Skip to content

Fix LLama RoPE #910

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Apr 30, 2025
Merged

Fix LLama RoPE #910

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
cf01dc4
Fix LLama RoPE
mntss Apr 11, 2025
c608eee
Merge branch 'dev' into fix_llama_rope
bryce13950 Apr 30, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions transformer_lens/HookedTransformerConfig.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -262,6 +262,7 @@ class HookedTransformerConfig:
NTK_by_parts_low_freq_factor: float = 1.0
NTK_by_parts_high_freq_factor: float = 4.0
NTK_by_parts_factor: float = 8.0
NTK_original_ctx_len: int = 8192

def __post_init__(self):
if self.n_heads == -1:
Expand Down
2 changes: 1 addition & 1 deletion transformer_lens/components/abstract_attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -504,7 +504,7 @@ def calculate_sin_cos_rotary(
factor = self.cfg.NTK_by_parts_factor
low_freq_factor = self.cfg.NTK_by_parts_low_freq_factor
high_freq_factor = self.cfg.NTK_by_parts_high_freq_factor
old_context_len = n_ctx
old_context_len = self.cfg.NTK_original_ctx_len

low_freq_wavelen = old_context_len / low_freq_factor
high_freq_wavelen = old_context_len / high_freq_factor
Expand Down
5 changes: 5 additions & 0 deletions transformer_lens/loading_from_pretrained.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -947,6 +947,7 @@ def convert_hf_model_config(model_name: str, **kwargs):
"NTK_by_parts_low_freq_factor": 1.0,
"NTK_by_parts_high_freq_factor": 4.0,
"NTK_by_parts_factor": 32.0,
"NTK_original_ctx_len": 8192,
}
elif "Llama-3.2-3B" in official_model_name:
cfg_dict = {
Expand All @@ -971,6 +972,7 @@ def convert_hf_model_config(model_name: str, **kwargs):
"NTK_by_parts_low_freq_factor": 1.0,
"NTK_by_parts_high_freq_factor": 4.0,
"NTK_by_parts_factor": 32.0,
"NTK_original_ctx_len": 8192,
}
elif "Llama-3.3-70B" in official_model_name:
cfg_dict = {
Expand All @@ -995,6 +997,7 @@ def convert_hf_model_config(model_name: str, **kwargs):
"NTK_by_parts_low_freq_factor": 1.0,
"NTK_by_parts_high_freq_factor": 4.0,
"NTK_by_parts_factor": 8.0,
"NTK_original_ctx_len": 8192,
}
elif "Llama-3.1-8B" in official_model_name:
cfg_dict = {
Expand All @@ -1019,6 +1022,7 @@ def convert_hf_model_config(model_name: str, **kwargs):
"NTK_by_parts_low_freq_factor": 1.0,
"NTK_by_parts_high_freq_factor": 4.0,
"NTK_by_parts_factor": 8.0,
"NTK_original_ctx_len": 8192,
}
elif "Llama-3.1-70B" in official_model_name:
cfg_dict = {
Expand All @@ -1043,6 +1047,7 @@ def convert_hf_model_config(model_name: str, **kwargs):
"NTK_by_parts_low_freq_factor": 1.0,
"NTK_by_parts_high_freq_factor": 4.0,
"NTK_by_parts_factor": 8.0,
"NTK_original_ctx_len": 8192,
}
elif architecture == "GPTNeoForCausalLM":
cfg_dict = {
Expand Down
Loading