Jan 25, 2025 14:52 Upload pseudo_labeled_data_final/pl_gemma2_27B_stage1_lora_707_merged_combined.parquet with huggingface_hub 14:34 Upload train_stage2_stage1_combined_v1.parquet 14:26 Rename pseudo_labeled_data_final/eval_data_qwen_2.5_7B_stage1_init_stage0_combined.parquet to pseudo_labeled_data_final/eval/eval_data_qwen_2.5_7B_stage1_init_stage0_combined.parquet 14:25 Rename pseudo_labeled_data_final/eval_data_phi4_stage1_lora_700_combined.parquet to pseudo_labeled_data_final/eval/eval_data_phi4_stage1_lora_700_combined.parquet 14:25 Rename pseudo_labeled_data_final/eval_data_gemma2_9b_stage1_init_stage0_combined.parquet to pseudo_labeled_data_final/eval/eval_data_gemma2_9b_stage1_init_stage0_combined.parquet 14:25 Rename pseudo_labeled_data_final/eval_data_gemma2_27B_stage1_lora_707_combined.parquet to pseudo_labeled_data_final/eval/eval_data_gemma2_27B_stage1_lora_707_combined.parquet 14:25 Rename pseudo_labeled_data_final/eval_data_deepseek_llama3.18B_stage1_init_stage0_combined.parquet to pseudo_labeled_data_final/eval/eval_data_deepseek_llama3.18B_stage1_init_stage0_combined.parquet 14:24 Create eval/keep.txt 13:50 Rename pseudo_labeled_data_final/eval_data_qwen_2.5_7B_stage1_init_stage0_combined_combined.parquet to pseudo_labeled_data_final/eval_data_qwen_2.5_7B_stage1_init_stage0_combined.parquet 13:50 Rename pseudo_labeled_data_final/eval_data_gemma2_27B_stage1_lora_707_combined_combined.parquet to pseudo_labeled_data_final/eval_data_gemma2_27B_stage1_lora_707_combined.parquet 13:49 Delete pseudo_labeled_data_final/eval_data_gemma2_27B_stage1_lora_707_combined.parquet 13:44 Delete pseudo_labeled_data_final/eval_data_qwen_2.5_7B_stage1_init_stage0_combined.parquet 13:38 Upload pseudo_labeled_data_final/eval_data_qwen_2.5_7B_stage1_init_stage0_combined_combined.parquet with huggingface_hub 13:14 Upload pseudo_labeled_data_final/eval_data_gemma2_27B_stage1_lora_707_combined_combined.parquet with huggingface_hub 12:39 Upload pseudo_labeled_data_final/eval_data_phi4_stage1_lora_700_combined.parquet with huggingface_hub 11:45 Upload pseudo_labeled_data_final/eval_data_phi4_stage1_lora_700_combined.parquet with huggingface_hub 06:27 Upload pseudo_labeled_data_final/eval_data_phi4_stage1_lora_700_combined.parquet with huggingface_hub Jan 24, 2025 17:20 Upload pseudo_labeled_data_final/eval_data_gemma2_27B_stage1_lora_707_combined.parquet with huggingface_hub 15:51 Upload eval_data_deepseek_llama3.18B_stage1_init_stage0_combined.parquet 15:51 Delete pseudo_labeled_data_final/deepseek_llama3.18B_stage1_init_stage0_combined.parquet 15:21 Upload pseudo_labeled_data_final/eval_data_qwen_2.5_7B_stage1_init_stage0_combined.parquet with huggingface_hub Jan 23, 2025 10:18 Upload lmsys1m_unlabeled_78k.parquet 07:47 Upload pseudo_labeled_data_final/deepseek_llama3.18B_stage1_init_stage0_combined.parquet with huggingface_hub 06:08 Delete pseudo_labeled_data_final/eval_data_gemma2_9b_stage1_init_stage0 05:46 Upload pseudo_labeled_data_final/eval_data_gemma2_9b_stage1_init_stage0_combined.parquet with huggingface_hub Jan 22, 2025 11:07 Upload pseudo_labeled_data_final/eval_data_gemma2_9b_stage1_init_stage0 with huggingface_hub 10:57 Upload pseudo_labeled_data_final/eval_data_gemma2_9b_stage1_init_stage0 with huggingface_hub 10:39 Upload pseudo_labeled_data_final/eval_data_gemma2_9b_stage1_init_stage0 with huggingface_hub 10:30 Upload pseudo_labeled_data_final/eval_data_gemma2_9b_stage1_init_stage0 with huggingface_hub Jan 17, 2025 13:58 Upload train_combined_stage_1_v3.parquet Jan 12, 2025 11:13 Upload train_combined_stage_1_v2.parquet add orpo dpo mix 40k data to stage 1 data (maybe for 27B model if i train it) Jan 11, 2025 16:36 Upload train_orig_plus_pseudo_v1.parquet here's how the logits are prepared:
##### gemma 2 stage 1 pseudo labels
gemma_logits = 0.6 * logits + 0.4 * logits_tta
##### llama3 stage 1 pseudo labels
llama_logits = 0.5 * logits + 0.5 * logits
#### final soft labels
logits = 0.7878 * gemma_logits + 0.2122 * llama_logits
soft_labels = softmax(logits) 16:30 Delete train_orig_plus_pseudo_v1.parquet 15:19 Upload lmsys1m_125k_pseudo_label_llama3stage1_tta.parquet with huggingface_hub 13:42 Upload lmsys1m_125k_pseudo_label_gemma2stage1_tta.parquet with huggingface_hub 11:58 Delete lmsys1m_125k_pseudo_label_gemma2stage1_tta.parquet 11:41 Upload lmsys1m_125k_pseudo_label_gemma2stage1_tta.parquet with huggingface_hub 11:33 Upload 2 files 11:33 Delete lmsys1m_125k_pseudo_label_gemma2stage1_tta.parquet 11:32 Delete lmsys1m_110k_pseudo_label_llama3stage1.parquet 11:32 Delete lmsys1m_110k_pseudo_label_gemma2stage1.parquet 11:02 Upload eval_data_llama3_stage1_logits_tta.parquet with huggingface_hub 10:45 Upload eval_data_gemma2_stage1_logits_tta.parquet with huggingface_hub 10:17 Upload lmsys1m_125k_pseudo_label_gemma2stage1_tta.parquet with huggingface_hub 06:31 original stage 1 data + dpo mix 40k + lmsys 120k pseudo labelled from stage 1 ensemble (0.8 * gemma + 0.2 * llama3) Jan 10, 2025 16:37 Upload eval_data_gemma2_stage1_logits.parquet with huggingface_hub 16:19 Upload eval_data_llama3_stage1_logits.parquet with huggingface_hub 06:22 Upload lmsys1m_110k_pseudo_label_llama3stage1.parquet with huggingface_hub Jan 9, 2025 20:14 Upload lmsys1m_110k_pseudo_label_gemma2stage1.parquet with huggingface_hub 18:21 Upload lmsys1m_110k_pseudo_label_gemma2stage1.parquet with huggingface_hub 
