Update mergekit_config.yml

mergekit_config.yml

@@ -1,424 +1,18 @@
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_task_level_merging_classification_gradient2_full/
-#     parameters:
-#       weight: 1
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_task_level_merging_clustering_gradient2_full/
-#     parameters:
-#       weight: 1
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_task_level_merging_retrieval_gradient2_full/
-#     parameters:
-#       weight: 1
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_task_level_merging_sts_gradient2_full/
-#     parameters:
-#       weight: 1
-# merge_method: multislerp
-# base_model: /mnt/data/models/Qwen/Qwen3-4B
-# dtype: float32
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_task_level_merging_classification_gradient2_full/
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_task_level_merging_sts_gradient2_full/
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_task_level_merging_classification_gradient2_full/
-# parameters:
-#   t: 0.5
-# dtype: float32
-
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_sts_classification_selerp
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_task_level_merging_clustering_gradient2_full/
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_sts_classification_selerp
-# parameters:
-#   t: 0.5
-# dtype: float32
-
-
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_sts_classification_clustering_selerp
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_task_level_merging_retrieval_gradient2_full/
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_sts_classification_clustering_selerp
-# parameters:
-#   t: 0.5
-# dtype: float32
-
-
-
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.2_clustering2_other_full
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_other_full
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.2_clustering2_other_full
-# parameters:
-#   t: 0.5
-# dtype: float32
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-0.6B_cut0.15_clustering1/
-#     parameters:
-#       weight: 0.863526622
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-0.6B_cut0.15_clustering2/
-#     parameters:
-#       weight: 0.020574888
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-0.6B_cut0.15_clustering3/
-#     parameters:
-#       weight: 0.11589849
-# merge_method: ties
-# base_model: /mnt/data/models/Qwen/Qwen3-0.6B
-# parameters:
-#   normalize: true
-#   int8_mask: true
-# dtype: float32
-
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering2_full/
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering3_full/
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering2_full/
-# parameters:
-#   t: 0.84
-# dtype: float32
-
-
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_clustering23_slerp
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_clustering23_slerp
-# parameters:
-#   t: 0.86
-# dtype: float32
-
-
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-#     parameters:
-#       weight: 0.863526622
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering2_full/
-#     parameters:
-#       weight: 0.020574888
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering3_full/
-#     parameters:
-#       weight: 0.11589849
-# merge_method: ties
-# base_model: /mnt/data/models/Qwen/Qwen3-4B
-# parameters:
-#   normalize: true
-#   int8_mask: true
-# dtype: float32
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-#     parameters:
-#       weight: 0.4
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering2_full/
-#     parameters:
-#       weight: 0.3
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering3_full/
-#     parameters:
-#       weight: 0.3
-# merge_method: ties
-# base_model: /mnt/data/models/Qwen/Qwen3-4B
-# parameters:
-#   normalize: true
-#   int8_mask: true
-# dtype: float32
-
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-#     parameters:
-#       weight: 1
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering2_full/
-#     parameters:
-#       weight: 1
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering3_full/
-#     parameters:
-#       weight: 1
-# merge_method: ties
-# base_model: /mnt/data/models/Qwen/Qwen3-4B
-# parameters:
-#   normalize: true
-#   int8_mask: true
-# dtype: float32
-
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering2_full/
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering3_full/
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering2_full/
-# parameters:
-#   t: 0.5
-# dtype: float32
-
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_clustering23_slerp55
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_clustering23_slerp55
-# parameters:
-#   t: 0.5
-# dtype: float32
-
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_clustering_retrieval123_slerp55/
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_data_mixing_gradient2_full/
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_clustering_retrieval123_slerp55/
-# parameters:
-#   t: 0.5
-# dtype: float32
-
-
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-gradient2-hotpotqa_0.02_8_5e-5_16_lora32_full/
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-grdient2-imdb_classification_0.02_2_5e-5_16_lora32_full
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-gradient2-hotpotqa_0.02_8_5e-5_16_lora32_full/
-# parameters:
-#   t: 0.5
-# dtype: float32
-
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-#     parameters:
-#       weight: 0.5
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.2_clustering2_full/
-#     parameters:
-#       weight: 0.5
-# merge_method: dare_ties
-# base_model: /mnt/data/models/Qwen/Qwen3-4B
-# parameters:
-#   normalize: true
-#   int8_mask: true
-# dtype: float32
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.2_clustering2_full/
-# merge_method: model_stock
-# base_model: /mnt/data/models/Qwen/Qwen3-4B
-# parameters:
-#   normalize: true
-#   int8_mask: true
-# dtype: float32
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.2_clustering2_full/
-# merge_method: karcher
-# dtype: float32
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-grdient2-medrxiv_title_0.02_8_5e-5_16_lora32_full
