	spinner.Stop()
 	return nil
 }
 
@@ -350,46 +363,51 @@ func PullHandler(cmd *cobra.Command, args []string) error {
 		return err
 	}
 
-	return pull(args[0], insecure)
-}
-
-func pull(model string, insecure bool) error {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
 
-	var currentDigest string
-	var bar *progressbar.ProgressBar
+	p := progress.NewProgress(os.Stderr)
+	defer p.Stop()
+
+	bars := make(map[string]*progress.Bar)
+
+	var status string
+	var spinner *progress.Spinner
 
-	request := api.PullRequest{Name: model, Insecure: insecure}
 	fn := func(resp api.ProgressResponse) error {
-		if resp.Digest != currentDigest && resp.Digest != "" {
-			currentDigest = resp.Digest
-			bar = progressbar.DefaultBytes(
-				resp.Total,
-				fmt.Sprintf("pulling %s...", resp.Digest[7:19]),
-			)
-
-			bar.Set64(resp.Completed)
-		} else if resp.Digest == currentDigest && resp.Digest != "" {
-			bar.Set64(resp.Completed)
-		} else {
-			currentDigest = ""
-			fmt.Println(resp.Status)
+		if resp.Digest != "" {
+			if spinner != nil {
+				spinner.Stop()
+			}
+
+			bar, ok := bars[resp.Digest]
+			if !ok {
+				bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
+				bars[resp.Digest] = bar
+				p.Add(resp.Digest, bar)
+			}
+
+			bar.Set(resp.Completed)
+		} else if status != resp.Status {
+			if spinner != nil {
+				spinner.Stop()
+			}
+
+			status = resp.Status
+			spinner = progress.NewSpinner(status)
+			p.Add(status, spinner)
 		}
 
 		return nil
 	}
 
+	request := api.PullRequest{Name: args[0], Insecure: insecure}
 	if err := client.Pull(context.Background(), &request, fn); err != nil {
 		return err
 	}
 
-	if bar != nil && !bar.IsFinished() {
-		return errors.New("unexpected end to pull model")
-	}
-
 	return nil
 }
 
@@ -442,8 +460,11 @@ func generate(cmd *cobra.Command, model, prompt string, wordWrap bool, format st
 		return err
 	}
 
-	spinner := NewSpinner("")
-	go spinner.Spin(60 * time.Millisecond)
+	p := progress.NewProgress(os.Stderr)
+	defer p.StopAndClear()
+
+	spinner := progress.NewSpinner("")
+	p.Add("", spinner)
 
 	var latest api.GenerateResponse
 
@@ -475,9 +496,7 @@ func generate(cmd *cobra.Command, model, prompt string, wordWrap bool, format st
 
 	request := api.GenerateRequest{Model: model, Prompt: prompt, Context: generateContext, Format: format}
 	fn := func(response api.GenerateResponse) error {
-		if !spinner.IsFinished() {
-			spinner.Finish()
-		}
+		p.StopAndClear()
 
 		latest = response
 
@@ -511,7 +530,6 @@ func generate(cmd *cobra.Command, model, prompt string, wordWrap bool, format st
 
 	if err := client.Generate(cancelCtx, &request, fn); err != nil {
 		if strings.Contains(err.Error(), "context canceled") && abort {
-			spinner.Finish()
 			return nil
 		}
 		return err

cmd/spinner.go

-package cmd
-
-import (
-	"fmt"
-	"os"
-	"time"
-
-	"github.com/jmorganca/ollama/progressbar"
-)
-
-type Spinner struct {
-	description string
-	*progressbar.ProgressBar
-}
-
-func NewSpinner(description string) *Spinner {
-	return &Spinner{
-		description: description,
-		ProgressBar: progressbar.NewOptions(-1,
-			progressbar.OptionSetWriter(os.Stderr),
-			progressbar.OptionThrottle(60*time.Millisecond),
-			progressbar.OptionSpinnerType(14),
-			progressbar.OptionSetRenderBlankState(true),
-			progressbar.OptionSetElapsedTime(false),
-			progressbar.OptionClearOnFinish(),
-			progressbar.OptionSetDescription(description),
-		),
-	}
-}
-
-func (s *Spinner) Spin(tick time.Duration) {
-	for range time.Tick(tick) {
-		if s.IsFinished() {
-			break
-		}
-
-		s.Add(1)
-	}
-}
-
-func (s *Spinner) Stop() {
-	s.Finish()
-	fmt.Println(s.description)
-}

docs/api.md

@@ -51,7 +51,9 @@ Advanced parameters (optional):
 
