Skip to content

Construct causal mask on-the-fly #493

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

Closed
andyrdt wants to merge 5 commits into from
Closed

Construct causal mask on-the-fly #493

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
e5cd850
construct causal mask on-the-fly
andyrdt Jan 24, 2024
b4d0583
Merge branch 'main' into attn_mask
bryce13950 Apr 8, 2024
336e3c6
Merge branch 'main' into attn_mask
bryce13950 Apr 11, 2024
d729191
ran formatting
bryce13950 Apr 11, 2024
02e5aca
Merge branch 'dev' into attn_mask
bryce13950 May 23, 2024
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
6 changes: 2 additions & 4 deletions tests/integration/test_grouped_query_attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,10 +52,9 @@ def test_grouped_query_attention_output_is_correct():
"b_K": b_K,
"W_V": W_V,
"b_V": b_V,
"mask": regular_attention.state_dict()["mask"],
"IGNORE": regular_attention.state_dict()["IGNORE"],
}
grouped_query_attemtion_state_dict = {
grouped_query_attention_state_dict = {
"W_Q": W_Q,
"b_Q": b_Q,
"W_O": W_O,
Expand All @@ -64,12 +63,11 @@ def test_grouped_query_attention_output_is_correct():
"_b_K": _b_K,
"_W_V": _W_V,
"_b_V": _b_V,
"mask": grouped_query_attention.state_dict()["mask"],
"IGNORE": grouped_query_attention.state_dict()["IGNORE"],
}

regular_attention.load_state_dict(regular_attention_state_dict)
grouped_query_attention.load_state_dict(grouped_query_attemtion_state_dict)
grouped_query_attention.load_state_dict(grouped_query_attention_state_dict)

query_input = torch.rand((1, 5, d_model))
key_input = torch.rand((1, 5, d_model))
Expand Down
2 changes: 1 addition & 1 deletion transformer_lens/loading_from_pretrained.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1039,7 +1039,7 @@ def convert_hf_model_config(model_name: str, **kwargs):
"n_heads": hf_config.num_attention_heads,
"d_mlp": hf_config.intermediate_size // 2,
"n_layers": hf_config.num_hidden_layers,
"n_ctx": 2048, # Capped bc the actual ctx length is 30k and the attn mask would be too big
"n_ctx": hf_config.seq_length,
"eps": hf_config.layer_norm_epsilon,
"d_vocab": hf_config.vocab_size,
"act_fn": "silu",
Expand Down