Skip to content

[QEff.Finetuning]CI enablement for Fine-Tuning #629

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 13 commits into
base: main
Choose a base branch
from
+325 −235
Open

[QEff.Finetuning]CI enablement for Fine-Tuning #629

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
4f4982a
Reference values are updated
quic-akuruvil Nov 21, 2025
951ded4
jenkins env changed
quic-akuruvil Nov 21, 2025
359e98b
Separated the finetune test for functional and assertion cases
tchawada Nov 24, 2025
d01717d
Separated the finetune test for functional and assertion cases
tchawada Nov 24, 2025
844f223
Skip assert test
quic-akuruvil Nov 24, 2025
4f383d2
Preflight env activate for FT with extra torch whl
quic-akuruvil Nov 24, 2025
a8956eb
Comment torch-qaic-env
quic-akuruvil Nov 24, 2025
ef30c44
jenkins script update
quic-akuruvil Nov 24, 2025
dd55003
jenkins script update
quic-akuruvil Nov 25, 2025
48f06b0
jenkins script update
quic-akuruvil Nov 27, 2025
ad8897d
jenkins script update
quic-akuruvil Nov 27, 2025
0a81fa7
Update dataset path
quic-akuruvil Nov 28, 2025
0009cac
Update atol
quic-akuruvil Nov 28, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion scripts/Jenkinsfile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,6 @@ pipeline {
pip install junitparser pytest-xdist &&
pip install librosa==0.10.2 soundfile==0.13.1 && #packages needed to load example for whisper testing
pip install --extra-index-url https://download.pytorch.org/whl/cpu timm==1.0.14 torchvision==0.22.0+cpu einops==0.8.1 && #packages to load VLMs
pip install /opt/qti-aic/integrations/torch_qaic/py310/torch_qaic-0.1.0-cp310-cp310-linux_x86_64.whl && # For finetuning tests
rm -rf QEfficient"
'''
}
Expand Down Expand Up @@ -168,6 +167,7 @@ pipeline {
sudo docker exec ${BUILD_TAG} bash -c "
cd /efficient-transformers &&
. preflight_qeff/bin/activate &&
pip install /opt/qti-aic/integrations/torch_qaic/py310/torch_qaic-0.1.0-cp310-cp310-linux_x86_64.whl &&
mkdir -p $PWD/cli_qaic_finetuning &&
export TOKENIZERS_PARALLELISM=false &&
export QEFF_HOME=$PWD/cli_qaic_finetuning &&
Expand Down
334 changes: 167 additions & 167 deletions tests/finetune/reference_data.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,206 +13,206 @@
"llama_3.2_1B_config_alpaca_single_device": {
"description": "Baseline for Llama on Alpaca single-device",
"train_step_losses": [
1.5112206935882568,
1.2211230993270874,
1.9942185878753662,
2.093623161315918,
0.9168124198913574,
1.2125635147094727,
0.3648962676525116,
1.6231939792633057,
0.8259601593017578,
0.7741442918777466,
1.7359141111373901,
2.118462085723877,
2.061161994934082,
0.8256913423538208,
0.8088029623031616,
1.761340618133545,
1.6828027963638306,
1.3538823127746582,
2.0672550201416016,
3.1532647609710693,
1.5110896825790405,
1.2206485271453857,
1.9950776100158691,
2.091615676879883,
0.9182446599006653,
1.1993569135665894,
0.36413607001304626,
1.6241482496261597,
0.8270177245140076,
0.7749958634376526,
1.73696768283844,
2.120077610015869,
2.061460256576538,
0.8267984390258789,
0.8105809688568115,
1.7627557516098022,
1.6819559335708618,
1.3528242111206055,
2.0654125213623047,
3.156151294708252,
],
"eval_step_losses": [
1.462059736251831,
0.24527676403522491,
1.046107292175293,
1.6403586864471436,
1.395291805267334,
2.8664817810058594,
1.035412311553955,
1.8670039176940918,
3.8079662322998047,
0.6516809463500977,
1.4607517719268799,
0.24302150309085846,
1.0471211671829224,
1.642044186592102,
1.3949533700942993,
2.8850066661834717,
1.0366586446762085,
1.8661959171295166,
3.81632924079895,
0.6577113270759583,
],
"train_step_metrics": [
4.532259941101074,
3.390994071960449,
7.34645938873291,
8.114261627197266,
2.5013046264648438,
3.3620924949645996,
1.4403645992279053,
5.069255828857422,
2.2840728759765625,
2.1687355041503906,
5.674112319946289,
8.318334579467773,
7.855090141296387,
2.283458948135376,
2.2452187538146973,
5.820234775543213,
5.380615711212158,
3.872429847717285,
7.903097629547119,
23.412376403808594,
4.531666278839111,
3.389385223388672,
7.352773189544678,
8.09798812866211,
2.504889488220215,
3.3179824352264404,
1.43927001953125,
5.074095249176025,
2.286489486694336,
2.1705832481384277,
5.680093288421631,
8.33178424835205,
7.857433319091797,
2.2859883308410645,
2.2492144107818604,
5.828476905822754,
5.376060962677002,
3.8683345317840576,
7.8885498046875,
23.480052947998047,
],
"eval_step_metrics": [ # steps 0-9
4.31483793258667,
1.2779749631881714,
2.8465487957000732,
5.157018661499023,
4.036152362823486,
17.575077056884766,
2.816267251968384,
6.468885898590088,
45.05870819091797,
1.9187631607055664,
4.309197902679443,
1.27509605884552,
