noonghunnaandClaude Opus 4.8
180adb083e
beellama.cpp composes (ghcr.io/{anbeeld/beellama.cpp,noonghunna/beellama-cpp})
were not matched by the is_llamacpp image regex, so they fell into the vLLM
HF-cache branch and skipped the model-presence check entirely. And even on the
llama.cpp branch, only `-m`/`--model` (the target) was checked — never the
`--spec-draft-model` drafter. Result: a missing target OR drafter GGUF passed
preflight and surfaced as a cryptic in-container "failed to open GGUF file"
crash instead of preflight's "download this:" hint (#288 eddie + George).
- Add `beellama` to the is_llamacpp image regex.
- Collect drafter paths from --spec-draft-model/--model-draft/-md into their own
presence check, with a DRAFT_FILE env override mirroring GGUF_FILE/MMPROJ_FILE.
- New test case: beellama target-present + drafter-missing must refuse with the
drafter path + its `hf download Anbeeld/...` hint; both present must pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
197 lines
7.1 KiB
Bash
Executable File
197 lines
7.1 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
set -euo pipefail
|
|
|
|
ROOT_DIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd)"
|
|
TMP_DIR="$(mktemp -d)"
|
|
trap 'rm -rf "$TMP_DIR"' EXIT
|
|
|
|
assert_contains() {
|
|
local haystack="$1"
|
|
local needle="$2"
|
|
if [[ "$haystack" != *"$needle"* ]]; then
|
|
echo "ASSERTION FAILED: expected output to contain: $needle" >&2
|
|
echo "--- output ---" >&2
|
|
echo "$haystack" >&2
|
|
exit 1
|
|
fi
|
|
}
|
|
|
|
assert_not_contains() {
|
|
local haystack="$1"
|
|
local needle="$2"
|
|
if [[ "$haystack" == *"$needle"* ]]; then
|
|
echo "ASSERTION FAILED: expected output not to contain: $needle" >&2
|
|
echo "--- output ---" >&2
|
|
echo "$haystack" >&2
|
|
exit 1
|
|
fi
|
|
}
|
|
|
|
run_deps() {
|
|
local compose="$1"
|
|
local model_dir="$2"
|
|
(
|
|
export ROOT_DIR MODEL_DIR="$model_dir"
|
|
source "${ROOT_DIR}/scripts/preflight.sh"
|
|
preflight_compose_deps "$compose"
|
|
) 2>&1
|
|
}
|
|
|
|
run_deps_unset_model_dir() {
|
|
local compose="$1"
|
|
(
|
|
unset MODEL_DIR
|
|
export ROOT_DIR
|
|
source "${ROOT_DIR}/scripts/preflight.sh"
|
|
preflight_compose_deps "$compose"
|
|
) 2>&1
|
|
}
|
|
|
|
expect_missing() {
|
|
local compose="$1"
|
|
local model_dir="$2"
|
|
local output
|
|
|
|
if output="$(run_deps "$compose" "$model_dir")"; then
|
|
echo "ASSERTION FAILED: expected missing-model failure for $compose" >&2
|
|
echo "--- output ---" >&2
|
|
echo "$output" >&2
|
|
exit 1
|
|
fi
|
|
printf '%s' "$output"
|
|
}
|
|
|
|
ik_compose="${TMP_DIR}/ik.yml"
|
|
cat > "$ik_compose" <<'YAML'
|
|
services:
|
|
ik:
|
|
image: ${IK_LLAMA_IMAGE:-ghcr.io/ikawrakow/ik-llama-cpp:cu13-server}
|
|
command: >-
|
|
--model /models/${GGUF_FILE:-qwen3.6-27b-gguf/ubergarm-mtp-iq4ks/Qwen3.6-27B-MTP-IQ4_KS.gguf}
|
|
YAML
|
|
|
|
out="$(expect_missing "$ik_compose" "${TMP_DIR}/empty-models")"
|
|
assert_contains "$out" "qwen3.6-27b-gguf/ubergarm-mtp-iq4ks/Qwen3.6-27B-MTP-IQ4_KS.gguf"
|
|
