ollama model download and test script

download_test_models.sh

@@ -0,0 +1,132 @@
+#!/bin/bash
+
+# Display help information
+usage() {
+    echo "Usage: $0 [Options] [Arguments...]"
+    echo ""
+    echo "Options:"
+    echo "  -h,  --help                Show this help message"
+    echo "  -m,  --models              Specify one or more models, separated by ','"
+    echo "  -f,  --file FILE           Specify the input file(.txt),with model names separated by spaces or newlines(default: all models of the ollama library)"
+    echo "  -s,  --maxsize             Specify the maximum size of the model that can be downloaded (default: 40GB)"
+    echo "  -d, --down_model_path      Specify the model download directory (default: current directory)"
+    echo "  -b, --llama-bench_path     Absolute path of llama-bench"
+    exit 0
+}
+
+# default value
+file_name=""
+models=""
+maxsize=40
+model_path="."
+llama_bench_path=""
+
+declare -a model_names
+
+# Parsing command line arguments
+while getopts "hm:f:s:d:b:" opt; do
+    case $opt in
+        h)
+            usage
+            exit 0
+            ;;
+	    m)
+            models=$OPTARG
+            ;;
+        f)
+            file_name=$OPTARG
+            if [[ ! -f "$file_name" ]]; then
+            echo "Error: File '$file_name' not found!"
+                    exit 1
+            fi
+            ;;
+        s)
+            maxsize=$OPTARG
+            ;;
+	    d)
+            model_path=$OPTARG
+            ;;
+	    b)
+            llama_bench_path=$OPTARG
+            ;;
+        *)
+            usage
+            exit 1
+            ;;
+    esac
+done
+
+# Check if omdd exists in the current directory
+if [ ! -f "./omdd" ];then
+    wget -O omdd https://github.com/amirrezaDev1378/ollama-model-direct-download/releases/download/v2.1.1/omdd-linux-arm64
+    chmod +x omdd
+fi
+
+# acquire model name
+if [ ! -z "$models" ]; then
+    IFS=',' read -r -a model_names <<< "$models"
+
+elif [ ! -z "$file_name" ];then
+    echo "Reading contents of $file_name"
+    model_names=($(cat $file_name))
+
+else
+    url="https://ollama.com/search"
+    model_names=($(curl -s "$url" | grep "x-test-search-response-title" | awk -F '>' '{print $2}' | awk -F '<' '{print $1}'))
+
+fi
+
+
+if [ -f "./model_urls.txt" ]; then
+    rm "./model_urls.txt"
+fi
+
+# Filter tags with the smallest number of parameters according to the model name.
+for model_name in "${model_names[@]}";do
+	url="https://ollama.com/library/$model_name/tags"
+   	echo "Processing: $url, model_name = $model_name"
+	model_tag=($(curl -s "$url" | grep "library/$model_name:"  | grep -oP '(?<=/library/)[^"]+' | grep -E 'q4_0|q8_0|fp16' | sort -t ':' -k2,2 -n | perl -ne 'print if /q4_0/ && !$q4++; print if /q8_0/ && !$q8++; print if /fp16/ && !$fp16++; exit if $q4 && $q8 && $fp16'))
+	for tag in "${model_tag[@]}";do
+		echo "$tag `./omdd get $tag | grep  '1 -' | awk '{print $3}'`" >> model_urls.txt
+	done
+done
+
+
+all_model_args=""
+
+# Limiting the size of downloaded files
+max_size=$(($maxsize * 1024 * 1024 * 1024)) 
+while read -r name url; do
+    echo "Checking size for $name..."
+
+    size=$(curl -sI "$url" | grep -i "Content-Length" | cut -d ' ' -f2 | tr -d '\r')
+
+    if [[ -z "$size" ]]; then
+        echo "Warning: Unable to determine size for $name. Skipping..."
+        continue
+    fi
+
+    if ((size > max_size)); then
+	    echo "Skipping $name, file size exceeds $maxsize GB ($((size / 1024 / 1024 / 1024)) GB)"
+        continue
+    fi
+
+    all_model_args="$all_model_args -m $model_path/$name"
+    echo "Downloading $name from $url (Size: $((size / 1024 / 1024)) MB)..."
+    if [[ -f "$model_path/$name" ]]; then
+        local_size=$(stat -c %s "$model_path/$name")
+	echo $local_size
+        if ((local_size == size)); then
+            echo "$name is already fully downloaded. Skipping..."
+	    continue
+        fi
+    fi
+
+    wget -c --progress=bar -P "$model_path" -O "$model_path/$name" "$url"
+
+done < model_urls.txt
+
+#echo $llama_bench_path $all_model_args -p \"what is a car?\" -ngl 99 -o csv
+llama_bench_res=$($llama_bench_path $all_model_args -p "what is a car?" -ngl 99 -o csv)
+
+echo $llama_bench_res >> llama_bench_data.csv

hf-models.txt

@@ -6,23 +6,29 @@ microsoft/Phi-3.5-MoE-instruct
 
 aisingapore/Llama-SEA-LION-v3-8B-IT
 
-allenai/OLMoE-1B-7B-0924
+allenai/OLMoE-1B-7B-0924-Instruct-GGUF/olmoe-1b-7b-0924-instruct-f16.gguf
+allenai/OLMoE-1B-7B-0924-Instruct-GGUF/olmoe-1b-7b-0924-instruct-q4_0.gguf
+allenai/OLMoE-1B-7B-0924-Instruct-GGUF/olmoe-1b-7b-0924-instruct-q8_0.gguf
 
