Add vicuna prune 0.60 model

vicuna_prune60/2024-05-01-03-26-22/description.txt

@@ -0,0 +1,28 @@
+- Training Parameters: 
+  - base_model: lmsys/vicuna-7b-v1.3
+  - save_ckpt_log_name: vicuna_prune60
+  - pruning_ratio: 0.6
+  - pruner_type: taylor
+  - temperature: 1.0
+  - top_p: 0.95
+  - max_seq_len: 128
+  - channel_wise: False
+  - block_wise: True
+  - layer_wise: False
+  - layer: 12
+  - block_attention_layer_start: 4
+  - block_attention_layer_end: 30
+  - block_mlp_layer_start: 4
+  - block_mlp_layer_end: 30
+  - iterative_steps: 1
+  - grouping_strategy: sum
+  - global_pruning: False
+  - taylor: param_first
+  - num_examples: 10
+  - device: cpu
+  - test_before_train: False
+  - eval_device: cuda
+  - test_after_train: True
+  - seed: 42
+  - save_model: True
+  - torch_version: 2.2

vicuna_prune60/2024-05-01-03-26-22/train.sh

@@ -0,0 +1 @@
+python hf_prune.py --pruning_ratio 0.60 --block_wise --block_mlp_layer_start 4 --block_mlp_layer_end 30 --block_attention_layer_start 4 --block_attention_layer_end 30 --pruner_type taylor --test_after_train --device cpu --eval_device cuda --save_ckpt_log_name vicuna_prune60 --save_model --base_model lmsys/vicuna-7b-v1.3

vicuna_prune60/2024-05-01-03-26-22/training.log

@@ -0,0 +1,57 @@
+2024-05-01 03:26:26 - INFO :       Use taylor pruner...
+2024-05-01 03:26:26 - INFO :       Pruning Attention Layer = [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]
+2024-05-01 03:26:26 - INFO :       Pruning MLP Layer = [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]
+2024-05-01 03:26:32 - INFO :       Start Pruning
+2024-05-01 03:26:43 - INFO :       Start Backwarding in iterative steps = 0...
+2024-05-01 03:26:55 - INFO :       Loss = 3.7534372806549072
+2024-05-01 03:27:46 - INFO :       After Iter 1/1, #parameters: 3592204288
+2024-05-01 03:27:46 - INFO :       #Param before: 6738415616, #Param after: 3592204288, Ratio = 53.3093%
+2024-05-01 03:28:03 - INFO :       
+==================Generation Results After Pruning================
+
+2024-05-01 03:28:14 - INFO :       <s>I believe the meaning of life is the following road, however sometimes it’s difficult to make the journey. We will continue to search and explore the unknown for a little while longer, until we find the right path to follow.”
+During our conversation I recall the memory of the forest, when i walked so long in the pouring rain.
+I continue my journey, the darkness of the unknown road ahead of me creating a blend of light and stars. In the distance, I can see your silhououlouout, I can sense the shake of your hand.
+A light wind blows through my hair
+2024-05-01 03:28:23 - INFO :       <s>Simply put, the theory of relativity states that 50% pounds off a whip, and the next at the bottom who’s gone to school.
+He who does not, at the beginning learn to cut, will reap the blame.
+To the brave who will stand with him.
+An empty mind and a piece in hand, Stril every cheak to day a bitter truth.
+This time was registered by my heart. “
+And then, with tears in his eyes, he turned to me, telling the life-threiling tale.
+The hard life of the poor,
+2024-05-01 03:28:33 - INFO :       <s>Building a website can be done in 10 simple steps:
+You want to ensure the efficiency of these actions and facilitate finding an audience. Therefore, you need to ensure that the website and Google will be looking at each other.
+Before you start, take note of the people who are watching you in order to determine your mood. Whether your colleagues are at their desks, or when they suddenly find inspiration, head down, you will have to make another step towards their coffee, if they get in the way and block their colleagues in front. This will make your face and red lighting your eyes
+2024-05-01 03:28:36 - INFO :       <s>Tweet: "I hate it when my phone battery dies."
+Sentiment: Negative
+###
+Tweet: "My day has been 👍"
+Sentiment: Positive
+###
+Tweet: "This is the link to the article"
+Sentiment: Neutral
+###
+Tweet: "This new music video was incredibile"
+Sentiment:
+AWE. I am amazed at how well I stay when the news isn'ters good. "Noise-free zone"
+This is a major high to have control
+2024-05-01 03:28:42 - INFO :       <s>Translate English to French:
+
+sea otter => loutre de mer
+
+peppermint => menthe poivrée
+
+plush girafe => girafe peluche
+
+cheese => ra Montague qui tala à chent me, i scared) – English – The Laran cels
+Cry – The donkey
+Ferozly – The animal cries
+as the woman is lifted to the ground, and her killer growls.
+The next night, a robber wearing a mask and red roat comes in and kicks
+2024-05-01 03:28:42 - INFO :       
+==================Finish================
+
+2024-05-01 03:32:12 - INFO :       PPL after pruning: {'wikitext2': 154.32526724792663, 'ptb': 507.9935496710831}
+2024-05-01 03:32:12 - INFO :       Memory Requirement: 6995.72802734375 MiB
+

vicuna_prune60/2024-05-01-03-43-15/description.txt

@@ -0,0 +1,28 @@
+- Training Parameters: 
+  - base_model: lmsys/vicuna-7b-v1.3
+  - save_ckpt_log_name: vicuna_prune60
+  - pruning_ratio: 0.6
+  - pruner_type: taylor
+  - temperature: 1.0
+  - top_p: 0.95
+  - max_seq_len: 128
+  - channel_wise: False
+  - block_wise: True
+  - layer_wise: False
+  - layer: 12
+  - block_attention_layer_start: 4
+  - block_attention_layer_end: 30
+  - block_mlp_layer_start: 4
+  - block_mlp_layer_end: 30
+  - iterative_steps: 1
+  - grouping_strategy: sum
+  - global_pruning: False
+  - taylor: param_first
+  - num_examples: 10
+  - device: cpu
+  - test_before_train: False
+  - eval_device: cuda
+  - test_after_train: True
+  - seed: 42
+  - save_model: True
+  - torch_version: 2.2