 ### JSON mode
 
-Enable JSON mode by setting the `format` parameter to `json` and specifying the model should use JSON in the `prompt`. This will structure the response as valid JSON. See the JSON mode [example](#request-json-mode) below.
+Enable JSON mode by setting the `format` parameter to `json`. This will structure the response as valid JSON. See the JSON mode [example](#request-json-mode) below.
+
+> Note: it's important to instruct the model to use JSON in the `prompt`. Otherwise, the model may generate large amounts whitespace.
 
 ### Examples
 

docs/faq.md

@@ -149,8 +149,8 @@ docker build -t ollama-with-ca .
 docker run -d -e HTTPS_PROXY=https://my.proxy.example.com -p 11434:11434 ollama-with-ca
 ```
 
-### How do I use Ollama with GPU acceleration in Docker?
+## How do I use Ollama with GPU acceleration in Docker?
 
-The Ollama Docker container can be configured with GPU acceleration in Linux or Windows (with WSL2). This requires the [nvidia-container-toolkit](https://github.com/NVIDIA/nvidia-container-toolkit).
+The Ollama Docker container can be configured with GPU acceleration in Linux or Windows (with WSL2). This requires the [nvidia-container-toolkit](https://github.com/NVIDIA/nvidia-container-toolkit). See [ollama/ollama](https://hub.docker.com/r/ollama/ollama) for more details.
 
 GPU acceleration is not available for Docker Desktop in macOS due to the lack of GPU passthrough and emulation.

docs/modelfile.md

@@ -41,6 +41,8 @@ INSTRUCTION arguments
 
 ## Examples
 
+### Basic `Modelfile`
+
 An example of a `Modelfile` creating a mario blueprint:
 
 ```modelfile
@@ -63,6 +65,35 @@ To use this:
 
 More examples are available in the [examples directory](../examples).
 
+### `Modelfile`s in [ollama.ai/library][1]
+
+There are two ways to view `Modelfile`s underlying the models in [ollama.ai/library][1]:
+
+- Option 1: view a details page from a model's tags page:
+   1. Go to a particular model's tags (e.g. https://ollama.ai/library/llama2/tags)
+   2. Click on a tag (e.g. https://ollama.ai/library/llama2:13b)
+   3. Scroll down to "Layers"
+      - Note: if the [`FROM` instruction](#from-required) is not present,
+        it means the model was created from a local file
+- Option 2: use `ollama show` to print the `Modelfile` like so:
+
+  ```bash
+  > ollama show --modelfile llama2:13b
+  # Modelfile generated by "ollama show"
+  # To build a new Modelfile based on this one, replace the FROM line with:
+  # FROM llama2:13b
+
+  FROM /root/.ollama/models/blobs/sha256:123abc
+  TEMPLATE """[INST] {{ if and .First .System }}<<SYS>>{{ .System }}<</SYS>>
+
+  {{ end }}{{ .Prompt }} [/INST] """
+  SYSTEM """"""
+  PARAMETER stop [INST]
+  PARAMETER stop [/INST]
+  PARAMETER stop <<SYS>>
+  PARAMETER stop <</SYS>>
+  ```
+
 ## Instructions
 
 ### FROM (Required)
@@ -177,3 +208,5 @@ LICENSE """
 
 - the **`Modelfile` is not case sensitive**. In the examples, we use uppercase for instructions to make it easier to distinguish it from arguments.
 - Instructions can be in any order. In the examples, we start with FROM instruction to keep it easily readable.
+
+[1]: https://ollama.ai/library

examples/jupyter-notebook/README.md

+# Ollama Jupyter Notebook
+
+This example downloads and installs Ollama in a Jupyter instance such as Google Colab. It will start the Ollama service and expose an endpoint using `ngrok` which can be used to communicate with the Ollama instance remotely.
+
+For best results, use an instance with GPU accelerator.

