 oobabooga
|
316e07f06a
auto-assign gpu memory with --auto-devices alone
|
3 éve |
|
oobabooga
|
219366342b
Sort imports according to PEP8 (based on #67)
|
3 éve |
 81300
|
20dbef9623
Extend bfloat16 support
|
3 éve |
|
oobabooga
|
cadd100405
min_length has to be 0 when streaming is on
|
3 éve |
|
oobabooga
|
6be571cff7
Better variable names
|
3 éve |
|
oobabooga
|
58b07cca81
length_penalty can be negative (apparently)
|
3 éve |
|
oobabooga
|
7e4c25691d
Repetition penalty has to be < 5
|
3 éve |
|
oobabooga
|
1c30e1b49a
Add even more sliders
|
3 éve |
|
oobabooga
|
24dc705eca
Add lots of sliders
|
3 éve |
 Martin J
|
06a4664805
Fix a regex issue in `tokenize_dialogue`.
|
3 éve |
|
oobabooga
|
2fe235738e
Reorganize chat buttons
|
3 éve |
|
oobabooga
|
2207d44986
Windows doesn't like : in filenames
|
3 éve |
|
oobabooga
|
65266f3349
Fix loading official colab chat logs
|
3 éve |
|
oobabooga
|
44e8c671f9
Fix API documentation formatting in chat mode
|
3 éve |
|
oobabooga
|
a28f0d8bd7
Show it/s in the same units with or without streaming
|
3 éve |
|
oobabooga
|
4e4cd67223
Save chat history with name/date in filename
|
3 éve |
|
oobabooga
|
3af3ffeb90
Make --help output more readable
|
3 éve |
|
oobabooga
|
638495b633
Simplify generate() function
|
3 éve |
|
oobabooga
|
3f05cf5ddd
Simplify encode() function
|
3 éve |
|
oobabooga
|
2583bc5840
Simplify deepspeed implementation (#40)
|
3 éve |
|
oobabooga
|
f38c9bf428
Fix deepspeed (oops)
|
3 éve |
|
oobabooga
|
90f1067598
Move deepspeed parameters to another file
|
3 éve |
|
81300
|
248ec4fa21
Merge branch 'oobabooga:main' into ds
|
3 éve |
|
81300
|
a6f4760772
Add arg for bfloat16
|
3 éve |
|
81300
|
c515282f5c
no_split_module_classes not needed
|
3 éve |
|
81300
|
0a0d289537
Fix issue with generating on multiple GPUs
|
3 éve |
|
81300
|
a97afa6965
Add DeepSpeed ZeRO-3 integration
|
3 éve |
|
oobabooga
|
6b13816c47
Change default --disk behavior
|
3 éve |
|
oobabooga
|
119be56390
Add back low_cpu_mem_usage=True
|
3 éve |
|
oobabooga
|
d4a0b377ab
Allow standalone --cpu-memory
|
3 éve |
