FIX: Make adaptation prompt CI happy for transformers 4.39.0 (#1551)

* fix for transformers 4.39.0 * Update src/peft/tuners/adaption_prompt/utils.py Co-authored-by: Benjamin Bossan <[email protected]> --------- Co-authored-by: Benjamin Bossan <[email protected]>
huggingface · Mar 11, 2024 · a1fe368 · a1fe368
1 parent 234bbab
commit a1fe368
Showing 1 changed file with 6 additions and 1 deletion.
diff --git a/src/peft/tuners/adaption_prompt/utils.py b/src/peft/tuners/adaption_prompt/utils.py
@@ -95,7 +95,12 @@ def llama_compute_query_states(model: nn.Module, **kwargs) -> torch.Tensor:
             new_cache_positions = torch.arange(past_seen_tokens, past_seen_tokens + q_len, device=value_states.device)
         position_ids = new_cache_positions.unsqueeze(0)
 
-    cos, sin = model.rotary_emb(value_states, seq_len=q_len + past_seen_tokens, position_ids=position_ids)
+    rotary_emb_kwargs = {"position_ids": position_ids}
+    # The `seq_len` argument has been officially removed in transformers >= 4.39.0
+    if "seq_len" in inspect.signature(model.rotary_emb.forward).parameters:
+        rotary_emb_kwargs["seq_len"] = q_len + past_seen_tokens
+
+    cos, sin = model.rotary_emb(value_states, **rotary_emb_kwargs)
 
     # For batched inference unsqueeze it on the correct dim
     # since: https://github.com/huggingface/transformers/pull/29109