examples/jupyter-notebook/ollama.ipynb

+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "93f59dcb-c588-41b8-a792-55d88ade739c",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Download and run the Ollama Linux install script\n",
+    "!curl https://ollama.ai/install.sh | sh\n",
+    "!command -v systemctl >/dev/null && sudo systemctl stop ollama"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "658c147e-c7f8-490e-910e-62b80f577dda",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!pip install aiohttp pyngrok\n",
+    "\n",
+    "import os\n",
+    "import asyncio\n",
+    "from aiohttp import ClientSession\n",
+    "\n",
+    "# Set LD_LIBRARY_PATH so the system NVIDIA library becomes preferred\n",
+    "# over the built-in library. This is particularly important for \n",
+    "# Google Colab which installs older drivers\n",
+    "os.environ.update({'LD_LIBRARY_PATH': '/usr/lib64-nvidia'})\n",
+    "\n",
+    "async def run(cmd):\n",
+    "  '''\n",
+    "  run is a helper function to run subcommands asynchronously.\n",
+    "  '''\n",
+    "  print('>>> starting', *cmd)\n",
+    "  p = await asyncio.subprocess.create_subprocess_exec(\n",
+    "      *cmd,\n",
+    "      stdout=asyncio.subprocess.PIPE,\n",
+    "      stderr=asyncio.subprocess.PIPE,\n",
+    "  )\n",
+    "\n",
+    "  async def pipe(lines):\n",
+    "    async for line in lines:\n",
+    "      print(line.strip().decode('utf-8'))\n",
+    "\n",
+    "  await asyncio.gather(\n",
+    "      pipe(p.stdout),\n",
+    "      pipe(p.stderr),\n",
+    "  )\n",
+    "\n",
+    "\n",
+    "await asyncio.gather(\n",
+    "    run(['ollama', 'serve']),\n",
+    "    run(['ngrok', 'http', '--log', 'stderr', '11434']),\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "e7735a55-9aad-4caf-8683-52e2163ba53b",
+   "metadata": {},
+   "source": [
+    "The previous cell starts two processes, `ollama` and `ngrok`. The log output will show a line like the following which describes the external address.\n",
+    "\n",
+    "```\n",
+    "t=2023-11-12T22:55:56+0000 lvl=info msg=\"started tunnel\" obj=tunnels name=command_line addr=http://localhost:11434 url=https://8249-34-125-179-11.ngrok.io\n",
+    "```\n",
+    "\n",
+    "The external address in this case is `https://8249-34-125-179-11.ngrok.io` which can be passed into `OLLAMA_HOST` to access this instance.\n",
+    "\n",
+    "```bash\n",
+    "export OLLAMA_HOST=https://8249-34-125-179-11.ngrok.io\n",
+    "ollama list\n",
+    "ollama run mistral\n",
+    "```"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

format/bytes.go

 package format
 
-import "fmt"
+import (
+	"fmt"
+	"math"
+)
 
 const (
 	Byte     = 1
 	KiloByte = Byte * 1000
 	MegaByte = KiloByte * 1000
 	GigaByte = MegaByte * 1000
+	TeraByte = GigaByte * 1000
 )
 
 func HumanBytes(b int64) string {
+	var value float64
+	var unit string
+
 	switch {
-	case b > GigaByte:
-		return fmt.Sprintf("%.1f GB", float64(b)/GigaByte)
-	case b > MegaByte:
-		return fmt.Sprintf("%.1f MB", float64(b)/MegaByte)
-	case b > KiloByte:
-		return fmt.Sprintf("%.1f KB", float64(b)/KiloByte)
+	case b >= TeraByte:
+		value = float64(b) / TeraByte
+		unit = "TB"
+	case b >= GigaByte:
+		value = float64(b) / GigaByte
+		unit = "GB"
+	case b >= MegaByte:
+		value = float64(b) / MegaByte
+		unit = "MB"
+	case b >= KiloByte:
+		value = float64(b) / KiloByte
+		unit = "KB"
 	default:
 		return fmt.Sprintf("%d B", b)
 	}
+
+	switch {
+	case value >= 100:
+		return fmt.Sprintf("%d %s", int(value), unit)
+	case value != math.Trunc(value):
+		return fmt.Sprintf("%.1f %s", value, unit)
+	default:
+		return fmt.Sprintf("%d %s", int(value), unit)
+	}
 }

