Update modeling_chatglm.py

fix mask and position bug for batch generation

Fix attention score on mps

Update dependency

Merge branch 'main' of https://huggingface.co/THUDM/chatglm-6b

Fix tokenizer config saving

Fix LogitsProcessor using slim checkpoint (#29)

Use gmask in first place

Update slim checkpoint (#28)

Merge branch 'slim' of https://huggingface.co/THUDM/chatglm-6b into slim

Add gmask token id

Update slim checkpoint

Update code for slim

Drop icetk dependency

fix typo in use_gmask (#21)

Fix decode method for torch tensor

Support single integer or empty list as input to decode (#7)

Fix position ids expand

Fix generate

Fix attention mask for prefix prompt

No padding for chat function

Fix attention_mask and position_ids

Fix encode method

Fix batch input

Implement batch generation

Fix position id for training

fix GLM6BBlock name typo (#20)

Add support for loading quantized model

Use dynamic dtype for prompts

Fix backward for quantization

Implement gradient checkpointing

Fix bugs

Add pad_token_id in config.json

Change padding side

Set ignore_index for CrossEntropyLoss

Support batch training

Merge branch 'main' into dev_pt

Add p-tuning v2

Fix context length in get_position_ids

Close CPU fusion on Mac

Fix Chinese punctuation

Add English

Fix typo in tokenization_chatglm.py

Remove hardcode bos_token_id

Add support for streaming output

Fix overflow in FP16

Set is_parallelizable to False

Remove image tokens when decoding

Revert "Update quantization"

Update quantization