Commit Graph

27 Commits

Author SHA1 Message Date
duzx16 08bc85104d Fix attention mask for prefix prompt 2023-04-02 02:25:03 +08:00
duzx16 4b7ffbf070 No padding for chat function 2023-04-02 02:03:05 +08:00
duzx16 cc96a2271a Implement batch generation 2023-04-01 19:41:28 +08:00
duzx16 11c270c26c Fix position id for training 2023-03-31 20:18:10 +08:00
Zhengxiao Du 9c7416d834 fix GLM6BBlock name typo (#20)
- fix GLM6BBlock name typo (2a180534c5f2b4860598a9caef798b55a77cfb72)


Co-authored-by: rich brain <richbrain@users.noreply.huggingface.co>
2023-03-31 07:34:06 +00:00
duzx16 2e1be30ac4 Add support for loading quantized model 2023-03-31 10:48:38 +08:00
duzx16 c949d03152 Use dynamic dtype for prompts 2023-03-31 01:13:32 +08:00
duzx16 0cfae21ef8 Fix backward for quantization 2023-03-30 21:49:06 +08:00
duzx16 aea6cefcf5 Implement gradient checkpointing 2023-03-30 19:42:01 +08:00
duzx16 0564795e6e Fix bugs 2023-03-30 17:35:58 +08:00
duzx16 2200e2bc52 Add pad_token_id in config.json
Fix position_ids in ChatGLMModel
Add batch position_ids
2023-03-29 21:52:46 +08:00
duzx16 5c64357295 Set ignore_index for CrossEntropyLoss 2023-03-29 21:19:38 +08:00
duzx16 8127ab6abf Support batch training 2023-03-29 21:15:30 +08:00
duzx16 fbda1206cb Merge branch 'main' into dev_pt
# Conflicts:
#	modeling_chatglm.py
2023-03-29 20:37:39 +08:00
duzx16 812f43f9ff Add p-tuning v2 2023-03-29 20:22:57 +08:00
duzx16 096f3de6b4 Fix context length in get_position_ids 2023-03-28 17:37:46 +08:00
duzx16 4a9b711e61 Close CPU fusion on Mac 2023-03-23 22:43:06 +08:00
duzx16 d2bbc82a2c Fix Chinese punctuation 2023-03-22 14:37:21 +08:00
duzx16 2460dc2430 Remove hardcode bos_token_id 2023-03-19 14:56:15 +08:00
duzx16 42095d42ff Add support for streaming output 2023-03-19 14:31:26 +08:00
duzx16 220f772e9a Fix overflow in FP16 2023-03-16 09:26:05 +08:00
duzx16 f9f74fda55 Set is_parallelizable to False 2023-03-16 00:30:43 +08:00
duzx16 c3dece3f01 Add logit processor for NaN or Inf scores 2023-03-15 18:14:34 +08:00
duzx16 9d1509a1ad Fix default history argument 2023-03-14 18:38:49 +08:00
duzx16 d4832e8142 Add support for float32 2023-03-14 14:49:14 +08:00
duzx16 cd8041ea53 Fix past_key_values 2023-03-14 02:08:43 +08:00
Sengxian d11c6aaed8 Add chatglm-6b 2023-03-14 00:21:01 +08:00