use bigendian models when testing bigendian arches

Signed-off-by: Mike Bonnet <[email protected]>

Author: mikebonnet

shortnames/shortnames.conf

@@ -24,6 +24,8 @@
   "granite:2b" = "ollama://granite3.1-dense:2b"
   "granite:7b" = "huggingface://instructlab/granite-7b-lab-GGUF/granite-7b-lab-Q4_K_M.gguf"
   "granite:8b" = "ollama://granite3.1-dense:8b"
+  "granite-be-3.0:1b" = "hf://taronaeo/Granite-3.0-1B-A400M-Instruct-BE-GGUF/granite-3.0-1b-a400m-instruct-be.Q2_K.gguf"
+  "granite-be-3.3:2b" = "hf://taronaeo/Granite-3.3-2B-Instruct-BE-GGUF/granite-3.3-2b-instruct-be.Q4_K_M.gguf"
   "hermes" = "huggingface://NousResearch/Hermes-2-Pro-Mistral-7B-GGUF/Hermes-2-Pro-Mistral-7B.Q4_K_M.gguf"
   "ibm/granite" = "ollama://granite3.1-dense:8b"
   "ibm/granite:2b" = "ollama://granite3.1-dense:2b"

test/system/002-bench.bats

@@ -16,7 +16,10 @@ function setup() {
 # bats test_tags=distro-integration
 @test "ramalama bench" {
     skip_if_no_llama_bench
-    run_ramalama bench -t 2 smollm:135m
+    if is_s390x; then
+        local RAMALAMA_TIMEOUT=1200
+    fi
+    run_ramalama bench -t 2 $(test_model smollm:135m)
     is "$output" ".*model.*size.*" "model and size in output"
 }
 

test/system/030-run.bats

@@ -125,19 +125,20 @@ EOF
     is "$output" ".*-e HSA_OVERRIDE_GFX_VERSION=0.0.0" "ensure HSA_OVERRIDE_GFX_VERSION is set from environment"
 }
 
-@test "ramalama run smollm with prompt" {
-    run_ramalama run --temp 0 ${MODEL} "What is the first line of the declaration of independence?"
+@test "ramalama run with prompt" {
+    run_ramalama run --temp 0 $(test_model ${MODEL} granite-be-3.3:2b) "What is the first line of the declaration of independence?"
 }
 
 @test "ramalama run --keepalive" {
     # timeout within 1 second and generate a 124 error code.
-    run_ramalama 0 --debug run --keepalive 1s tiny
+    run_ramalama 0 --debug run --keepalive 1s $(test_model tiny)
 }
 
 @test "ramalama run --image bogus" {
     skip_if_nocontainer
     skip_if_darwin
     skip_if_docker
+    run_ramalama pull tiny
     run_ramalama 22 run --image bogus --pull=never tiny
     is "$output" ".*Error: bogus: image not known"
     run_ramalama 125 run --image bogus1 --rag quay.io/ramalama/rag --pull=never tiny

test/system/040-serve.bats

@@ -126,7 +126,9 @@ verify_begin=".*run --rm"
     run_ramalama -q --dryrun serve smollm
     is "$output" ".*ai.ramalama.model=ollama://library/smollm:latest" "smollm should be expanded to fullname"
 
-    model=ollama://smollm:135m
+    model=$(test_model smollm:135m)
+    run_ramalama info
+    full_model=$(jq -r --arg model $model -r '.Shortnames.Names[$model]' <<<"$output")
     container1=c_$(safename)
     container2=c_$(safename)
 
@@ -139,7 +141,7 @@ verify_begin=".*run --rm"
     port=${output: -8:4}
 
     run_ramalama chat --ls --url http://127.0.0.1:${port}/v1
-    is "$output" "smollm:135m" "list of models available correct"
+    is "$output" ${full_model#*://} "list of models available correct"
 
     run_ramalama containers --noheading
     is "$output" ".*${container1}" "list correct for container1"
@@ -159,7 +161,7 @@ verify_begin=".*run --rm"
 @test "ramalama --detach serve multiple" {
     skip_if_nocontainer
 
