blob: 25347b58968636b4116ce05a19f2a56f76603d5a (
plain) (
blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
|
#!/usr/bin/env bash
set -euo pipefail
ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/.." && pwd -P)"
DEFAULT_MODEL_GLOB='models--perplexity-ai--pplx-embed-v1-0.6b/snapshots/*'
DEFAULT_CONTEXT_MODEL_GLOB='models--perplexity-ai--pplx-embed-context-v1-0.6B/snapshots/*'
model_dir=""
context_model_dir=""
binary="./ffwd-cli"
batch_size=4
skip_build=0
skip_mlx=0
platform_skip_mlx=0
skip_contextual=0
context_runs=0
late_model_dir=""
long=0
skip_code_quality=0
skip_hermetic=0
skip_tokenizer=0
skip_mlx_quantized=0
die() {
echo "error: $*" >&2
exit 2
}
usage() {
cat <<EOF
Usage: $(basename "$0") [options]
Options:
--model-dir DIR 0.6B model snapshot; defaults to the HF cache
--context-model-dir DIR context 0.6B snapshot; defaults to the HF cache
--binary PATH binary to test, default: ./ffwd-cli
--batch-size N batch size, default: 4
--skip-build skip make cpu / make gpu
--skip-hermetic skip the model-free make test suite
--skip-mlx stop after the CPU build and CPU checks
--skip-contextual skip contextual parity checks
--skip-tokenizer skip Hugging Face tokenizer parity
--skip-mlx-quantized skip the MLX Q8 parity check
--context-runs N timed contextual batches per backend after parity
--late-model-dir DIR optional late-interaction 0.6B snapshot to verify
--long also run the long CPU attention parity check
--skip-code-quality skip the final compiler/analyzer/tidy CI gate
-h, --help show this help
EOF
}
find_cached_snapshot() {
local glob="$1"
local label="$2"
local hub="$HOME/.cache/huggingface/hub"
local last=""
local p
while IFS= read -r p; do
if [[ -d "$p" && -f "$p/config.json" ]]; then
last="$p"
fi
done < <({ compgen -G "$hub/$glob" || true; } | sort)
if [[ -z "$last" ]]; then
echo "no cached $label snapshot found; pass the model dir" >&2
exit 1
fi
printf '%s\n' "$last"
}
find_default_model() {
find_cached_snapshot "$DEFAULT_MODEL_GLOB" "pplx-embed-v1-0.6b"
}
find_default_context_model() {
find_cached_snapshot "$DEFAULT_CONTEXT_MODEL_GLOB" "pplx-embed-context-v1-0.6b"
}
run() {
printf '+'
printf ' %q' "$@"
printf '\n'
(cd "$ROOT" && "$@")
}
run_parity() {
local binary_arg="$1"
local model_dir_arg="$2"
local backend="$3"
local batch_size_arg="$4"
local use_long="$5"
local cmd=(
"$ROOT/all.bash" check-batch-parity
--binary "$binary_arg"
--model-dir "$model_dir_arg"
--backend "$backend"
--batch-size "$batch_size_arg"
)
if (( use_long )); then
cmd+=(--long)
fi
run "${cmd[@]}"
}
run_contextual() {
local model_dir_arg="$1"
local backend="$2"
local runs="$3"
run \
"$ROOT/all.bash" check-context-batch-parity \
--model-dir "$model_dir_arg" \
--backend "$backend" \
--runs "$runs"
}
run_late() {
local model_dir_arg="$1"
local backend="$2"
run \
"$ROOT/all.bash" check-late-interaction \
--model-dir "$model_dir_arg" \
--backend "$backend"
}
while (($#)); do
case "$1" in
--model-dir)
(($# >= 2)) || die "--model-dir requires a value"
model_dir="$2"
shift 2
;;
--model-dir=*)