llm/llama.cpp/generate_darwin_amd64.go

@@ -7,13 +7,13 @@ package llm
 //go:generate git -C ggml apply ../patches/0002-34B-model-support.patch
 //go:generate git -C ggml apply ../patches/0003-metal-fix-synchronization-in-new-matrix-multiplicati.patch
 //go:generate git -C ggml apply ../patches/0004-metal-add-missing-barriers-for-mul-mat-2699.patch
-//go:generate cmake -S ggml -B ggml/build/cpu -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64 -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0
+//go:generate cmake -S ggml -B ggml/build/cpu -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_NAME=Darwin -DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64 -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0
 //go:generate cmake --build ggml/build/cpu --target server --config Release
 //go:generate mv ggml/build/cpu/bin/server ggml/build/cpu/bin/ollama-runner
 
 //go:generate git submodule update --force gguf
 //go:generate git -C gguf apply ../patches/0001-update-default-log-target.patch
 //go:generate git -C gguf apply ../patches/0001-metal-handle-ggml_scale-for-n-4-0-close-3754.patch
-//go:generate cmake -S gguf -B gguf/build/cpu -DLLAMA_METAL=off -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64 -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0 -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off
+//go:generate cmake -S gguf -B gguf/build/cpu -DLLAMA_METAL=off -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_NAME=Darwin -DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64 -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0 -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=on -DLLAMA_AVX512=off -DLLAMA_FMA=on -DLLAMA_F16C=on
 //go:generate cmake --build gguf/build/cpu --target server --config Release
 //go:generate mv gguf/build/cpu/bin/server gguf/build/cpu/bin/ollama-runner

llm/llama.go

@@ -226,7 +226,7 @@ type llama struct {
 }
 
 var (
-	errNvidiaSMI     = errors.New("nvidia-smi command failed")
+	errNvidiaSMI     = errors.New("warning: gpu support may not be enabled, check that you have installed GPU drivers: nvidia-smi command failed")
 	errAvailableVRAM = errors.New("not enough VRAM available, falling back to CPU only")
 )
 
@@ -343,6 +343,10 @@ func newLlama(model string, adapters []string, runners []ModelRunner, numLayers
 		"--embedding",
 	}
 
+	if opts.MainGPU > 0 {
+		params = append(params, "--main-gpu", fmt.Sprintf("%d", opts.MainGPU))
+	}
+
 	if opts.RopeFrequencyBase > 0 {
 		params = append(params, "--rope-freq-base", fmt.Sprintf("%f", opts.RopeFrequencyBase))
 	}
@@ -544,6 +548,7 @@ func (llm *llama) Predict(ctx context.Context, prevContext []int, prompt string,
 		"stream":            true,
 		"n_predict":         llm.NumPredict,
 		"n_keep":            llm.NumKeep,
+		"main_gpu":          llm.MainGPU,
 		"temperature":       llm.Temperature,
 		"top_k":             llm.TopK,
 		"top_p":             llm.TopP,

llm/llm.go

@@ -41,20 +41,13 @@ func New(workDir, model string, adapters []string, opts api.Options) (LLM, error
 
 	if runtime.GOOS == "darwin" {
 		switch ggml.FileType() {
-		case "Q8_0":
+		case "F32", "Q5_0", "Q5_1", "Q8_0":
 			if ggml.Name() != "gguf" && opts.NumGPU != 0 {
 				// GGML Q8_0 do not support Metal API and will
 				// cause the runner to segmentation fault so disable GPU
 				log.Printf("WARNING: GPU disabled for F32, Q5_0, Q5_1, and Q8_0")
 				opts.NumGPU = 0
 			}
-		case "F32", "Q5_0", "Q5_1":
-			if opts.NumGPU != 0 {
-				// F32, Q5_0, Q5_1, and Q8_0 do not support Metal API and will
-				// cause the runner to segmentation fault so disable GPU
-				log.Printf("WARNING: GPU disabled for F32, Q5_0, Q5_1, and Q8_0")
-				opts.NumGPU = 0
-			}
 		}
 