-#     parameters:
-#       weight: 0.5
-#   - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-gradient2-squad_0.02_8_5e-5_16_lora32_full
-#     parameters:
-#       weight: 0.5
-# merge_method: multislerp
-# dtype: float32
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-grdient2-medrxiv_title_0.02_8_5e-5_16_lora32_full
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-gradient2-squad_0.02_8_5e-5_16_lora32_full
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-grdient2-medrxiv_title_0.02_8_5e-5_16_lora32_full
-# parameters:
-#   t: 0.5
-# dtype: float32
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-#     parameters:
-#       weight: 0.863
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.2_clustering2_full/
-#     parameters:
-#       weight: 0.137
-# merge_method: task_arithmetic
-# base_model: /mnt/data/models/Qwen/Qwen3-4B
-# parameters:
-#   normalize: true
-#   int8_mask: true
-# dtype: float32
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-#     parameters:
-#       weight: 0.863
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.2_clustering2_full/
-#     parameters:
-#       weight: 0.137
-# merge_method: ties
-# base_model: /mnt/data/models/Qwen/Qwen3-4B
-# parameters:
-#   normalize: true
-#   int8_mask: true
-# dtype: float32
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.2_clustering2_full/
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.15_clustering1_full/
-# parameters:
-#   t: 0.137
-# dtype: float32
-
-
-
-
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.2_clustering2_other_cut0.15_clustering1_slerp_same_weighted
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_data_mixing_gradient2_sampled_full
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_cut0.2_clustering2_other_cut0.15_clustering1_slerp_same_weighted
-# parameters:
-#   t: 0.5
-# dtype: float32
-
-
-# slices:
-#   - sources:
-#       - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-gradient2-squad_0.02_8_5e-5_16_lora32_full
-#         layer_range: [0, 36]
-#       - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-grdient2-medrxiv_abstract_0.02_8_5e-5_16_lora32_full
-#         layer_range: [0, 36]
-# merge_method: slerp
-# base_model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-gradient2-squad_0.02_8_5e-5_16_lora32_full
-# parameters:
-#   t: 0.5
-# dtype: float32
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-grdient2-sts_0.02_8_5e-5_16_lora32_full
-#     parameters:
-#       weight: 0.5
-#   - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-grdient2-toxic_conversations_classification_0.02_2_5e-5_16_lora32_full
-#     parameters:
-#       weight: 0.5
-# merge_method: multislerp
-# dtype: float32
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-grdient2-sts_0.02_8_5e-5_16_lora32_full
-#     parameters:
-#       weight: 0.07
-#   - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-grdient2-tweet_sentiment_extraction_classification_0.02_3_5e-5_16_lora32_full
-#     parameters:
-#       weight: 0.93
-# merge_method: multislerp
-# dtype: float32
-
-
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-grdient2-sts_0.02_8_5e-5_16_lora32_full
-#     parameters:
-#       weight: 0.5
-#   - model: /root/paddlejob/workspace/env_run/output/model_merge_checkpoint/Qwen3-4B-grdient2-tweet_sentiment_extraction_classification_0.02_3_5e-5_16_lora32_full
-#     parameters:
-#       weight: 0.5
-# merge_method: multislerp
-# dtype: float32
-
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_data_mixing_gradient2_sampled50_full
-#     parameters:
-#       weight: 0.5
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_data_mixing_gradient2_sampled_other50_full
-#     parameters:
-#       weight: 0.5
-# merge_method: multislerp
-# dtype: float32
-
-
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_data_mixing_gradient2_sampled_jsonl_output60_seed1_full
-#     parameters:
-#       weight: 0.5
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_data_mixing_gradient2_sampled_60_seed100_full
-#     parameters:
-#       weight: 0.5
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_data_mixing_gradient2_sampled60_full
-#     parameters:
-#       weight: 0.5
-# merge_method: multislerp
-# dtype: float32
-
-
-# models:
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_general_all_data_sampled20_full
-#     parameters:
-#       weight: 0.2
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_original_4B_with_mrl_general_all_data_sampled40_full
-#     parameters:
-#       weight: 0.4
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_original_4B_with_mrl_general_all_data_sampled60_full
-#     parameters:
-#       weight: 0.6
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_general_all_data_sampled80_full
-#     parameters:
-#       weight: 0.8
-#   - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_original_4B_with_mrl_general_all_data_full
-#     parameters:
-#       weight: 1.0
-# merge_method: multislerp
-# dtype: float32
-
 
 models:
-  - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_general_all_data_sampled20_without_mrl_full
+  - model: Qwen3-4B_general_all_data_sampled20_without_mrl_full
     parameters:
       weight: 0.2
-  - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_original_4B_without_mrl_general_all_data_sampled40_full
+  - model: Qwen3-4B_original_4B_without_mrl_general_all_data_sampled40_full
     parameters:
       weight: 0.4
-  - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_original_4B_without_mrl_general_all_data_sampled60_full
+  - model: Qwen3-4B_original_4B_without_mrl_general_all_data_sampled60_full
     parameters:
       weight: 0.6
-  - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_general_all_data_sampled80_without_mrl_full
+  - model: Qwen3-4B_general_all_data_sampled80_without_mrl_full
     parameters:
       weight: 0.8
-  - model: /root/paddlejob/workspace/env_run/output/FlagEmbedding-master/Qwen3-4B_data_mixing_gradient2_plus_code_multilingual_full
+  - model: Qwen3-4B_data_mixing_gradient2_plus_code_multilingual_full
     parameters:
       weight: 1.0
 merge_method: multislerp