Linux uninstall instructions (#894)

Separate GGUF and PyTorch guides

docs/import.md

@@ -1,8 +1,43 @@
 # Import a model
 
-This guide walks through importing a PyTorch, Safetensors or GGUF model.
+This guide walks through importing a GGUF, PyTorch or Safetensors model.
 
-## Supported models
+## Importing (GGUF)
+
+### Step 1: Write a `Modelfile`
+
+Start by creating a `Modelfile`. This file is the blueprint for your model, specifying weights, parameters, prompt templates and more.
+
+```
+FROM ./mistral-7b-v0.1.Q4_0.gguf
+```
+
+(Optional) many chat models require a prompt template in order to answer correctly. A default prompt template can be specified with the `TEMPLATE` instruction in the `Modelfile`:
+
+```
+FROM ./q4_0.bin
+TEMPLATE "[INST] {{ .Prompt }} [/INST]"
+```
+
+### Step 2: Create the Ollama model
+
+Finally, create a model from your `Modelfile`:
+
+```
+ollama create example -f Modelfile
+```
+
+### Step 3: Run your model
+
+Next, test the model with `ollama run`:
+
+```
+ollama run example "What is your favourite condiment?"
+```
+
+## Importing (PyTorch & Safetensors)
+
+### Supported models
 
 Ollama supports a set of model architectures, with support for more coming soon:
 
@@ -13,8 +48,6 @@ Ollama supports a set of model architectures, with support for more coming soon:
 
 To view a model's architecture, check the `config.json` file in its HuggingFace repo. You should see an entry under `architectures` (e.g. `LlamaForCausalLM`).
 
-## Importing
-
 ### Step 1: Clone the HuggingFace repository (optional)
 
 If the model is currently hosted in a HuggingFace repository, first clone that repository to download the raw model.
@@ -44,7 +77,7 @@ This will output two files into the directory:
 
 ### Step 3: Write a `Modelfile`
 
-Next, create a `Modelfile` for your model. This file is the blueprint for your model, specifying weights, parameters, prompt templates and more.
+Next, create a `Modelfile` for your model:
 
 ```
 FROM ./q4_0.bin
@@ -65,13 +98,15 @@ Finally, create a model from your `Modelfile`:
 ollama create example -f Modelfile
 ```
 
+### Step 5: Run your model
+
 Next, test the model with `ollama run`:
 
 ```
 ollama run example "What is your favourite condiment?"
 ```
 
-### Step 5: Publish your model (optional – early alpha)
+## Publishing your model (optional – early alpha)
 
 Publishing models is in early alpha. If you'd like to publish your model to share with others, follow these steps:
 

docs/linux.md

@@ -80,3 +80,23 @@ To view logs of Ollama running as a startup service, run:
 ```bash
 journalctl -u ollama
 ```
+
+## Uninstall
+
+Remove the ollama service:
+```bash
+systemctl stop ollama
+systemctl disable ollama
+rm /etc/systemd/system/ollama.service
+```
+
+Remove the ollama binary from your bin directory (either `/usr/local/bin`, `/usr/bin`, or `/bin`):
+```bash
+rm /usr/local/bin/ollama
+```
+
+Remove the downloaded models and Ollama service user:
+```bash
+rm /usr/share/ollama
+userdel ollama
+```

llm/llama.cpp/generate_darwin_amd64.go

@@ -13,6 +13,7 @@ package llm
 
 //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_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 --build gguf/build/cpu --target server --config Release
 //go:generate mv gguf/build/cpu/bin/server gguf/build/cpu/bin/ollama-runner

llm/llama.cpp/generate_darwin_arm64.go

@@ -13,6 +13,7 @@ package llm
 
 //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/metal -DLLAMA_METAL=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_PROCESSOR=arm64 -DCMAKE_OSX_ARCHITECTURES=arm64 -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0
 //go:generate cmake --build gguf/build/metal --target server --config Release
 //go:generate mv gguf/build/metal/bin/server gguf/build/metal/bin/ollama-runner

