Delete plm/plm7b(adapter,llm,mlp)_mini

plm/plm7b(adapter,llm,mlp)_mini/adaptor_config.json

@@ -1,7 +0,0 @@
-{
-    "adaptor_name": null,
-    "bottle_dim": 512,
-    "sparse_k": 128,
-    "index_heads": 4,
-    "index_dim": 16
-}

plm/plm7b(adapter,llm,mlp)_mini/logs/log.txt

@@ -1,72 +0,0 @@
-No existing adaptor_config.json found, using command line arguments.
-Using adaptor arguments from command line.
-No modification for Qwen2.5-VL vision block's attention.
-train model visual:  False
-train model visual merger:  True
-train llm:  True
-Unfreezing the panoramic adapter modules.
-
---- Whole Model Parameter Status (ZeRO-3) ---
-Total Parameters:      3754.62 M
-Trainable Parameters:  3122.65 M
-Trainable Percentage:  83.17%
-----------------------------
-
---- Visual Tower Parameter Status (ZeRO-3) ---
-Total Parameters:      668.68 M
-Trainable Parameters:  36.71 M
-Trainable Percentage:  5.49%
-----------------------------
-
---- Vision-Language Merger Parameter Status (ZeRO-3) ---
-Total Parameters:      36.71 M
-Trainable Parameters:  36.71 M
-Trainable Percentage:  100.00%
-----------------------------
-
---- Language Model Parameter Status (ZeRO-3) ---
-Total Parameters:      3085.94 M
-Trainable Parameters:  3085.94 M
-Trainable Percentage:  100.00%
-----------------------------
-
---- Panoramic Adapter Parameter Status (ZeRO-3) ---
-Total Parameters:      0.00 M
-Trainable Parameters:  0.00 M
-Trainable Percentage:  N/A (0 total parameters)
-----------------------------
-
-model trainable parameters:
-Vision Module - Attention Blocks:
-Trainable Block Indices: None
-Non-Trainable Block Indices: [0, 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]
-Merger Module Trainable: True
-LLM Module - Embed Tokens Trainable: True
-LLM Module - Trainable Layer Indices: [0, 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]
-LLM Module - Non-Trainable Layer Indices: None
-Parameter Offload: Total persistent parameters: 755712 in 408 params
-{'loss': 2.1779, 'grad_norm': 12.600039825072235, 'learning_rate': 5.555555555555555e-07, 'epoch': 0.0}
-{'loss': 2.1999, 'grad_norm': 11.75674110267262, 'learning_rate': 1.111111111111111e-06, 'epoch': 0.01}
-{'loss': 2.2641, 'grad_norm': 14.388746619277908, 'learning_rate': 1.6666666666666667e-06, 'epoch': 0.01}
-{'loss': 2.3198, 'grad_norm': 16.134130218330654, 'learning_rate': 2.222222222222222e-06, 'epoch': 0.01}
-{'loss': 2.117, 'grad_norm': 12.833794194691453, 'learning_rate': 2.7777777777777783e-06, 'epoch': 0.02}
-{'loss': 1.9118, 'grad_norm': 10.883819347888405, 'learning_rate': 3.3333333333333333e-06, 'epoch': 0.02}
-{'loss': 2.0464, 'grad_norm': 12.5427286237243, 'learning_rate': 3.88888888888889e-06, 'epoch': 0.03}
-{'loss': 2.0049, 'grad_norm': 13.35092546241232, 'learning_rate': 4.444444444444444e-06, 'epoch': 0.03}
-{'loss': 2.1246, 'grad_norm': 11.730189682620821, 'learning_rate': 5e-06, 'epoch': 0.03}
-{'loss': 1.861, 'grad_norm': 9.609551384949736, 'learning_rate': 4.999820277698202e-06, 'epoch': 0.04}
-{'loss': 2.0375, 'grad_norm': 11.189411268114222, 'learning_rate': 4.999281136632893e-06, 'epoch': 0.04}
-{'loss': 2.156, 'grad_norm': 8.82310667571, 'learning_rate': 4.998382654320609e-06, 'epoch': 0.04}
-{'loss': 1.9998, 'grad_norm': 7.257002244156749, 'learning_rate': 4.997124959943201e-06, 'epoch': 0.05}
-{'loss': 1.7428, 'grad_norm': 6.320183265944286, 'learning_rate': 4.99550823432925e-06, 'epoch': 0.05}
-{'loss': 1.7066, 'grad_norm': 6.986843170051151, 'learning_rate': 4.993532709928075e-06, 'epoch': 0.06}
-{'loss': 1.882, 'grad_norm': 7.922332527495501, 'learning_rate': 4.99119867077631e-06, 'epoch': 0.06}
-{'loss': 1.2039, 'grad_norm': 5.234099566150457, 'learning_rate': 4.9885064524570665e-06, 'epoch': 0.06}
-{'loss': 1.4106, 'grad_norm': 6.863986255992783, 'learning_rate': 4.985456442051683e-06, 'epoch': 0.07}
-{'loss': 1.5842, 'grad_norm': 7.12568239277001, 'learning_rate': 4.982049078084072e-06, 'epoch': 0.07}
-{'loss': 1.5061, 'grad_norm': 6.576928153268851, 'learning_rate': 4.978284850457669e-06, 'epoch': 0.07}
-{'loss': 1.3006, 'grad_norm': 6.4389022399712665, 'learning_rate': 4.974164300384998e-06, 'epoch': 0.08}
-{'loss': 1.4786, 'grad_norm': 6.07725748748085, 'learning_rate': 4.969688020309853e-06, 'epoch': 0.08}
-{'loss': 1.2155, 'grad_norm': 7.060642353490742, 'learning_rate': 4.9648566538221224e-06, 'epoch': 0.08}
-{'loss': 1.2863, 'grad_norm': 5.571732169440192, 'learning_rate': 4.959670895565248e-06, 'epoch': 0.09}
-{'loss': 1.2208, 'grad_norm': 6.252915176043189, 'learning_rate': 4.954131491136362e-06, 'epoch': 0.09}

