[PT2] weight_compression

[AlexanderDokuchaev]

Changes

Implement weight compression algorithms for experimental PT tracing

Related tickets

152996

[AlexanderDokuchaev]

Model	Metric value	Metric diff	Num int8	Num int4	Compr. time	Stat. collection time	Mixed-Precision search time	Apply Compression time	Total time	RAM MiB	RAM MiB System	Status
tinyllama_data_aware	0.85767	-0.14233	123	94	0:01:00	0:00:51	0:00:04	0:00:02	0:03:09	5099	1618	XFAIL: num_compressed_xfail_reason - Regression: The number of int8 ops is different than reference 124 != 123
tinyllama_int8_data_free	0.95624	-0.04376	311	0	0:00:03			0:00:02	0:02:25	5336	2499	XFAIL: num_compressed_xfail_reason - Regression: The number of int8 ops is different than reference 312 != 311
tinyllama_int4_data_free	0.73873	-0.26127	83	114	0:00:08		0:00:04	0:00:02	0:02:16	5201	1990	XFAIL: num_compressed_xfail_reason - Regression: The number of int8 ops is different than reference 84 != 83
tinyllama_scale_estimation_per_channel	0.80799	-0.19201	123	188	0:02:31	0:00:55	0:00:04	0:00:01	0:04:40	5280	4012	XFAIL: num_compressed_xfail_reason - Regression: The number of int8 ops is different than reference 124 != 123
tinyllama_scale_estimation_group_size_64	0.8566	-0.1434	123	94	0:03:21	0:00:57	0:00:04	0:00:02	0:05:29	5461	4287	XFAIL: num_compressed_xfail_reason - Regression: The number of int8 ops is different than reference 124 != 123