Commit
•
93ba188
1
Parent(s):
005361e
Update README.md
Browse files
README.md
CHANGED
@@ -54,6 +54,15 @@ reference-data-model:
|
|
54 |
https://github.com/tomaarsen/attention_sinks
|
55 |
https://arxiv.org/abs/2309.17453
|
56 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
Version:
|
58 |
- Link:
|
59 |
https://huggingface.co/NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v1
|
@@ -113,6 +122,7 @@ model = AutoModelForCausalLM.from_pretrained(model_id,
|
|
113 |
torch_dtype=torch.bfloat16,
|
114 |
load_in_4bit=True,
|
115 |
low_cpu_mem_usage= True,
|
|
|
116 |
|
117 |
attention_sink_size=4,
|
118 |
attention_sink_window_size=1024, #512, # <- Low for the sake of faster generation
|
|
|
54 |
https://github.com/tomaarsen/attention_sinks
|
55 |
https://arxiv.org/abs/2309.17453
|
56 |
|
57 |
+
TRL:
|
58 |
+
- Link:
|
59 |
+
https://huggingface.co/docs/trl/index
|
60 |
+
https://huggingface.co/docs/trl/sft_trainer
|
61 |
+
|
62 |
+
flash-attention:
|
63 |
+
https://github.com/Dao-AILab/flash-attention
|
64 |
+
https://arxiv.org/abs/2205.14135
|
65 |
+
|
66 |
Version:
|
67 |
- Link:
|
68 |
https://huggingface.co/NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v1
|
|
|
122 |
torch_dtype=torch.bfloat16,
|
123 |
load_in_4bit=True,
|
124 |
low_cpu_mem_usage= True,
|
125 |
+
#use_flash_attention_2=True, #GPU A100 or GPU supported
|
126 |
|
127 |
attention_sink_size=4,
|
128 |
attention_sink_window_size=1024, #512, # <- Low for the sake of faster generation
|