plm/plm7b(adapter,llm,mlp)_mini/logs/log.txt-2026-03-15-00-58-56

@@ -1,47 +0,0 @@
-No existing adaptor_config.json found, using command line arguments.
-Using adaptor arguments from command line.
-No modification for Qwen2.5-VL vision block's attention.
-train model visual:  False
-train model visual merger:  True
-train llm:  True
-Unfreezing the panoramic adapter modules.
-
---- Whole Model Parameter Status (ZeRO-3) ---
-Total Parameters:      8292.17 M
-Trainable Parameters:  7660.19 M
-Trainable Percentage:  92.38%
-----------------------------
-
---- Visual Tower Parameter Status (ZeRO-3) ---
-Total Parameters:      676.55 M
-Trainable Parameters:  44.57 M
-Trainable Percentage:  6.59%
-----------------------------
-
---- Vision-Language Merger Parameter Status (ZeRO-3) ---
-Total Parameters:      44.57 M
-Trainable Parameters:  44.57 M
-Trainable Percentage:  100.00%
-----------------------------
-
---- Language Model Parameter Status (ZeRO-3) ---
-Total Parameters:      7070.62 M
-Trainable Parameters:  7070.62 M
-Trainable Percentage:  100.00%
-----------------------------
-
---- Panoramic Adapter Parameter Status (ZeRO-3) ---
-Total Parameters:      0.00 M
-Trainable Parameters:  0.00 M
-Trainable Percentage:  N/A (0 total parameters)
-----------------------------
-
-model trainable parameters:
-Vision Module - Attention Blocks:
-Trainable Block Indices: None
-Non-Trainable Block Indices: [0, 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]
-Merger Module Trainable: True
-LLM Module - Embed Tokens Trainable: True
-LLM Module - Trainable Layer Indices: [0, 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]
-LLM Module - Non-Trainable Layer Indices: None
-Parameter Offload: Total persistent parameters: 848896 in 368 params