Upload 9 files

config.json

@@ -0,0 +1,26 @@
+{
+  "_name_or_path": "simcse-roberta-large-with-mask",
+  "architectures": [
+    "RobertaForCL"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": 0,
+  "eos_token_id": 2,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 514,
+  "model_type": "roberta",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "pad_token_id": 1,
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.2.1",
+  "type_vocab_size": 1,
+  "use_cache": true,
+  "vocab_size": 50265
+}

eval_results.txt

@@ -0,0 +1,11 @@
+------ test ------
++-------+-------+-------+-------+-------+--------------+-----------------+-------+
+| STS12 | STS13 | STS14 | STS15 | STS16 | STSBenchmark | SICKRelatedness |  Avg. |
++-------+-------+-------+-------+-------+--------------+-----------------+-------+
+| 62.53 | 75.35 | 68.75 | 80.49 | 76.69 |    76.67     |      69.24      | 72.82 |
++-------+-------+-------+-------+-------+--------------+-----------------+-------+
++-------+-------+-------+-------+-------+-------+-------+-------+
+|   MR  |   CR  |  SUBJ |  MPQA |  SST2 |  TREC |  MRPC |  Avg. |
++-------+-------+-------+-------+-------+-------+-------+-------+
+| 84.85 | 88.29 | 95.20 | 88.35 | 89.29 | 91.40 | 71.94 | 87.05 |
++-------+-------+-------+-------+-------+-------+-------+-------+

run_unsup_example.sh

@@ -0,0 +1,29 @@
+#!/bin/bash
+
+# In this example, we show how to train SimCSE on unsupervised Wikipedia data.
+# If you want to train it with multiple GPU cards, see "run_sup_example.sh"
+# about how to use PyTorch's distributed data parallel.
+
+python train.py \
+    --model_name_or_path $BC_ROBERTA_LARGE_PATH \
+    --train_file $BC_WIKI1M_PATH \
+    --output_dir result/simcse-roberta-large-with-mask \
+    --num_train_epochs 1 \
+    --per_device_train_batch_size 64 \
+    --learning_rate 5e-6 \
+    --max_seq_length 32 \
+    --pooler_type cls \
+    --mlp_only_train \
+    --overwrite_output_dir \
+    --temp 0.05 \
+    --do_train \
+    --fp16 \
+    --do_mlm \
+    --save_steps 5000 \
+    
+    # --load_best_model_at_end \
+    # --eval_steps 125 \
+    # --evaluation_strategy steps \
+    # --metric_for_best_model stsb_spearman \
+    # --do_eval \
+    # "$@"

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "sep_token": "</s>", "pad_token": "<pad>", "cls_token": "<s>", "mask_token": {"content": "<mask>", "single_word": false, "lstrip": true, "rstrip": false, "normalized": false}}

tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>", "add_prefix_space": false, "errors": "replace", "sep_token": "</s>", "cls_token": "<s>", "pad_token": "<pad>", "mask_token": "<mask>", "special_tokens_map_file": null, "name_or_path": "/mnt/nfs-storage-pvc-n26-20241218/rizejin/wzl/model-files/roberta-large"}

train_results.txt

@@ -0,0 +1,3 @@
+epoch = 1.0
+train_runtime = 3654.5011
+train_samples_per_second = 4.276

trainer_state.json

@@ -0,0 +1,208 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "global_step": 15625,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.03,
+      "learning_rate": 4.84e-06,
+      "loss": 0.6713,
+      "step": 500
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.680000000000001e-06,
+      "loss": 0.1946,
+      "step": 1000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 4.520000000000001e-06,
+      "loss": 0.1884,
+      "step": 1500
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.360000000000001e-06,
+      "loss": 0.1858,
+      "step": 2000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 4.2000000000000004e-06,
+      "loss": 0.1833,
+      "step": 2500
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 4.04e-06,
+      "loss": 0.1839,
+      "step": 3000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.88e-06,
+      "loss": 0.1816,
+      "step": 3500
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 3.7200000000000004e-06,
+      "loss": 0.1807,
+      "step": 4000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 3.5600000000000002e-06,
+      "loss": 0.1804,
+      "step": 4500
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 3.4000000000000005e-06,
+      "loss": 0.1803,
+      "step": 5000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 3.2400000000000003e-06,
+      "loss": 0.179,
+      "step": 5500
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 3.08e-06,
+      "loss": 0.18,
+      "step": 6000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.92e-06,
+      "loss": 0.1774,
+      "step": 6500
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.7600000000000003e-06,
+      "loss": 0.178,
+      "step": 7000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.6e-06,
+      "loss": 0.178,
+      "step": 7500
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 2.4400000000000004e-06,
+      "loss": 0.1775,
+      "step": 8000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 2.28e-06,
+      "loss": 0.1759,
+      "step": 8500
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 2.12e-06,
+      "loss": 0.1773,
+      "step": 9000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.9600000000000003e-06,
+      "loss": 0.1747,
+      "step": 9500
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.8000000000000001e-06,
+      "loss": 0.1756,
+      "step": 10000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.6400000000000002e-06,
+      "loss": 0.1768,
+      "step": 10500
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.48e-06,
+      "loss": 0.1753,
+      "step": 11000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.32e-06,
+      "loss": 0.175,
+      "step": 11500
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 1.1600000000000001e-06,
+      "loss": 0.1764,
+      "step": 12000
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 0.1758,
+      "step": 12500
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.400000000000001e-07,
+      "loss": 0.1752,
+      "step": 13000
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.800000000000001e-07,
+      "loss": 0.1755,
+      "step": 13500
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.2e-07,
+      "loss": 0.1751,
+      "step": 14000
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6e-07,
+      "loss": 0.1755,
+      "step": 14500
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0000000000000002e-07,
+      "loss": 0.1738,
+      "step": 15000
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4e-08,
+      "loss": 0.1747,
+      "step": 15500
+    },
+    {
+      "epoch": 1.0,
+      "step": 15625,
+      "train_runtime": 3654.5011,
+      "train_samples_per_second": 4.276
+    }
+  ],
+  "max_steps": 15625,
+  "num_train_epochs": 1,
+  "total_flos": 313292557056000000,
+  "trial_name": null,
+  "trial_params": null
+}