Commit History
more fixes
0aa7fc9
changes
66ac3c1
Fixes exceeding maximum sequence length when using generate().
759d148
Uses native torch decorator for disabling autocast.
5819d04
Adds disable_autocast support for different device types.
67ecc75
Fixes any potential overflow when calculating attention weights.
b5c5161
Delete modeling_mixformer_sequential.py
470e18a
Delete configuration_mixformer_sequential.py
bd98e4e
Upload pytorch_model.bin
34b22f4
Update to new model interface.
bbace88
Improves type hinting on configuration arguments.
8d2c4ce
Fixes flash-attn import with a try/except statement
9ed5987
Adds support for flash-attn rotary embedding and fused dense layers.
90c38d9
Adds support for MQA/GQA and attention mask during training / fine-tuning.
371fd51
Upload modeling_mixformer_sequential.py
633bca1
Upload README.md
769684a
fix(phi-1): Checks length of `attention_mask`if it is passed as direct tensor.
1f890f7
Support for `attention_mask` in forward pass.
d22f35e
Update README.md
621f844
Upload tokenizer
7a24267
Upload MixFormerSequentialForCausalLM
44cca9f
Update README.md
3034d33
Update README.md
1121e12
Update README.md
3e86fe1
Update generation_config.json
a85c61b
Update generation_config.json
cb13e96
Update generation_config.json
a15ded7
Update README.md
ba068cc
Update README.md
4cb33c4
Update README.md
3cf35a2
Update README.md
9e27d7d
Upload Research License.docx
046a667
Update README.md
3670ef4
Update README.md
9c0466b
Update README.md
9691b01
Upload tokenizer
ebfa940
Upload MixFormerSequentialForCausalLM
e96b200
Upload tokenizer
91817f9
Upload MixFormerSequentialForCausalLM
0f4ae0e
initial commit
47c069f
Gunasekar
commited on