llm/llama.cpp/patches/0001-metal-handle-ggml_scale-for-n-4-0-close-3754.patch

@@ -0,0 +1,91 @@
+From 469c9addef75893e6be12edda852d12e840bf064 Mon Sep 17 00:00:00 2001
+From: Georgi Gerganov <ggerganov@gmail.com>
+Date: Tue, 24 Oct 2023 09:46:50 +0300
+Subject: [PATCH 1/2] metal : handle ggml_scale for n%4 != 0 (close #3754)
+
+ggml-ci
+---
+ ggml-metal.m     | 18 +++++++++++++-----
+ ggml-metal.metal | 10 +++++++++-
+ 2 files changed, 22 insertions(+), 6 deletions(-)
+
+diff --git a/ggml-metal.m b/ggml-metal.m
+index c908106..c1901dc 100644
+--- a/ggml-metal.m
++++ b/ggml-metal.m
+@@ -62,6 +62,7 @@
+     GGML_METAL_DECL_KERNEL(mul);
+     GGML_METAL_DECL_KERNEL(mul_row); // TODO: avoid this extra kernel, instead extend the "mul" kernel to support broadcast
+     GGML_METAL_DECL_KERNEL(scale);
++    GGML_METAL_DECL_KERNEL(scale_4);
+     GGML_METAL_DECL_KERNEL(silu);
+     GGML_METAL_DECL_KERNEL(relu);
+     GGML_METAL_DECL_KERNEL(gelu);
+@@ -249,6 +250,7 @@ static void ggml_metal_log(enum ggml_log_level level, const char* format, ...){
+         GGML_METAL_ADD_KERNEL(mul);
+         GGML_METAL_ADD_KERNEL(mul_row);
+         GGML_METAL_ADD_KERNEL(scale);
++        GGML_METAL_ADD_KERNEL(scale_4);
+         GGML_METAL_ADD_KERNEL(silu);
+         GGML_METAL_ADD_KERNEL(relu);
+         GGML_METAL_ADD_KERNEL(gelu);
+@@ -347,6 +349,7 @@ void ggml_metal_free(struct ggml_metal_context * ctx) {
+     GGML_METAL_DEL_KERNEL(mul);
+     GGML_METAL_DEL_KERNEL(mul_row);
+     GGML_METAL_DEL_KERNEL(scale);
++    GGML_METAL_DEL_KERNEL(scale_4);
+     GGML_METAL_DEL_KERNEL(silu);
+     GGML_METAL_DEL_KERNEL(relu);
+     GGML_METAL_DEL_KERNEL(gelu);
+@@ -923,15 +926,20 @@ void ggml_metal_graph_compute(
+ 
+                             const float scale = *(const float *) src1->data;
+ 
+-                            [encoder setComputePipelineState:ctx->pipeline_scale];
++                            int64_t n = ggml_nelements(dst);
++
++                            if (n % 4 == 0) {
++                                n /= 4;
++                                [encoder setComputePipelineState:ctx->pipeline_scale_4];
++                            } else {
++                                [encoder setComputePipelineState:ctx->pipeline_scale];
++                            }
++
+                             [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
+                             [encoder setBuffer:id_dst  offset:offs_dst  atIndex:1];
+                             [encoder setBytes:&scale length:sizeof(scale) atIndex:2];
+ 
+-                            const int64_t n = ggml_nelements(dst);
+-                            GGML_ASSERT(n % 4 == 0);
+-
+-                            [encoder dispatchThreadgroups:MTLSizeMake(n/4, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
++                            [encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
+                         } break;
+                     case GGML_OP_UNARY:
+                         switch (ggml_get_unary_op(gf->nodes[i])) {
+diff --git a/ggml-metal.metal b/ggml-metal.metal
+index 69fc713..f4b4605 100644
+--- a/ggml-metal.metal
++++ b/ggml-metal.metal
+@@ -125,9 +125,17 @@ kernel void kernel_mul_row(
+ }
+ 
+ kernel void kernel_scale(
++        device const float * src0,
++        device       float * dst,
++        constant     float & scale,
++        uint tpig[[thread_position_in_grid]]) {
++    dst[tpig] = src0[tpig] * scale;
++}
++
++kernel void kernel_scale_4(
+         device const float4 * src0,
+         device       float4 * dst,
+-        constant     float & scale,
++        constant     float  & scale,
+         uint tpig[[thread_position_in_grid]]) {
+     dst[tpig] = src0[tpig] * scale;
+ }
+-- 
+2.39.3 (Apple Git-145)
+

scripts/install.sh

@@ -74,6 +74,9 @@ configure_systemd() {
         $SUDO useradd -r -s /bin/false -m -d /usr/share/ollama ollama
     fi
 
+    status "Adding current user to ollama group..."
+    $SUDO usermod -a -G ollama $(whoami)
+
     status "Creating ollama systemd service..."
     cat <<EOF | $SUDO tee /etc/systemd/system/ollama.service >/dev/null
 [Unit]