 		var requiredMemory int64

progress/bar.go

+package progress
+
+import (
+	"fmt"
+	"math"
+	"os"
+	"strings"
+	"time"
+
+	"github.com/jmorganca/ollama/format"
+	"golang.org/x/term"
+)
+
+type Stats struct {
+	rate      int64
+	value     int64
+	remaining time.Duration
+}
+
+type Bar struct {
+	message      string
+	messageWidth int
+
+	maxValue     int64
+	initialValue int64
+	currentValue int64
+
+	started time.Time
+
+	stats   Stats
+	statted time.Time
+}
+
+func NewBar(message string, maxValue, initialValue int64) *Bar {
+	return &Bar{
+		message:      message,
+		messageWidth: -1,
+		maxValue:     maxValue,
+		initialValue: initialValue,
+		currentValue: initialValue,
+		started:      time.Now(),
+	}
+}
+
+func (b *Bar) String() string {
+	termWidth, _, err := term.GetSize(int(os.Stderr.Fd()))
+	if err != nil {
+		termWidth = 80
+	}
+
+	var pre, mid, suf strings.Builder
+
+	if b.message != "" {
+		message := strings.TrimSpace(b.message)
+		if b.messageWidth > 0 && len(message) > b.messageWidth {
+			message = message[:b.messageWidth]
+		}
+
+		fmt.Fprintf(&pre, "%s", message)
+		if b.messageWidth-pre.Len() >= 0 {
+			pre.WriteString(strings.Repeat(" ", b.messageWidth-pre.Len()))
+		}
+
+		pre.WriteString(" ")
+	}
+
+	fmt.Fprintf(&pre, "%3.0f%% ", math.Floor(b.percent()))
+	fmt.Fprintf(&suf, "(%s/%s", format.HumanBytes(b.currentValue), format.HumanBytes(b.maxValue))
+
+	stats := b.Stats()
+	rate := int64(stats.rate)
+	if rate > 0 {
+		fmt.Fprintf(&suf, ", %s/s", format.HumanBytes(rate))
+	}
+
+	fmt.Fprintf(&suf, ")")
+
+	elapsed := time.Since(b.started)
+	if b.percent() < 100 && rate > 0 {
+		fmt.Fprintf(&suf, " [%s:%s]", elapsed.Round(time.Second), stats.remaining)
+	} else {
+		fmt.Fprintf(&suf, "        ")
+	}
+
+	mid.WriteString("▕")
+
+	// add 3 extra spaces: 2 boundary characters and 1 space at the end
+	f := termWidth - pre.Len() - suf.Len() - 3
+	n := int(float64(f) * b.percent() / 100)
+
+	if n > 0 {
+		mid.WriteString(strings.Repeat("█", n))
+	}
+
+	if f-n > 0 {
+		mid.WriteString(strings.Repeat(" ", f-n))
+	}
+
+	mid.WriteString("▏")
+
+	return pre.String() + mid.String() + suf.String()
+}
+
+func (b *Bar) Set(value int64) {
+	if value >= b.maxValue {
+		value = b.maxValue
+	}
+
+	b.currentValue = value
+}
+
+func (b *Bar) percent() float64 {
+	if b.maxValue > 0 {
+		return float64(b.currentValue) / float64(b.maxValue) * 100
+	}
+
+	return 0
+}
+
+func (b *Bar) Stats() Stats {
+	if time.Since(b.statted) < time.Second {
+		return b.stats
+	}
+
+	switch {
+	case b.statted.IsZero():
+		b.stats = Stats{
+			value:     b.initialValue,
+			rate:      0,
+			remaining: 0,
+		}
+	case b.currentValue >= b.maxValue:
+		b.stats = Stats{
+			value:     b.maxValue,
+			rate:      0,
+			remaining: 0,
+		}
+	default:
+		rate := b.currentValue - b.stats.value
+		var remaining time.Duration
+		if rate > 0 {
+			remaining = time.Second * time.Duration((float64(b.maxValue-b.currentValue))/(float64(rate)))
+		}
+
+		b.stats = Stats{
+			value:     b.currentValue,
+			rate:      rate,
+			remaining: remaining,
+		}
+	}
+
+	b.statted = time.Now()
+
+	return b.stats
+}