assert_contains "$out" "hf download ubergarm/Qwen3.6-27B-GGUF Qwen3.6-27B-MTP-IQ4_KS.gguf"
|
|
assert_contains "$out" "WEIGHTS=iq4ks bash scripts/setup.sh qwen3.6-27b"
|
|
assert_not_contains "$out" "qwen3.6-27b-autoround-int4"
|
|
|
|
llama_compose="${TMP_DIR}/llama.yml"
|
|
cat > "$llama_compose" <<'YAML'
|
|
services:
|
|
llama:
|
|
image: ${LLAMACPP_IMAGE:-ghcr.io/ggml-org/llama.cpp:server-cuda-b9246}
|
|
command: >-
|
|
-m /models/${GGUF_FILE:-qwen3.6-27b-gguf/unsloth-mtp-q4km/Qwen3.6-27B-Q4_K_M.gguf}
|
|
YAML
|
|
mkdir -p "${TMP_DIR}/models/qwen3.6-27b-gguf/unsloth-mtp-q4km"
|
|
touch "${TMP_DIR}/models/qwen3.6-27b-gguf/unsloth-mtp-q4km/Qwen3.6-27B-Q4_K_M.gguf"
|
|
out="$(run_deps "$llama_compose" "${TMP_DIR}/models")"
|
|
[[ -z "$out" ]]
|
|
|
|
vllm_compose="${TMP_DIR}/gemma.yml"
|
|
cat > "$vllm_compose" <<'YAML'
|
|
services:
|
|
vllm:
|
|
image: vllm/vllm-openai:v0.21.0
|
|
command:
|
|
- --model
|
|
- /root/.cache/huggingface/gemma-4-31b-autoround-int4
|
|
- --speculative-config
|
|
- '{"model":"/root/.cache/huggingface/gemma-4-31b-it-assistant","num_speculative_tokens":4}'
|
|
YAML
|
|
out="$(expect_missing "$vllm_compose" "${TMP_DIR}/empty-models")"
|
|
assert_contains "$out" "gemma-4-31b-autoround-int4/config.json"
|
|
assert_contains "$out" "gemma-4-31b-it-assistant/config.json"
|
|
assert_contains "$out" "hf download Intel/gemma-4-31B-it-int4-AutoRound"
|
|
assert_contains "$out" "hf download google/gemma-4-31B-it-assistant"
|
|
assert_contains "$out" "bash scripts/setup.sh gemma-4-31b"
|
|
assert_not_contains "$out" "qwen3.6-27b-autoround-int4"
|
|
mkdir -p "${TMP_DIR}/models/gemma-4-31b-autoround-int4" "${TMP_DIR}/models/gemma-4-31b-it-assistant"
|
|
touch "${TMP_DIR}/models/gemma-4-31b-autoround-int4/config.json" "${TMP_DIR}/models/gemma-4-31b-it-assistant/config.json"
|
|
out="$(run_deps "$vllm_compose" "${TMP_DIR}/models")"
|
|
[[ -z "$out" ]]
|
|
|
|
|
|
extends_base="${TMP_DIR}/base.yml"
|
|
extends_stub="${TMP_DIR}/stub.yml"
|
|
cat > "$extends_base" <<'YAML'
|
|
services:
|
|
vllm-base:
|
|
image: vllm/vllm-openai:v0.21.0
|
|
command:
|
|
- --model
|
|
- /root/.cache/huggingface/qwen3.6-27b-autoround-int4
|
|
YAML
|
|
cat > "$extends_stub" <<'YAML'
|
|
services:
|
|
vllm-stub:
|
|
extends:
|
|
file: base.yml
|
|
service: vllm-base
|
|
YAML
|
|
out="$(expect_missing "$extends_stub" "${TMP_DIR}/empty-models")"
|
|
assert_contains "$out" "qwen3.6-27b-autoround-int4/config.json"
|
|
|
|
if out="$(run_deps_unset_model_dir "$extends_stub")"; then
|
|
echo "ASSERTION FAILED: expected missing-model failure with unset MODEL_DIR" >&2
|
|
echo "--- output ---" >&2
|
|
echo "$out" >&2
|
|
exit 1
|
|
fi
|
|
assert_contains "$out" "MODEL_DIR not set"
|
|
assert_contains "$out" "defaulting to ${ROOT_DIR}/models-cache"
|
|
|
|
out="$(CLUB3090_WEIGHTS_READER_DISABLE=1 expect_missing "$extends_stub" "${TMP_DIR}/empty-models")"
|
|
assert_contains "$out" "qwen3.6-27b-autoround-int4/config.json"
|
|
assert_contains "$out" "Check the compose header for its model-specific hf download command."
