update eval_strategy for new transformers version

move import wandb to conditional

fully qualified imports to resolve name-space conflicts (#532)

ddp_mtl_nccl_support (#524)

edit imports in perturber_utils

fix exact_mean subselection to be for index rather than col name

fix exact_mean and exact_median to subselect dataframe emb cols, not cell rows

add quantization for pretrained model

Prepare data cell level pass cell state dict instead of genes (#483)

ensure gene class dict only inverted when using gene classifier

make plot roc compatible with eval-only metrics

Merge remote-tracking branch 'origin/main'

fix gene class dict labeling

Fix how mappings and train and val data are returned (#473)

Upload gene_name_id_dict_gc95M.pkl (#470)

Add checks for custom attributes and n_counts prior to sum ensembl id (#461)

remove unused imports while no longer using distributed sampler

Raise error for train and validation mismatch (#459)

Upload in_silico_perturber.py (#432)

Update geneformer/emb_extractor.py (#453)

Update geneformer/tokenizer.py (#450)

update isp default mode to cls

update pretrainer to not use distributed sampler (Trainer uses accelerate)

update function for N_Detections for mixture_model without anchor_token

add check to ensure emb_label is None for getting state embs dict

change default emb to cls

Update trainer output dir (#427)

Update geneformer/tokenizer.py (#415)

CUDA kernels incompatible with standard PyTorch device movement with 4bit/8bit, necessitating device-specific handling (#416)

dictionaries from parent dir (#405)

remove token dictionary and unpickling from init (#403)

move dict loading to function in eval utils

edit docs formatting

update tokenizer to defaults for 95M models for special token and input size

pointing dictionaries from the mtl module's init (#397)

Refactored token dictionary loading and encapsulated dictionary (#398)

Refactor: Convert mask_token_id, pad_token_id, and all_special_ids to properties (#395)

sync token_dictionary variable name w/ classifier

fix imports mtl/eval_utils

allow model_type valid options to take params model_type : {"Pretrained", "GeneClassifier", "CellClassifier", "MTLCellClassifier", "MTLCellClassifier-Quantized"} (#390)

"save_model_without_heads" is redundant (#385)

comment out "def save_model_without_heads(original_model_save_directory)"; redundant for ISP/Emb extractor (#382)

fixed bug related to dynamic ranges in dictionary with 'min' and 'max' value mismatch in optuna suggest fn (#380)

precommit formatting

update with 12L and 20L i4096 gc95M models, multitask and quantiz code

rename for consistency

delete old gene name dict

update to only have gene names as keys in gene_name_id_dict

Add function for summing of Ensembl IDs (#377)

save pval