progress/progress.go

+package progress
+
+import (
+	"fmt"
+	"io"
+	"sync"
+	"time"
+)
+
+type State interface {
+	String() string
+}
+
+type Progress struct {
+	mu sync.Mutex
+	w  io.Writer
+
+	pos int
+
+	ticker *time.Ticker
+	states []State
+}
+
+func NewProgress(w io.Writer) *Progress {
+	p := &Progress{w: w}
+	go p.start()
+	return p
+}
+
+func (p *Progress) stop() bool {
+	for _, state := range p.states {
+		if spinner, ok := state.(*Spinner); ok {
+			spinner.Stop()
+		}
+	}
+
+	if p.ticker != nil {
+		p.ticker.Stop()
+		p.ticker = nil
+		p.render()
+		return true
+	}
+
+	return false
+}
+
+func (p *Progress) Stop() bool {
+	stopped := p.stop()
+	if stopped {
+		fmt.Fprint(p.w, "\n")
+	}
+	return stopped
+}
+
+func (p *Progress) StopAndClear() bool {
+	fmt.Fprint(p.w, "\033[?25l")
+	defer fmt.Fprint(p.w, "\033[?25h")
+
+	stopped := p.stop()
+	if stopped {
+		// clear all progress lines
+		for i := 0; i < p.pos; i++ {
+			if i > 0 {
+				fmt.Fprint(p.w, "\033[A")
+			}
+			fmt.Fprint(p.w, "\033[2K\033[1G")
+		}
+	}
+
+	return stopped
+}
+
+func (p *Progress) Add(key string, state State) {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+
+	p.states = append(p.states, state)
+}
+
+func (p *Progress) render() error {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+
+	fmt.Fprint(p.w, "\033[?25l")
+	defer fmt.Fprint(p.w, "\033[?25h")
+
+	// clear already rendered progress lines
+	for i := 0; i < p.pos; i++ {
+		if i > 0 {
+			fmt.Fprint(p.w, "\033[A")
+		}
+		fmt.Fprint(p.w, "\033[2K\033[1G")
+	}
+
+	// render progress lines
+	for i, state := range p.states {
+		fmt.Fprint(p.w, state.String())
+		if i < len(p.states)-1 {
+			fmt.Fprint(p.w, "\n")
+		}
+	}
+
+	p.pos = len(p.states)
+
+	return nil
+}
+
+func (p *Progress) start() {
+	p.ticker = time.NewTicker(100 * time.Millisecond)
+	for range p.ticker.C {
+		p.render()
+	}
+}

progress/spinner.go

+package progress
+
+import (
+	"fmt"
+	"strings"
+	"time"
+)
+
+type Spinner struct {
+	message      string
+	messageWidth int
+
+	parts []string
+
+	value int
+
+	ticker  *time.Ticker
+	started time.Time
+	stopped time.Time
+}
+
+func NewSpinner(message string) *Spinner {
+	s := &Spinner{
+		message: message,
+		parts: []string{
+			"⠋", "⠙", "⠹", "⠸", "⠼", "⠴", "⠦", "⠧", "⠇", "⠏",
+		},
+		started: time.Now(),
+	}
+	go s.start()
+	return s
+}
+
+func (s *Spinner) String() string {
+	var sb strings.Builder
+	if len(s.message) > 0 {
+		message := strings.TrimSpace(s.message)
+		if s.messageWidth > 0 && len(message) > s.messageWidth {
+			message = message[:s.messageWidth]
+		}
+
+		fmt.Fprintf(&sb, "%s", message)
+		if s.messageWidth-sb.Len() >= 0 {
+			sb.WriteString(strings.Repeat(" ", s.messageWidth-sb.Len()))
+		}
+
+		sb.WriteString(" ")
+	}
+
+	if s.stopped.IsZero() {
+		spinner := s.parts[s.value]
+		sb.WriteString(spinner)
+		sb.WriteString(" ")
+	}
+
+	return sb.String()
+}
+
+func (s *Spinner) start() {
+	s.ticker = time.NewTicker(100 * time.Millisecond)
+	for range s.ticker.C {
+		s.value = (s.value + 1) % len(s.parts)
+		if !s.stopped.IsZero() {
+			return
+		}
+	}
+}
+
+func (s *Spinner) Stop() {
+	if s.stopped.IsZero() {
+		s.stopped = time.Now()
+	}
+}

progressbar/LICENSE

-MIT License
-
-Copyright (c) 2017 Zack
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.