vicuna_prune60/2024-05-01-03-43-15/train.sh

@@ -0,0 +1 @@
+python hf_prune.py --pruning_ratio 0.60 --block_wise --block_mlp_layer_start 4 --block_mlp_layer_end 30 --block_attention_layer_start 4 --block_attention_layer_end 30 --pruner_type taylor --test_after_train --device cpu --eval_device cuda --save_ckpt_log_name vicuna_prune60 --save_model --base_model lmsys/vicuna-7b-v1.3

vicuna_prune60/2024-05-01-03-43-15/training.log

@@ -0,0 +1,57 @@
+2024-05-01 03:43:24 - INFO :       Use taylor pruner...
+2024-05-01 03:43:24 - INFO :       Pruning Attention Layer = [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]
+2024-05-01 03:43:24 - INFO :       Pruning MLP Layer = [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]
+2024-05-01 03:43:28 - INFO :       Start Pruning
+2024-05-01 03:43:43 - INFO :       Start Backwarding in iterative steps = 0...
+2024-05-01 03:43:54 - INFO :       Loss = 3.7534372806549072
+2024-05-01 03:44:22 - INFO :       After Iter 1/1, #parameters: 3592204288
+2024-05-01 03:44:23 - INFO :       #Param before: 6738415616, #Param after: 3592204288, Ratio = 53.3093%
+2024-05-01 03:44:44 - INFO :       
+==================Generation Results After Pruning================
+
+2024-05-01 03:44:53 - INFO :       <s>I believe the meaning of life is the following road, however sometimes it’s difficult to make the journey. We will continue to search and explore the unknown for a little while longer, until we find the right path to follow.”
+During our conversation I recall the memory of the forest, when i walked so long in the pouring rain.
+I continue my journey, the darkness of the unknown road ahead of me creating a blend of light and stars. In the distance, I can see your silhououlouout, I can sense the shake of your hand.
+A light wind blows through my hair
+2024-05-01 03:45:02 - INFO :       <s>Simply put, the theory of relativity states that 50% pounds off a whip, and the next at the bottom who’s gone to school.
+He who does not, at the beginning learn to cut, will reap the blame.
+To the brave who will stand with him.
+An empty mind and a piece in hand, Stril every cheak to day a bitter truth.
+This time was registered by my heart. “
+And then, with tears in his eyes, he turned to me, telling the life-threiling tale.
+The hard life of the poor,
+2024-05-01 03:45:09 - INFO :       <s>Building a website can be done in 10 simple steps:
+You want to ensure the efficiency of these actions and facilitate finding an audience. Therefore, you need to ensure that the website and Google will be looking at each other.
+Before you start, take note of the people who are watching you in order to determine your mood. Whether your colleagues are at their desks, or when they suddenly find inspiration, head down, you will have to make another step towards their coffee, if they get in the way and block their colleagues in front. This will make your face and red lighting your eyes
+2024-05-01 03:45:10 - INFO :       <s>Tweet: "I hate it when my phone battery dies."
+Sentiment: Negative
+###
+Tweet: "My day has been 👍"
+Sentiment: Positive
+###
+Tweet: "This is the link to the article"
+Sentiment: Neutral
+###
+Tweet: "This new music video was incredibile"
+Sentiment:
+AWE. I am amazed at how well I stay when the news isn'ters good. "Noise-free zone"
+This is a major high to have control
+2024-05-01 03:45:12 - INFO :       <s>Translate English to French:
+
+sea otter => loutre de mer
+
+peppermint => menthe poivrée
+
+plush girafe => girafe peluche
+
+cheese => ra Montague qui tala à chent me, i scared) – English – The Laran cels
+Cry – The donkey
+Ferozly – The animal cries
+as the woman is lifted to the ground, and her killer growls.
+The next night, a robber wearing a mask and red roat comes in and kicks
+2024-05-01 03:45:12 - INFO :       
+==================Finish================
+
+2024-05-01 03:47:39 - INFO :       PPL after pruning: {'wikitext2': 154.32526724792663, 'ptb': 507.9935496710831}
+2024-05-01 03:47:39 - INFO :       Memory Requirement: 6995.72802734375 MiB
+

vicuna_prune60/description.txt

@@ -0,0 +1,28 @@
+- Training Parameters: 
+  - base_model: lmsys/vicuna-7b-v1.3
+  - save_ckpt_log_name: vicuna_prune60
+  - pruning_ratio: 0.6
+  - pruner_type: taylor
+  - temperature: 1.0
+  - top_p: 0.95
+  - max_seq_len: 128
+  - channel_wise: False
+  - block_wise: True
+  - layer_wise: False
+  - layer: 12
+  - block_attention_layer_start: 4
+  - block_attention_layer_end: 30
+  - block_mlp_layer_start: 4
+  - block_mlp_layer_end: 30
+  - iterative_steps: 1
+  - grouping_strategy: sum
+  - global_pruning: False
+  - taylor: param_first
+  - num_examples: 10
+  - device: cpu
+  - test_before_train: False
+  - eval_device: cuda
+  - test_after_train: True
+  - seed: 42
+  - save_model: True
+  - torch_version: 2.2

vicuna_prune60/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18bb341f86cd97bc3de036278bb2083e34729129b5e1386ab36119a2242ac341
+size 7218842434