Investigate accuracy issues in working models

[ayerofieiev-tt]

Among the 66 models running end-to-end in TT-NN, the following models exhibit accuracy deviations:

Models with Accuracy Issues:

1. Bloom: 43.3
2. albert/albert-base-v2: 69.04
3. albert/albert-large-v2: 23.63
4. albert/albert-xlarge-v2: 53.42
5. mobilenetv1_100.ra4_e3600_r224_in1k: -6.46
6. regnet_x_16gf: 1.01
7. regnet_x_32gf: 3.94

Model with Accuracy Marked as N/A:

1. Unet-brain: N/A

Action Required:

1. Investigate and diagnose the causes of the accuracy deviations.
2. If the issue is in TT-NN operation, determine what operation on what input causes an issue. Fire a ticket to tt-metal, link here
3. Validate expected accuracy baselines for each model.
4. Assess Unet-brain to determine if accuracy reporting is feasible.