-abeja/gpt-neox-japanese-2.7b
+mmnga/stockmark-gpt-neox-japanese-1.4b-gguf/stockmark-gpt-neox-japanese-1.4b-q4_0.gguf
+mmnga/stockmark-gpt-neox-japanese-1.4b-gguf/stockmark-gpt-neox-japanese-1.4b-q8_0.gguf
 EleutherAI/pythia-6.9b
 
 abacusai/Llama-3-Smaug-8B
 TheBloke/Poro-34B-GGUF/poro-34b.Q4_0.gguf
 TheBloke/Poro-34B-GGUF/poro-34b.Q8_0.gguf
 1bitLLM/bitnet_b1_58-xl
 google/flan-t5-base
-apple/OpenELM-1_1B-Instruct
+QuietImpostor/OpenELM-3B-Instruct-GGUFs/OpenELM-3B-Instruct-F16.gguf
+QuietImpostor/OpenELM-3B-Instruct-GGUFs/OpenELM-3B-Instruct-Q4_0.gguf
+QuietImpostor/OpenELM-3B-Instruct-GGUFs/OpenELM-3B-Instruct-Q8_0.gguf
 THUDM/glm-edge-4b-chat
 THUDM/GLM-4-9B-0414
 HuggingFaceTB/SmolLM-135M
 LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct
 tiiuae/falcon-mamba-7b
-erfanvaredi/jais-7b-chat
+mradermacher/jais-family-1p3b-chat-GGUF/jais-family-1p3b-chat.f16.gguf
+mradermacher/jais-family-1p3b-chat-GGUF/jais-family-1p3b-chat.Q8_0.gguf
 speakleash/Bielik-11B-v2.3-Instruct
 bartowski/rwkv-6-world-7b-GGUF/rwkv-6-world-7b-Q4_0.gguf
 recursal/QRWKV6-32B-Instruct-Preview-v0.1
@@ -33,16 +39,20 @@ inclusionAI/Ling-lite
 
 
 
-
-liuhaotian/llava-v1.6-mistral-7b
-SkunkworksAI/BakLLaVA-1
+llava-hf/llava-v1.6-34b-hf
+llava-hf/llava-1.5-7b-hf
+cjpais/llava-v1.6-vicuna-7b-gguf/llava-v1.6-vicuna-7b.Q8_0.gguf
+abetlen/BakLLaVA-1-GGUF/bakllava-1.Q4_0.gguf
+abetlen/BakLLaVA-1-GGUF/bakllava-1.Q8_0.gguf
+abetlen/BakLLaVA-1-GGUF/bakllava-1.f16.gguf
 NousResearch/Obsidian-3B-V0.5
-Lin-Chen/ShareGPT4V-7B
-mtgv/MobileVLM_V2-1.7B
-01-ai/Yi-VL-6B
+NousResearch/Obsidian-3B-V0.5-GGUF/obsidian-f16.gguf
+Galunid/ShareGPT4V-gguf/ShareGPT4V-f16.gguf
+ZiangWu/MobileVLM_V2-1.7B-GGUF/mmproj-model-f16.gguf
+cmp-nct/Yi-VL-34B-GGUF/mmproj-model-f16.gguf
 openbmb/MiniCPM-o-2_6-gguf/Model-7.6B-Q8_0.gguf
 openbmb/MiniCPM-o-2_6-gguf/Model-7.6B-F16.gguf
 vikhyatk/moondream2
-BAAI/Bunny-v1_0-3B
+BAAI/Bunny-v1_0-4B-gguf/ggml-model-f16.gguf
 THUDM/glm-edge-1.5b-chat
 Qwen/Qwen2-VL-2B-Instruct

hf_model_dowload.sh

@@ -146,7 +146,7 @@ END
                     fi
                 fi
 
-                run_inference_with_expect "$gguf_path" "what is a car?" 30 "$llamaCpp_path/build/bin/llama-cli" > /dev/null 2>&1
+                run_inference_with_expect "$gguf_path" "what is a car?" 60 "$llamaCpp_path/build/bin/llama-cli" > /dev/null 2>&1
                 ret=$?
                 if [ $ret -eq 0 ]; then
                     echo "$model Success inference"
@@ -160,7 +160,7 @@ END
                 download_path="$model_path/hf-models/$model_name"
 
                 echo "============Attempting to download model: $model (Attempt: $((attempt + 1)))============"
-                huggingface-cli download "$model" --local-dir "$download_path" > /de v/null 2>&1
+                huggingface-cli download "$model" --local-dir "$download_path" > /dev/null 2>&1
                 status=$?
 
                 if [ $status -eq 0 ]; then
@@ -196,7 +196,7 @@ END
                             fi
                         fi
 
-                        run_inference_with_expect "$quant_out" "what is a car?" 30 "$llamaCpp_path/build/bin/llama-cli" > /dev/null 2>&1
+                        run_inference_with_expect "$quant_out" "what is a car?" 60 "$llamaCpp_path/build/bin/llama-cli" > /dev/null 2>&1
                         ret=$?
                         if [ $ret -eq 0 ]; then
                             echo "$model Success inference ($quant_type)"