model_dir="${1#*=}"
shift
;;
--context-model-dir)
(($# >= 2)) || die "--context-model-dir requires a value"
context_model_dir="$2"
shift 2
;;
--context-model-dir=*)
context_model_dir="${1#*=}"
shift
;;
--binary)
(($# >= 2)) || die "--binary requires a value"
binary="$2"
shift 2
;;
--binary=*)
binary="${1#*=}"
shift
;;
--batch-size)
(($# >= 2)) || die "--batch-size requires a value"
batch_size="$2"
shift 2
;;
--batch-size=*)
batch_size="${1#*=}"
shift
;;
--skip-build)
skip_build=1
shift
;;
--skip-hermetic)
skip_hermetic=1
shift
;;
--skip-mlx)
skip_mlx=1
shift
;;
--skip-contextual)
skip_contextual=1
shift
;;
--skip-tokenizer)
skip_tokenizer=1
shift
;;
--skip-mlx-quantized)
skip_mlx_quantized=1
shift
;;
--context-runs)
(($# >= 2)) || die "--context-runs requires a value"
context_runs="$2"
shift 2
;;
--context-runs=*)
context_runs="${1#*=}"
shift
;;
--late-model-dir)
(($# >= 2)) || die "--late-model-dir requires a value"
late_model_dir="$2"
shift 2
;;
--late-model-dir=*)
late_model_dir="${1#*=}"
shift
;;
--long)
long=1
shift
;;
--skip-code-quality)
skip_code_quality=1
shift
;;
-h|--help)
usage
exit 0
;;
*)
die "unknown argument: $1"
;;
esac
done
if [[ "$(uname -s)" != "Darwin" && $skip_mlx -eq 0 ]]; then
echo "info: MLX checks require macOS; skipping on $(uname -s)"
skip_mlx=1
platform_skip_mlx=1
fi
if [[ -z "$model_dir" ]]; then
model_dir="$(find_default_model)"
fi
if (( skip_contextual )); then
context_model_dir=""
elif [[ -z "$context_model_dir" ]]; then
context_model_dir="$(find_default_context_model)"
fi
if (( ! skip_hermetic )); then
run make test
fi
if (( ! skip_build )); then
run make cpu
fi
run "$ROOT/all.bash" check-kernel-golden --blas
if (( ! skip_tokenizer )); then
run "$ROOT/all.bash" check-tokenizer-parity --model-dir "$model_dir"
fi
run "$ROOT/all.bash" check-workspace-api --model-dir "$model_dir"
run_parity "$binary" "$model_dir" cpu "$batch_size" 0
if (( long )); then
run_parity "$binary" "$model_dir" cpu "$batch_size" 1
fi
if [[ -n "$context_model_dir" ]]; then
run_contextual "$context_model_dir" cpu "$context_runs"
fi
if [[ -n "$late_model_dir" ]]; then
run_late "$late_model_dir" cpu
fi
if (( skip_mlx )); then
if (( platform_skip_mlx && ! skip_code_quality )); then
run "$ROOT/all.bash" check-code-quality
fi
if (( skip_build )); then
echo "ok: CPU checks passed; builds and MLX checks skipped"
else
echo "ok: CPU build verified; MLX build skipped"
fi
exit 0
fi
if (( ! skip_build )); then
run make gpu
fi
run_parity "$binary" "$model_dir" mlx "$batch_size" 0
run \
"$ROOT/all.bash" check-mlx-model-limits \
--binary "$binary" \
--model-dir "$model_dir"
if (( ! skip_mlx_quantized )); then
run \
"$ROOT/all.bash" check-mlx-quantized-parity \
--binary "$binary" \
--model-dir "$model_dir" \
--batch-size "$batch_size"
fi
if [[ -n "$context_model_dir" ]]; then
run_contextual "$context_model_dir" mlx "$context_runs"
fi
if [[ -n "$late_model_dir" ]]; then
run_late "$late_model_dir" mlx
fi
if (( ! skip_code_quality )); then
run "$ROOT/all.bash" check-code-quality
fi
echo "verify ok"
|