progressbar/README.md

-# progressbar
-
-[![CI](https://github.com/schollz/progressbar/actions/workflows/ci.yml/badge.svg?branch=main&event=push)](https://github.com/schollz/progressbar/actions/workflows/ci.yml)
-[![go report card](https://goreportcard.com/badge/github.com/schollz/progressbar)](https://goreportcard.com/report/github.com/schollz/progressbar) 
-[![coverage](https://img.shields.io/badge/coverage-84%25-brightgreen.svg)](https://gocover.io/github.com/schollz/progressbar)
-[![godocs](https://godoc.org/github.com/schollz/progressbar?status.svg)](https://godoc.org/github.com/schollz/progressbar/v3) 
-
-A very simple thread-safe progress bar which should work on every OS without problems. I needed a progressbar for [croc](https://github.com/schollz/croc) and everything I tried had problems, so I made another one. In order to be OS agnostic I do not plan to support [multi-line outputs](https://github.com/schollz/progressbar/issues/6).
-
-
-## Install
-
-```
-go get -u github.com/schollz/progressbar/v3
-```
-
-## Usage 
-
-### Basic usage
-
-```golang
-bar := progressbar.Default(100)
-for i := 0; i < 100; i++ {
-    bar.Add(1)
-    time.Sleep(40 * time.Millisecond)
-}
-```
-
-which looks like:
-
-![Example of basic bar](examples/basic/basic.gif)
-
-
-### I/O operations
-
-The `progressbar` implements an `io.Writer` so it can automatically detect the number of bytes written to a stream, so you can use it as a progressbar for an `io.Reader`.
-
-```golang
-req, _ := http.NewRequest("GET", "https://dl.google.com/go/go1.14.2.src.tar.gz", nil)
-resp, _ := http.DefaultClient.Do(req)
-defer resp.Body.Close()
-
-f, _ := os.OpenFile("go1.14.2.src.tar.gz", os.O_CREATE|os.O_WRONLY, 0644)
-defer f.Close()
-
-bar := progressbar.DefaultBytes(
-    resp.ContentLength,
-    "downloading",
-)
-io.Copy(io.MultiWriter(f, bar), resp.Body)
-```
-
-which looks like:
-
-![Example of download bar](examples/download/download.gif)
-
-
-### Progress bar with unknown length
-
-A progressbar with unknown length is a spinner. Any bar with -1 length will automatically convert it to a spinner with a customizable spinner type. For example, the above code can be run and set the `resp.ContentLength` to `-1`.
-
-which looks like:
-
-![Example of download bar with unknown length](examples/download-unknown/download-unknown.gif)
-
-
-### Customization
-
-There is a lot of customization that you can do - change the writer, the color, the width, description, theme, etc. See [all the options](https://pkg.go.dev/github.com/schollz/progressbar/v3?tab=doc#Option).
-
-```golang
-bar := progressbar.NewOptions(1000,
-    progressbar.OptionSetWriter(ansi.NewAnsiStdout()),
-    progressbar.OptionEnableColorCodes(true),
-    progressbar.OptionShowBytes(true),
-    progressbar.OptionSetWidth(15),
-    progressbar.OptionSetDescription("[cyan][1/3][reset] Writing moshable file..."),
-    progressbar.OptionSetTheme(progressbar.Theme{
-        Saucer:        "[green]=[reset]",
-        SaucerHead:    "[green]>[reset]",
-        SaucerPadding: " ",
-        BarStart:      "[",
-        BarEnd:        "]",
-    }))
-for i := 0; i < 1000; i++ {
-    bar.Add(1)
-    time.Sleep(5 * time.Millisecond)
-}
-```
-
-which looks like:
-
-![Example of customized bar](examples/customization/customization.gif)
-
-
-## Contributing
-
-Pull requests are welcome. Feel free to...
-
-- Revise documentation
-- Add new features
-- Fix bugs
-- Suggest improvements
-
-## Thanks
-
-Thanks [@Dynom](https://github.com/dynom) for massive improvements in version 2.0!
-
-Thanks [@CrushedPixel](https://github.com/CrushedPixel) for adding descriptions and color code support!
-
-Thanks [@MrMe42](https://github.com/MrMe42) for adding some minor features!
-
-Thanks [@tehstun](https://github.com/tehstun) for some great PRs!
-
-Thanks [@Benzammour](https://github.com/Benzammour) and [@haseth](https://github.com/haseth) for helping create v3!
-
-Thanks [@briandowns](https://github.com/briandowns) for compiling the list of spinners.
-
-## License
-
-MIT