2.8494362831115723,
5.1657185554504395,
4.034786224365234,
17.9036865234375,
2.819779396057129,
6.463661193847656,
45.437110900878906,
1.9303690195083618,
],
},
# Scenario 2: Single-device llama 3.2-1B training on GSM8k dataset.
"llama_3.2_1B_config_gsm8k_single_device": {
"description": "Baseline for Llama on GSM8k single-device",
"train_step_losses": [
2.250276803970337,
2.3231687545776367,
1.9379945993423462,
1.5981022119522095,
1.9867562055587769,
1.4573354721069336,
1.8969658613204956,
1.2177824974060059,
1.6489791870117188,
1.5380687713623047,
1.4025083780288696,
1.5301083326339722,
1.6858205795288086,
1.383747935295105,
1.7968919277191162,
1.4075607061386108,
1.6447738409042358,
1.2807793617248535,
0.8450672030448914,
1.5795941352844238,
2.250361204147339,
2.3252110481262207,
1.9360781908035278,
1.5984115600585938,
1.9874038696289062,
1.4579044580459595,
1.8975679874420166,
1.2175723314285278,
1.6473736763000488,
1.537960410118103,
1.4019465446472168,
1.5310447216033936,
1.6878201961517334,
1.3849903345108032,
1.7976438999176025,
1.4060133695602417,
1.646375060081482,
1.2835280895233154,
0.8465587496757507,
1.5783095359802246,
],
"eval_step_losses": [
1.7081595659255981,
1.719305157661438,
1.153528094291687,
2.0051634311676025,
1.3372926712036133,
1.3009852170944214,
1.2207027673721313,
1.3452664613723755,
1.329830288887024,
1.307450532913208,
1.707140326499939,
1.7226355075836182,
1.1531383991241455,
2.0035903453826904,
1.3362350463867188,
1.3013248443603516,
1.2195535898208618,
1.3454742431640625,
1.3299248218536377,
1.3073854446411133,
],
"train_step_metrics": [
9.490362167358398,
10.207969665527344,
6.944809913635254,
4.943641662597656,
7.291841506958008,
4.294501304626465,
6.6656389236450195,
3.3796849250793457,
5.201667308807373,
4.655590534210205,
4.065384864807129,
4.618677139282227,
5.396877765655518,
3.989826202392578,
6.030873775482178,
4.0859761238098145,
5.179838180541992,
3.5994436740875244,
2.328134298324585,
4.852985858917236,
9.49116325378418,
10.228837966918945,
6.93151330947876,
4.945170879364014,
7.296566009521484,
4.296945571899414,
6.66965389251709,
3.378974676132202,
5.193322658538818,
4.655086040496826,
4.063101291656494,
4.623003959655762,
5.407680034637451,
3.994786262512207,
6.0354108810424805,
4.0796589851379395,
5.188138961791992,
3.60935115814209,
2.3316092491149902,
4.846755504608154,
],
"eval_step_metrics": [ # steps 0-9
5.518795013427734,
5.580649375915527,
3.1693549156188965,
7.42730712890625,
3.8087174892425537,
3.672913074493408,
3.38956880569458,
3.8392088413238525,
3.7804012298583984,
3.6967368125915527,
5.5131731033325195,
5.599266052246094,
3.1681201457977295,
7.415632247924805,
3.8046915531158447,
3.674160957336426,
3.3856759071350098,
3.8400065898895264,
3.7807586193084717,
3.69649600982666,
],
},
# Scenario 3: Single-device google-bert/bert-base-uncased training on IMDB dataset.
"bert_base_uncased_config_imdb_single_device": {
"description": "Baseline for google-bert/bert-base-uncased on IMDB single-device",
"train_step_losses": [
0.357421875,
0.546875,
0.98486328125,
0.35302734375,
1.23828125,
0.60791015625,
0.44384765625,
0.791015625,
0.7861328125,
0.51318359375,
0.50244140625,
0.90087890625,
0.8818359375,
0.86279296875,
0.6396484375,
0.49267578125,
0.97119140625,
0.7451171875,
0.798828125,
0.7080078125,
0.390625,
0.51220703125,
0.9208984375,
0.4052734375,
1.1640625,
0.6533203125,
0.5087890625,
0.76171875,
0.63525390625,
0.50146484375,
0.5439453125,
0.947265625,
0.89013671875,
0.80419921875,
0.6533203125,
0.4580078125,
0.92041015625,
0.7412109375,
0.7197265625,
0.62158203125,
],
"eval_step_losses": [
0.634765625,
0.8173828125,
0.6044921875,
0.798828125,
0.9072265625,
0.7177734375,
0.59423828125,
0.69921875,
0.7109375,
0.7216796875,
0.6064453125,
0.7041015625,
0.70361328125,
0.59912109375,
0.66357421875,
0.6962890625,
0.75390625,
0.61328125,
0.6806640625,
],
"train_step_metrics": [
1.0,
1.0,
0.5,
0.5,
0.5,
0.5,
0.5,
0.5,
0.5,
0.5,
0.5,
0.5,
0.5,
0.5,
0.5,
0.5,
0.5,
0.5,
0.449951171875,
0.4091796875,
0.49999988079071045,
0.49999988079071045,
0.5,
0.5000002384185791,
0.5000002384185791,
0.6250002384185791,
0.6249998807907104,
0.625,
0.6000000238418579,
0.5833332538604736,
0.5714285373687744,
0.5714285373687744,
0.5714285373687744,
0.5625,
0.555555522441864,
0.5055557489395142,
0.5101010203361511,
],
"eval_step_metrics": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0],
"eval_step_metrics": [1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0],
},
# Scenario 4: Distributed google-bert/bert-base-uncased training (world_size=2)
"bert_base_uncased_config_imdb_distributed_ws2": {
Expand Down
Loading
Loading