|
|
assert_not_contains "$out" "hf download Lorbus/Qwen3.6-27B-int4-AutoRound"
|
|
|
|
sglang_compose="${TMP_DIR}/sglang.yml"
|
|
cat > "$sglang_compose" <<'YAML'
|
|
services:
|
|
sglang:
|
|
image: ghcr.io/sgl-project/sglang:v0.5.12
|
|
volumes:
|
|
- "${MODEL_DIR:?MODEL_DIR is required}/qwen3.6-27b-autoround-int4:/models/target:ro"
|
|
- "${MODEL_DIR}/qwen3.6-27b-prism-eagle3/compressed:/models/drafter:ro"
|
|
YAML
|
|
out="$(expect_missing "$sglang_compose" "${TMP_DIR}/empty-models")"
|
|
assert_contains "$out" "qwen3.6-27b-autoround-int4"
|
|
assert_contains "$out" "qwen3.6-27b-prism-eagle3/compressed"
|
|
assert_contains "$out" "hf download Ex0bit/Qwen3.6-27B-PRISM-EAGLE3"
|
|
assert_contains "$out" "WITH_PRISM_EAGLE3=1 bash scripts/setup.sh qwen3.6-27b"
|
|
mkdir -p "${TMP_DIR}/models/qwen3.6-27b-autoround-int4" "${TMP_DIR}/models/qwen3.6-27b-prism-eagle3/compressed"
|
|
out="$(run_deps "$sglang_compose" "${TMP_DIR}/models")"
|
|
[[ -z "$out" ]]
|
|
|
|
# beellama.cpp DFlash: the image is NOT ggml-org/ik-llama (so it must still be
|
|
# detected as a llama.cpp-family GGUF server), and the drafter is named via
|
|
# --spec-draft-model (not -m). A present target + MISSING drafter must refuse
|
|
# with the drafter path + its hf-download hint — the #288 George report, where
|
|
# a missing drafter otherwise crashed cryptically in-container.
|
|
beellama_compose="${TMP_DIR}/beellama.yml"
|
|
cat > "$beellama_compose" <<'YAML'
|
|
services:
|
|
beellama:
|
|
image: ${BEELLAMA_IMAGE:-ghcr.io/anbeeld/beellama.cpp:server-cuda-v0.3.0-e0663be2713c}
|
|
command: >-
|
|
-m /models/${GGUF_FILE:-qwen3.6-27b-gguf/unsloth-q5ks/Qwen3.6-27B-Q5_K_S.gguf}
|
|
--spec-draft-model /models/${DRAFT_FILE:-qwen3.6-27b-gguf/anbeeld-dflash-iq4xs/Qwen3.6-27B-DFlash-IQ4_XS.gguf}
|
|
YAML
|
|
mkdir -p "${TMP_DIR}/bl-models/qwen3.6-27b-gguf/unsloth-q5ks"
|
|
touch "${TMP_DIR}/bl-models/qwen3.6-27b-gguf/unsloth-q5ks/Qwen3.6-27B-Q5_K_S.gguf"
|
|
out="$(expect_missing "$beellama_compose" "${TMP_DIR}/bl-models")"
|
|
assert_contains "$out" "qwen3.6-27b-gguf/anbeeld-dflash-iq4xs/Qwen3.6-27B-DFlash-IQ4_XS.gguf"
|
|
assert_contains "$out" "speculative drafter GGUF"
|
|
assert_contains "$out" "hf download Anbeeld/Qwen3.6-27B-DFlash-GGUF Qwen3.6-27B-DFlash-IQ4_XS.gguf"
|
|
# the target is present, so it must NOT be reported missing
|
|
assert_not_contains "$out" "unsloth-q5ks/Qwen3.6-27B-Q5_K_S.gguf (llama.cpp GGUF weights)"
|
|
# add the drafter → must pass
|
|
mkdir -p "${TMP_DIR}/bl-models/qwen3.6-27b-gguf/anbeeld-dflash-iq4xs"
|
|
touch "${TMP_DIR}/bl-models/qwen3.6-27b-gguf/anbeeld-dflash-iq4xs/Qwen3.6-27B-DFlash-IQ4_XS.gguf"
|
|
out="$(run_deps "$beellama_compose" "${TMP_DIR}/bl-models")"
|
|
[[ -z "$out" ]]
|
|
|
|
echo "test-preflight-compose-deps: ok"
|