-    model=ollama://smollm:135m
+    model=$(test_model ollama://smollm:135m)
     container=c_$(safename)
     port1=8100
     port2=8200
@@ -445,6 +447,8 @@ verify_begin=".*run --rm"
 @test "ramalama serve --api llama-stack" {
     skip_if_docker
     skip_if_nocontainer
+    skip_if_ppc64le
+    skip_if_s390x
     model=tiny
     name=c_$(safename)
     run_ramalama pull ${model}

test/system/050-pull.bats

@@ -171,6 +171,7 @@ load setup_suite
     if ! is_bigendian; then
         skip "Testing pulls of opposite-endian models"
     fi
+    run_ramalama rm --ignore tiny
     run_ramalama 1 pull --verify=on tiny
     is "$output" ".*Endian mismatch of host (BIG) and model (LITTLE).*" "detected little-endian model"
 }
@@ -179,6 +180,7 @@ load setup_suite
     if is_bigendian; then
         skip "Testing pulls of opposite-endian models"
     fi
+    run_ramalama rm --ignore granite-be-3.0:1b
     run_ramalama 1 pull --verify=on granite-be-3.0:1b
     is "$output" ".*Endian mismatch of host (LITTLE) and model (BIG).*" "detected big-endian model"
 }

test/system/055-convert.bats

@@ -19,6 +19,9 @@ load helpers
 
 @test "ramalama convert file to image" {
     skip_if_nocontainer
+    # Requires the -rag images which are not available on these arches yet
+    skip_if_ppc64le
+    skip_if_s390x
     echo "hello" > $RAMALAMA_TMPDIR/aimodel
     run_ramalama convert file://$RAMALAMA_TMPDIR/aimodel foobar
     run_ramalama list
@@ -45,6 +48,9 @@ load helpers
 @test "ramalama convert tiny to image" {
     skip_if_nocontainer
     skip_if_docker
+    # Requires the -rag images which are not available on these arches yet
+    skip_if_ppc64le
+    skip_if_s390x
     run_ramalama pull tiny
     run_ramalama convert tiny oci://quay.io/ramalama/tiny
     run_ramalama list
@@ -75,6 +81,9 @@ load helpers
 @test "ramalama convert tiny to GGUF image" {
     skip_if_nocontainer
     skip_if_docker
+    # Requires the -rag images which are not available on these arches yet
+    skip_if_ppc64le
+    skip_if_s390x
     run_ramalama pull hf://TinyLlama/TinyLlama-1.1B-Chat-v1.0
     run_ramalama convert --gguf Q4_0 hf://TinyLlama/TinyLlama-1.1B-Chat-v1.0 oci://quay.io/ramalama/tiny-q4-0
     run_ramalama list

test/system/helpers.bash

@@ -289,5 +289,43 @@ function skip_if_no_llama_bench() {
     fi
 }
 
+function is_ppc64le() {
+    [ "$(uname -m)" == "ppc64le" ]
+}
+
+function skip_if_ppc64le() {
+    if is_ppc64le; then
+        skip "Not yet supported on ppc64le"
+    fi
+}
+
+function is_s390x() {
+    [ "$(uname -m)" == "s390x" ]
+}
+
+function skip_if_s390x() {
+    if is_s390x; then
+        skip "Not yet supported on s390x"
+    fi
+}
+
+function is_bigendian() {
+    is_s390x
+}
+
+function test_model() {
+    if is_bigendian; then
+        # If there's a smaller, functional bigendian model, put it here
+        echo ${2:-granite-be-3.0:1b}
+    else
+        echo ${1:-smollm:135m}
+    fi
+}
+
+function model_base() {
+    local base=${1##*/}
+    echo ${base%:*}
+}
+
 # END   miscellaneous tools
 ###############################################################################